Marcel Danesi

AI in Foreign Language
Learning and Teaching
THEORY AND PRACTICE

No part of this digital document may be reproduced, stored in a retrieval system or transmitted in any form or
by any means. The publisher has taken reasonable care in the preparation of this digital document, but makes no
expressed or implied warranty of any kind and assumes no responsibility for any errors or omissions. No
liability is assumed for incidental or consequential damages in connection with or arising out of information
contained herein. This digital document is sold with the clear understanding that the publisher is not engaged in
rendering legal, medical or any other professional services.
Copyright © 2024 by Nova Science Publishers, Inc.

All rights reserved. No part of this book may be reproduced, stored in a retrieval system or transmitted in any form
or by any means: electronic, electrostatic, magnetic, tape, mechanical photocopying, recording or otherwise without
the written permission of the Publisher.

We have partnered with Copyright Clearance Center to make it easy for you to obtain permissions to reuse content
from this publication. Simply navigate to this publication’s page on Nova’s website and locate the “Get Permission”
button below the title description. This button is linked directly to the title’s permission page on copyright.com.
Alternatively, you can visit copyright.com and search by title, ISBN, or ISSN.

For further questions about using the service on copyright.com, please contact:
Copyright Clearance Center
Phone: +1-(978) 750-8400 Fax: +1-(978) 750-4470 E-mail: [email protected]

NOTICE TO THE READER

The Publisher has taken reasonable care in the preparation of this book, but makes no expressed or implied warranty
of any kind and assumes no responsibility for any errors or omissions. No liability is assumed for incidental or
consequential damages in connection with or arising out of information contained in this book. The Publisher shall
not be liable for any special, consequential, or exemplary damages resulting, in whole or in part, from the readers’
use of, or reliance upon, this material. Any parts of this book based on government reports are so indicated and
copyright is claimed for those parts to the extent applicable to compilations of such works.

Independent verification should be sought for any data, advice or recommendations contained in this book. In
addition, no responsibility is assumed by the Publisher for any injury and/or damage to persons or property arising
from any methods, products, instructions, ideas or otherwise contained in this publication.

This publication is designed to provide accurate and authoritative information with regard to the subject matter
covered herein. It is sold with the clear understanding that the Publisher is not engaged in rendering legal or any other
professional services. If legal or any other expert assistance is required, the services of a competent person should be
sought. FROM A DECLARATION OF PARTICIPANTS JOINTLY ADOPTED BY A COMMITTEE OF THE
AMERICAN BAR ASSOCIATION AND A COMMITTEE OF PUBLISHERS.

Additional color graphics may be available in the e-book version of this book.

Library of Congress Cataloging-in-Publication Data

Names: Danesi, Marcel, 1946- author.

Title: AI in foreign language learning and teaching : theory and practice /
Marcel Danesi.
Other titles: Artificial intelligence in foreign language learning and
teaching
Description: New York : Nova Science Publishers, [2024] | Series: Languages
and linguistics | Includes bibliographical references and index. |
Identifiers: LCCN 2023057308 (print) | LCCN 2023057309 (ebook) | ISBN
9798891133846 (paperback) | ISBN 9798891134355 (Adobe pdf)
Subjects: LCSH: Language and languages--Study and teaching--Technological
innovations. | Artificial intelligence--Educational applications.
Classification: LCC P53.855 D36 2024 (print) | LCC P53.855 (ebook) | DDC
418.0078/5--dc23/eng/20240109
LC record available at https://lccn.loc.gov/2023057308
LC ebook record available at https://lccn.loc.gov/2023057309

Published by Nova Science Publishers, Inc. † New York

Contents

Preface .......................................................................................... vii

Chapter 1 Technology in Foreign Language Teaching ....................1
Prologue...............................................................................1
The Language Laboratory ...................................................2
The Audio-Lingual Method ............................................... 4
The Role of the Native Language ...................................... 7
The Digital Language Laboratory .................................. 10
Computer-Assisted Language Learning
and Teaching .....................................................................12
CALL Pedagogy.............................................................. 13
The Role of the Teacher .................................................. 14
AI in Foreign Language Teaching .....................................15
Generative AI.................................................................. 16
General Features ............................................................ 17
Caveats ........................................................................... 19
The Contemporary Language Classroom ..........................19
Classrooms without Walls .............................................. 20
A Partnership.................................................................. 22
Epilogue.............................................................................23
Chapter 2 Foreign Language Learning ...........................................25
Prologue.............................................................................25
Theories and Models .........................................................26
Interlanguage Theory ..................................................... 27
The Input Hypothesis ...................................................... 31
Universal Grammar Theory ........................................... 33
Pragmatic Models........................................................... 35
Learning Variables ............................................................37
Motivational Factors ...................................................... 37
Enhancing Motivation .................................................... 38
iv Contents

Machine Learning ..............................................................39

Chatbots.......................................................................... 41
Conceptualization ........................................................... 42
Epilogue.............................................................................43
Chapter 3 Developing Linguistic Competence ................................45
Prologue.............................................................................45
Pronunciation and Spelling ................................................47
Listening and Speaking ................................................... 48
Writing ............................................................................ 49
Grammar ............................................................................51
Words and Sentences ...................................................... 52
Grammar and Usage ...................................................... 54
Vocabulary ........................................................................58
Polysemy ......................................................................... 58
Thematic Vocabulary ...................................................... 59
Epilogue.............................................................................61
Chapter 4 Promoting Communicative Competence .......................65
Prologue.............................................................................65
Communicating in a Foreign Language ............................67
Communicative Competence .......................................... 68
Role of Context ............................................................... 69
Conversations ....................................................................72
Structure of Conversations ............................................. 74
Conversation Analysis .................................................... 78
Interactional Verisimilitude ...............................................82
Focused Interactions ...................................................... 82
Interactional Pedagogy .................................................. 86
Epilogue.............................................................................86
Chapter 5 Enhancing Conceptual Competence ..............................89
Prologue.............................................................................89
Conceptual Fluency ...........................................................90
Cultural Meanings .......................................................... 92
Calquing ......................................................................... 96
Metaphorical Competence .................................................98
Conceptual Metaphor ..................................................... 99
Metonymy and Irony ..................................................... 102
Contrastive Conceptual Analysis.....................................103
Translation Analysis ..................................................... 104
 Contents v

Metaphor Plugin ........................................................... 106

Epilogue...........................................................................107
Chapter 6 Blended Foreign Language Teaching ..........................111
Prologue...........................................................................111
The Learner .....................................................................113
The Main Competencies ............................................... 114
Learner-Chatbot Interactions ....................................... 118
The Teacher .....................................................................119
Learner-Centric Pedagogy ........................................... 120
A Teaching Assistant .................................................... 121
The Course.......................................................................122
Blended Pedagogy ........................................................ 123
A Paradigm Shift .......................................................... 124
Epilogue...........................................................................127
Appendix .........................................................................................129
Language Learning Apps.................................................129
Babbel ........................................................................... 129
DALL-E......................................................................... 129
Duolingo ....................................................................... 130
Rosetta Stone ................................................................ 130
Memrise ........................................................................ 130
Busuu ............................................................................ 130
Drops ............................................................................ 131
Mondly .......................................................................... 131
References .........................................................................................133
Index .........................................................................................143
Preface

In 2011, the popular American television quiz show called Jeopardy featured
two human contestants competing against IBM’s Watson, an Artificial
Intelligence (AI) system programmed on purpose for the competition. Watson
won the match by a large margin. Then, in 2017, another AI system, called
AphaGo, beat the world’s Go champion at the time, Ke Jie, with a creative
move that was previously unknown, surprising Go experts. After the match,
an even more “intelligent,” self-taught system, called AlphaGo Zero, was built
and it achieved, amazingly, a complete victory against AlphaGo (Silver,
Hubert, Schrittwieser, Antonoglou, Lai, Guez, Lanctot, Sifre, Kumaran,
Graepel, Lillicrap, Simonyan, and Hassabis 2018). Significantly, the
boardgame Go was considered much more difficult for computers to win than
other games such as chess, because its strategic nature makes it hard to assess
a changing board configuration in a straightforward probabilistic way. It truly
requires what we call “creativity.”
Such mindboggling events have made it saliently obvious that AI has
come of age, bearing many implications for understanding what intelligence
is as well as how we learn as humans, given that the computer game systems
described above learned to adapt to an ever-changing game scenario with
apparent imagination and creativity. AI clearly has implications for how
learning itself unfolds. If an AI system can be devised to come up with a truly
intelligent move in the game of Go, previously unbeknownst to humans, then
the question arises: Can AI engage in creative conversations as well, making
it practically indistinguishable from a human interlocutor? A positive answer
to this question has emerged with the spread of chatbots throughout the social
sphere, from business to education, from the arts to science, and so on. A
chatbot is, essentially, a computer program that simulates and processes
human conversation (written or spoken), allowing humans to interact with the
chatbot as if they were communicating with a real human being. Although an
early chatbot prototype was developed in the mid-1960s by computer scientist
viii Marcel Danesi

Joseph Weizenbaum (1966), it has been only since 2022 that chatbot
technology and its everyday uses started spreading broadly—the year when
OpenAI’s ChatGPT came forward as a truly remarkable chatbot that was fine-
tuned to respond to prompts asking it to carry out specific tasks, such as
conversing with humans, and aiding learners acquire subject-matter in all
domains, from mathematics to foreign languages.
Chatbots are truly becoming partners or assistants to humans in various
fields of endeavor. As research in AI becomes ever more sophisticated, it
might even be possible for a chatbot to become even more “intelligent” than
humans in some tasks, as Ray Kurzweil claimed in his 2005 book, The
Singularity is Near, in which he maintains that there will come a moment in
time when AI will have progressed to a point that it will autonomously
outperform human intelligence. That moment, known as the (technological)
singularity, will occur when an upgradable software becomes self-sufficient
without human intervention, thus becoming capable of self-improvements.
Each new self-improvement will bring about an intelligence explosion that
will, in turn, lead to a powerful artificial intelligence (Kurzweil 2012). The
singularity may already be upon us, as the AlphaGo and chatbot systems
mentioned here make obvious. The implications are rather daunting at the
same time that they are exciting As Good (1965: 31) put it, enthusiastically,
decades ago: “Thus the first ultraintelligent machine is the last invention that
man need ever make.”
Whether or not the singularity is a plausible event, the point is that AI is
now a fact of everyday life, and it is moving into the educational domain,
influencing specific pedagogies such as foreign language teaching (FLT),
where it raises several key questions regarding both how a language is learned
and how language teaching is changing. As Robert Godwin-Jones (2022: 124)
has aptly observed, the new AI systems “have the potential to provide open-
ended conversational practice and language development which aligns with an
ecological, usage-based perspective on language development.” It is clear that
the time has come to look concretely at the pedagogical questions that the new
AI technologies raise for the traditional language teaching classroom. The
purpose of this book is, in fact, to examine the central issues related to the use
of AI in foreign language learning and teaching, including the following
questions: Can AI can teach languages independently of humans? What does
it tell us about teaching and learning? Can it discover new ways of teaching a
foreign language effectively? What are the implications for language
education in general?
 Preface ix

The thrust of this book concerns how to make FLT, at the very least,
relevant in an age of an ever-expanding AI culture, so as to make the language
learning experience consistent with what students are exposed to in their daily
lives as well as in other areas of education, which are increasingly opting for
AI as a complementary tool of instruction. Teaching and learning foreign
languages today is no longer restricted to the “walled-in classroom,” where
materials, pedagogy, and face-to-face interactions have always been
considered to be the central elements in bringing about successful learning
outcomes. Today, online courses, YouTube videos, digitally-available
materials, and the like have come forward to rival the traditional classroom as
the primary locus for foreign language learning. However, the question arises
as to whether nor not the outcomes are any more successful than they were in
the past. Some initial work in using Deep Learning AI systems, which have
the ability to interact with students and teachers in tandem, is showing how
classroom-teaching can be reinforced and perhaps even enhanced with an
“algorithmic tutor,” so to speak (for example, Fryer, Coniam, Carpenter, and
Lăpușneanu 2020, Bibauw, François, Van den Noortgate, and Desmet 2022).
This book will provide an overview of what the research says and what
prospects AI bears for FLT at present and in the future. It will delineate
concrete ideas that can potentially make foreign language teaching relevant to
the techno-savvy student, without compromising the role of the human
teacher. The book is based both on relevant research I have conducted on using
text analysis with AI (Neuman, Danesi, and Vilenchik 2023), and on
reviewing the relevant literature in AI-assisted FLT.
I should mention that I will be using ChatGPT as the main AI system in
the discussions and illustrations throughout this book for two reasons: (1) at
the time of writing this book it was the most commonly used one in foreign
language classrooms, and (2) it is the easiest one to use for teachers, at least
in my opinion. ChatGPT stands for “Chat Generative Pre-trained
Transformer,” which is a large language model chatbot developed by OpenAI.
It enables users to refine and guide a conversation in terms of desired length,
format, style, level of detail, and type of language used. Successive prompts
and replies are the cues that shape a human-ChatGPT conversation.
How will AI affect the human teacher and the overall structure of foreign
language teaching? Traditional pedagogy has often seemed incapable of
satisfying the individualized needs of learners; nevertheless, the teacher as part
of the learning system (wherever and however it occurs) cannot be replaced.
Rather, AI provides the teacher with a means to furnish individualized,
innovative, and cooperative learning instruction. Overall, the fact is that we
x Marcel Danesi

can no longer, as language teachers, ignore the ever-evolving technologies, or

else we will truly be left behind. AI is not going to replace teachers, but
teachers who use AI are likely going to gradually replace those who do not.
As entrepreneur Mark Cuban has bluntly put it (in Ringelberg 2023):
“Whatever you are studying right now, if you are not getting up to speed on
Deep Learning, neural networks, etc., you lose. We are going through the
process where software will automate software, automation will automate
automation.”

Marcel Danesi
University of Toronto, 2023
Chapter 1

Technology in Foreign Language Teaching

Prologue

Whenever there has been a methodological or technological innovation,

bringing with it the promise of facilitating and enhancing both language
learning and teaching, there has always been a concurrent period of reflection
and debate on the soundness of the promise itself. The classic example is the
language laboratory, which became a staple of foreign language teaching
(FLT) in the 1960s, leading to the establishment of the so-called Audio-
Lingual Method, which was initially welcomed with enthusiasm, but also with
some skepticism on the part of language teachers. As time went on, it became
obvious to everyone that it was not the “magic pill” it was thought to be. Since
then, teachers have proceeded with caution vis-à-vis new methods and
technologies that come with a similar promise.
The language laboratory brought about the first real “technological
revolution” in FLT; the second one was computer-assisted-language-learning
(CALL), which came about after the spread of personal computers in the
1980s. Both these events will be discussed further subsequently. Suffice it to
say at this point that, even though their relevant technologies have been
upgraded, the language lab and CALL remain essentially ancillary teaching
systems, designed to supplement or complement the pedagogical process that
occurs in a classroom environment. However, since the advent of Generative
AI, and the chatbots, which became available broadly at the start of the 2020s,
a third technological revolution has occurred in FLT which promises to go
beyond the adjunct function of the previous two. Some educators even believe
that Generative AI systems will even replace human teachers. But is this truly
possible? In a pertinent study, Carmelina Maurizio (2021), from the University
of Turin, expressed the dilemma facing language teachers today as follows:
Can AI algorithms truly enhance the learning of foreign languages and even
substitute the human instructor? In several relevant studies (Yin and Satar
2020, Dokukina and Gumanova 2020), the answer appears, apparently, to be
yes. Since we are at the beginning of a truly momentous revolution in language
teaching, examining the implications of this technological event is of obvious
importance for teachers and students alike. That is the aim of this book. This
2 Marcel Danesi

chapter will take an initial look at the role of technology in FLT, historically
and pedagogically.
For years, big tech companies have been developing algorithmic systems
that are getting closer and closer to simulating natural dialogue—they are able
to answer not only a wide range of questions but also understand and produce
metaphorical speech and even irony. The Duolingo platform, for example, can
simulate a human conversation in different languages; ELSA (English
Language Speech Assistant) supports English language learning, and is used
by millions of students worldwide. There are now chatbots that can be used
for different specific learning tasks. Given that Generative AI is a constantly
evolving technology, the two key questions it raises for FLT are the following
ones: Will AI actually replace the human teacher? Or is it simply another
technological tool that can be very useful in FLT as was the language lab and
CALL? There are those who enthusiastically say yes to the first question, and
there are those who express cynicism to both questions. Perhaps the sober
words of writer Douglas Adams (2002: 95) are needed in this regard, which,
though humorous, contain a grain of wisdom:

Anything that is in the world when you’re born is normal and ordinary
and is just a natural part of the way the world works. Anything that’s
invented between when you’re fifteen and thirty-five is new and exciting
and revolutionary and you can probably get a career in it. Anything
invented after you’re thirty-five is against the natural order of things.

The term foreign language teaching is used throughout this book in lieu
of second language teaching, although the two are often used interchangeably.
The reason is that the former term is typically constrained to indicate any
language studied after the first in a formal setting, whereas the latter can also
mean any language acquired after or at the same time with the native language
in any setting, or else a language learned in immigrant situations out of
necessity. In this book, the term foreign language is constrained to refer to any
language learned formally in a school or school-like environment.

The Language Laboratory

The first technology designed specifically to aid language students in

achieving successful learning outcomes was, as mentioned, the language
laboratory. The initial layout and structure of the language lab consisted of a
 Technology in Foreign Language Teaching 3

circumscribed space or room where students accessed audio or audio-visual

devices. The students could listen and respond to prerecorded content, through
headsets and microphones in isolated sound booths, to which a teacher could
simultaneously listen and thus manage, evaluate, and assess student oral
productions. Language labs were common in schools across the United States,
and many other parts of the world, starting in the two decades following World
War II (see Figure 1.1):

Figure 1.1. The Traditional Language Laboratory circa 1970 (Wikimedia

Commons).

For the sake of historical accuracy, it should be mentioned that

rudimentary pre-language labs go back considerably in time, emerging right
after the invention of the phonograph (Léon 1962). These consisted of nothing
more than phonographs to play vinyl records, and simple phonograph-based
tapes that allowed students to record their voices. They were not included in
individual booths. As Kitao (1995) comments:

It was Edison’s invention of the tin foil phonograph in 1877 that made
the first language laboratories possible. It was used for a foreign language
class for the first time in 1891. At first, records were mainly used to
preserve rare languages, but in the late 1800s and early 1900s,
correspondence courses were developed using records. Students listened
to records, recorded their own voices speaking the languages, and sent
their recordings back to the company for evaluation. The procedures used
by these early correspondence schools established methods that were
4 Marcel Danesi

later used in language laboratories. Between 1900 and 1950, equipment

became more sophisticated, with the invention and development of tape
recorders and television, and schools began establishing language
laboratories.

In the 1940s, linguists at the University of Michigan developed the so-

called Audio-Lingual Method (ALM) for foreign language learning. The
method relied on repeated listening and speaking drills, and hence the use of
new language labs became central to the method. By the mid-1950s, hundreds
of language labs were constructed in the US, with the majority in colleges and
universities. A decade later, thousands of language labs were built in all kinds
schools across the country as the ALM became a mainstream model of FLT.
However, by the early 1970s use declined as language teaching shifted away
from behaviorist models of learning, on which the ALM was based, to
different views of the learning process (discussed in chapter 2).

The Audio-Lingual Method

The language laboratory was incorporated into mainstream FLT from the late
1940s to the early 1970s—an era in which it subserved mainly the learning
goals of the ALM. It consisted, as mentioned, of individual booths for
students, equipped with a microphone, audio tapes, headphones, and a control
console operated by a teacher or assistant. The teacher console included a tape
recorder to play the selected recording, a headset and system of switches
enabling the teacher to monitor individual student responses, and a
microphone for communicating with the students.
The ALM was inspired by the confluence of behaviorist psychology and
structuralist linguistics in the late 1920s and early 1930s. The former stressed
habit-formation, and the latter structural patterns in language design. In the
early labs, there were several behaviorist-focused features built into the
recorded content, consistent with the ALM:

• The students listened to pre-recorded tapes, in which native speakers

of the foreign language (FL) enunciated syllables, words, phrases,
sentences, and larger texts in order of complexity. The idea was to
allow for a fine-tuning of student comprehension of the spoken FL.
• Students would repeat the FL items imitatively and then, in later
sessions, engage in question-and-answer drills with the recorded
 Technology in Foreign Language Teaching 5

material or with teachers so as to enhance their production skills in

the FL.
• In the playback mode, students could listen to their responses on tape,
and compare them to the responses given by native speakers, a phase
that was clearly intended to encourage self-correction and to boost
self-learning.
• As technology in lab design advanced, a visual component was added
to many labs, including slides, films, and photos designed to simulate
reality-based and vivid FL-usage situations. The visual component
was also used to impart images of the FL culture, so as to connect the
verbal learning process to the cultural-conceptual one indirectly.
• A separate method, called the Audio-Visual Method, arose from the
new technology, which retained the main pedagogical objectives of
the ALM, but stressed the role of visualization in the learning process
much more.
• Like the ALM, the new method stressed pattern practice, habit
formation, and the teaching of oral skills before reading and writing
skills. But it added an innovative feature to this basic blueprint—the
new dialogical and conversational material was introduced visually
via filmstrips and other kinds of visual aids either in the classroom
but more normally in the language lab.

Overall, the underlying learning principle behind the construction and

design of the early language labs was the behaviorist view that imitation,
repetition, and habit-formation through comparison and self-correction were
at the core of the learning process. As mentioned, the pre-recorded tape would
present the correct model of a syllable, word, or sentence spoken by a native
speaker and the students would listen to it, repeat it, listen to the model again,
and then repeat it again. The tape would then continue presenting new items
for the students to repeat and then apply within a structural pattern—a
pedagogical technique called pattern-practice. So, if the pattern was the
present tense of first conjugation verbs in Italian, the recording would first
present the conjugation paradigm of a verb such as parlare (“to speak”): Io
parlo. The student would repeat this and then the tape would leave out the
verb, allowing the student to fill it in: Io …. It would then introduce another
verb mangiare (“to eat”) and this time just the pronoun with a blank, for the
student to fill in: Io …. At first, therefore there was no explicit grammar
instruction. The belief was that by repeating the pattern, memorizing it, and
6 Marcel Danesi

then using it with novel information, the student would learn it by induction
and thus be able to apply it to other areas of the FL spontaneously. The lab
lessons were thus built on pattern-practice drills over which the students had
little or no control.
Pattern practice was introduced by American linguist Charles Fries (1927,
1945), the director of the English Language Institute at the University of
Michigan, the first of its kind in the United States. Fries believed that learning
FL structure and usage through imitation and controlled repetition was the
starting point for the student. The earliest types of language lab drills
conformed to Fries’ view of learning, consisting of a four-part sequence:

• Repetition: students repeated an utterance as soon as they heard it: “Io

amo la pizza” (“I love pizza”).
• Inflection: one word in the utterance is replaced by another form for
the students to repeat: “Io mangio la pizza” (“I am eating pizza”).
• Further replacement: another word is then replaced by another: “Io
amo la frutta” (“I love fruit”).
• Restatement: students are then asked to use the relevant patterns,
typically in response to a pre-recorded question: Question: “Ami la
pizza e la frutta?” (“Do you love pizza and fruit”). Response: “Si, amo
la pizza e la frutta”) (“Yes, I love pizza and fruit”).

Of course, more than one type of drill would be incorporated into a

practice session, and the drills and patterns would become increasingly
complex. The point was to get the students to extrapolate any grammatical or
lexical pattern and then be able to use it spontaneously. Native-like
competence was thought to crystallize on its own over time via the same
methodology. However, by the early 1970s this pedagogical model of FLT
was largely abandoned, and along with it, the language lab. The main reason
was, as the research started showing, that students could not map the patterns
induced from the drills to real-life communicative situations in the FL (Rivers
1964, Lamendella 1979). By discarding the use of controlled repetition and
pattern practice from FLT, subsequent approaches may have ignored the need
for such type of instruction—after all, it is central to other kinds of learning
tasks such as learning to play an instrument proficiently (Castagnaro 2006).
This topic will be taken up again subsequently. The point here is that the ALM
introduced the first technology used to support the learning process.
 Technology in Foreign Language Teaching 7

In updated, current uses of the language lab the stimuli presented have
been expanded in nature considerably and include the following:

• A dialogue is generally used as the initial input, which can also be

supported visually with video technologies and techniques.
• Pattern practice is based on material taken from the dialogue.
• The meanings of new vocabulary items are illustrated both through
pattern practice and visual aids.
• Students can interact with the content via role-playing exercises.
• Follow-up activities are designed to help with comprehension,
production, and creative uses of the FL.

While strict ALM-based teaching is no longer found, by and large, in FL

classrooms, the approach has embedded several pedagogical principles into
the profession broadly. One is that imitation and emulation cannot be
completely eliminated. Another is that correct pronunciation plays a much
more crucial role in acquiring the FL than some more communicatively-based
methods have tended to believe. As Pardede (2010) has aptly observed in this
regard: “Although the usefulness of teaching pronunciation is one of the most
widely debated subjects in the field of language teaching, current pedagogical
thinking and research on pronunciation reveals that intelligible pronunciation
is a very essential component of communicative competence.”

The Role of the Native Language

A third principle on which the ALM was based was that that FL-learning is
shaped, in large part, by the students’ knowledge of their native language
(NL); that is, the NL was thought to play a determinative role in shaping and
guiding the FL-learning process, especially during the initial stages. The
evidence for this was gleaned from the typical errors as well as the successful
efforts that students consistently demonstrate in learning the FL, which show
a direct influence from the NL—a process called transfer. Positive transfer, as
it came to be called, occurs when the patterns of the two languages, the NL
and the FL, are isomorphic in some way; and this facilitates the learning
process considerably. Negative transfer, on the other hand, occurs when the
patterns are asymmetrical or non-existent. Transfer Theory became a major
aspect of FL-learning theory and of FLT practice. By comparing and
8 Marcel Danesi

contrasting the FL and the NL in specific ways (pronunciation patterns,

grammatical patterns, and so on), the practitioners of the ALM aimed to
predict where positive and negative transfer might surface in the learning
process—a technique known as Contrastive Analysis (CA). The results of CA
were then used to prepare classroom and language lab materials in a sequential
flow, whereby patterns that involved positive transfer were presented at first,
while those that activated negative transfer processes were delayed to later
learning phases. The idea was to allow the students to transfer their NL
knowledge to the FL at first, focusing on differential patterns after this initial
phase.
The ALM, however, did not surface in a historical vacuum. It involved
refinements and elaborations of various aspects of the so-called Direct Method
(DM) that emerged as the first psychologically-designed approach to FLT at
the end of the nineteenth and early twentieth centuries. The main assumption
in that approach was actually rather different from the assumptions that shaped
the ALM. The DM believed that the same psychological mechanisms and
processes enlisted by the child in acquiring the NL were involved in FL-
learning. Initially, the DM was a reaction to the use of explicit grammatical
training in the classroom. The congeners of the DM saw grammar as
complicating the learning process; after all, children were not taught to speak
their native language through grammatical training, but by simple exposure to
it in context. The FLT reformers in that era, such as François Gouin (1880)
and Wilhelm Viëtor (1886), actually saw grammar training as an obstacle for
the typical student, since it delayed and even blocked access to the FL. So, for
the first time in foreign language teaching history, teachers turned their
attention to psychology for insights on how to develop a psychologically-
compatible model of language pedagogy. Their decision to do so was further
influenced by five crucial events:

• Psychology itself had emerged as a scientific enterprise. The

reformers claimed that teachers should use the findings of
psychologists on language learning to guide them in selecting what
was to be taught and when to teach it.
• Linguistics was emerging at the same time as an autonomous
discipline (Saussure 1916). This gave rise to a new scientific view of
language.
• The Modern Language Association of America was founded in 1883
and the Modern Language Association of Great Britain in 1889, both
 Technology in Foreign Language Teaching 9

of which focused attention on the importance of the teaching method

itself in enhancing learning outcomes for all types of learners.
• The International Phonetic Association was established in 1886,
raising awareness about the importance of accurate pronunciation in
language learning.
• National languages came to play increasingly important roles on the
stage of world relations after the Industrial Revolution. This led to a
practical need on the part of the average person to learn foreign
languages in school. The reformers wanted to turn this need into an
achievable goal for all, not just for those with a talent for languages.

This new Zeitgeist led to the construction of the Direct Method. Its main
features were as follows (Titone 1968: 100-101, Richards and Rodgers 1986:
7-10):

• All instruction and classroom activities were carried out in the FL

from the very outset. For this reason, it was considered preferable for
teachers themselves to be native speakers of the language, so that they
could provide native-like input constantly.
• Utilization of the students’ NL was discouraged. The new vocabulary
was imparted directly through actions, gestures, pictures, objects, etc.
(hence the term Direct Method).
• Every lesson began with listening and imitation activities revolving
around a dialogue. To ensure that these were learning-effective and
appropriate, uniformly-constructed textbooks were introduced as the
basis of instruction. These contained the dialogues and follow-up
exercises arranged in order of increasing complexity.
• Pattern practice was used consistently, whereby the learner was
expected to induce the relevant rules of grammar inherent in the
patterns.
• Reading and writing instruction and practice followed initial oral
learning phases. From this, the chronological teaching of the four
skills was introduced into pedagogy: listening-speaking-reading-
writing.

Although some of the DM’s techniques had historical parallels—pattern

practice drills, for instance, were actually traceable to the substitution tables
of sixteenth and seventeenth-century grammars, and pedagogically-designed
10 Marcel Danesi

situational dialogues to the writings of Comenius (the seventeenth-century

Czech educator). But never before were they organized as systematically and
coherently into a model that was believed to be compatible with the
psychology of foreign language learning. Many of the DM’s techniques
continue to be used to this day (whether or not it is known) in a variety of
guises.
The DM was also based on an implicit transfer learning model; in this
case, it was assumed that the learning of the FL followed the same
developmental path as the acquisition of the NL. Via imitation and repetition
of basic patterns the student, like the child, gradually builds competence in the
new language. However, the DM ignited controversy almost from the outset.
Henry Sweet (1899), for instance, argued that adolescents and adults, unlike
children, brought previous language experience to the learning task. He also
warned against the assumption that all students learned in the same way,
according to purported psychological principles. A little later, the British
educator Harold Palmer (1917, 1921, 1922) expressed similar reservations,
emphasizing that the diverse backgrounds and motivations of learners
warranted the need for instructional flexibility. Rather than develop a single,
standard method, it was better to have a set of principles that could be tailored
according to situation. By the early 1920s, disenchantment with the DM had
reached a critical mass. It was abandoned en masse shortly thereafter.
The Audio-Lingual Method was the second major FLT method. Its
promise to promote successful learning for all students was strengthened by
the language laboratory, as discussed. The method differed from the DM by
taking the NL directly into account via Transfer Theory and Contrastive
Analysis. So, while the DM believed that the FL learner acquired the new
language tabula rasa as did the child, the ALM saw the process as being
shaped directly by the habits and unconscious knowledge of the NL.
Classroom and language lab materials were thus designed according to
Transfer Theory.

The Digital Language Laboratory

The language lab was largely abandoned starting in the 1970s, because of the
shift in FLT models, which were becoming increasingly focused on
communication in the FL. However, the advent of the personal computer in
the 1980s brought back the language lab, bolstered by digital technologies.
Levy (1997: 1) described the promise of the computer and the digital lab as
 Technology in Foreign Language Teaching 11

“the search for and study of applications of the computer in language teaching
and learning.” One of the first such labs was constructed in 2005 at the
prominent women’s college at Kochi in southwestern India. This new type of
lab was a self-contained classroom, with teachers pegged as language coaches
or debriefers. As digital technologies expanded dramatically in the first two
decades of the millennium, the digital language lab became increasingly
“smart,” with software that was increasingly more responsive to individual
learner needs. It provided instant access to native-speakers interacting in the
FL via the Internet, immersing the students virtually into communicative
contexts that reflected situations of interaction that occurred commonly in the
foreign country (see Figure 1.2):

Figure 1.2. Digital Language Lab circa 2015 (Wikimedia Commons).

The purpose of the language labs of the ALM era was for students to gain
auditory exposure to the language they were studying, within the confines of
the limited technology of the times. Today, the language lab has made possible
a whole new dimension to FL-learning that can both reinforce and amplify
classroom learning, and even replace it. Among its features, the following
stand out:

• Multimedia-based input: the software allows learners to interact with

multimedia content constructively.
• Interactivity: the digital lab provides learners with interactive
capabilities that enable them to practice and expand language skills,
including games and simulations.
12 Marcel Danesi

• Feedback: the software makes instant feedback and assessment

possible, allowing for self-monitoring.
• Individualized learning: the software allows students to learn at their
own pace on the basis of individualized feedback.
• Virtuality: the lab has the ability to provide learners with virtual
environments that simulate realistic situations in the foreign culture.
• Artificial Intelligence: the software may incorporate AI to provide
learners with personalized and expanded learning experiences.
• Cloud-based learning: the digital lab allows cloud-based learning,
whereby students can access language learning resources from any
device with an Internet connection.

To indicate the radical break from the behaviorist-based ALM language

lab, the term “language lab” fell into disfavor at the millennium, replaced with
terms such as “language media center” or “learning resource center”
(Scinicariello 1997). However, today the historical memory with regard to the
ALM has faded considerably, and the initial term of language lab has
resurfaced, but with the expanded design and functions just discussed. So, it
is fair to say that the language laboratory today has made a kind of comeback,
having expanded its capacities from simple native speech models and drill
practices to using the Internet, virtual reality, and AI to support FL-learning.
And significantly, the lab does not have be located in a special room, given
that a computer can download the required software allowing students to learn
anywhere and at any time.

Computer-Assisted Language Learning and Teaching

The second major technological revolution in FLT dovetailed with the spread
of computer technologies in the late 1990s. It was named CALL (Levy 1997,
Warschauer and Healey 1998). The term CALI (computer-assisted language
instruction) was used before CALL, reflecting its origins in computer-assisted
instruction already in the late 1960s. CALL today emphasizes computer
training systems that allow learners to work on their own. Research has largely
corroborated the efficacy of CALL, both as an ancillary system of learning-
teaching and as a self-contained learner-based system. In the former, it is used,
more or less, like a portable language laboratory; in the latter case, it is
essentially a self-contained digital classroom, allowing for individualized
 Technology in Foreign Language Teaching 13

learner needs to be met concretely. The findings of a large-scale survey of over

2,000 college students studying nineteen different FLs, conducted by
Sydorenko, Hsieh, Ahn, and Arnold (2017), found that CALL was perceived
as effective, motivating, and beneficial overall to acquiring FL competence.
The higher the use of computer technology outside of the classroom also
resulted in more positive views, independently of the FL.
Among the features that any CALL system now makes available, the
following stand out, showing that digital technology now makes FLT and FL-
learning more and more dependent on systems that go beyond the classroom
and the single teacher in it—a development that will be discussed in more
detail in the final chapter:

• image storage and sharing, based on scenes and situations in the FL

culture;
• social bookmarking, allowing users to save links to web pages that
they want to remember or share;
• discussion lists, blogs, wikis, social networking platforms with native
speakers of the FL;
• chat rooms and virtual worlds that allow for interaction with FL
speakers;
• FL podcasting;
• sophisticated audio tools, video sharing applications, and screen
capture tools for self-instruction purposes;
• animation tools, such as comic strips, cartoons, etc. that provide
sophisticated exposure to the FL;
• mashups if various kinds in the FL;
• games and simulations in the FL.

CALL Pedagogy

It was the emergence of the Web at around the same time as the spread of
CALL that made integrated pedagogy plausible as a mode of training students
to learn via computer systems in tandem with human actors, such as native
speakers and classroom teachers.
A problem that arose, early on, was that this led to considerable time-
wasting through random Web navigations. This implied that CALL on its own
might not be as productive as it was claimed to be. As a corollary, it implied
14 Marcel Danesi

the need for the teacher to act as a central figure in the whole process. Overall,
CALL brought about a new lexicon for discussing FLT and for envisioning
what should occur in a typical classroom. Among the items in this lexicon,
four stand out as now embedded deeply into any view of FLT:

1. As ancillary tools: using digital technologies to amplify or extend

classroom learning.
2. As integrative tools: using the technologies in tandem with traditional
learning materials, such as print textbooks.
3. As collaborative tools: using technologies to get students to interact
among themselves and others outside the classroom, such as with
virtual language communities.
4. As embedded tools: taking into account what students bring to the
classroom in terms of technological skills that have become
embedded in their lives.

However, as Myrdene Anderson (2013: 298) has warned, we should be

very careful to embrace technology as an educational panacea simply because
it is new and trendy, because it can itself become a source of new and
unwanted problems in education if we do not understand what it is and how it
blends with social life:

I question silver bullet solutions to perceived social problems, among

which our public education makes an easy target. This allows me to slide
from teaching and learning in the early grades right into issues of later
training in the social sciences, all larded with prescriptive exposure to the
quantitative. Technology, computers included, while frequently
suggested as solutions to social and educational problems often
contribute to those very problems.

The Role of the Teacher

The main impact of CALL pedagogy was that it opened up FLT to an

integrated way of looking at the classroom, changing the role of the teacher as
a magister instructor to that of a facilitator or coach, along with computer
technology. The question even arose, early on, if the teacher was needed at all,
given that the computer could carry out most, if not all, of the normal
pedagogical duties associated with the human teacher. Training in computer
 Technology in Foreign Language Teaching 15

literacy for the students was thus seen as key to transforming the classroom,
empowering students to dictate the course of events in a teaching program.
However, as became obvious in experimental teacher-less programs, students
variegated considerably, achieving differential outcomes and even leading to
a high proportion of discouragement from further learning.
The traditional classroom in which a teacher and students are in physical
and dialogical contact is being altered constantly. Some educators (Gobbi
2012, Neilsen 2012) believe that the virtual classroom, without physically-
present teachers, is inevitable for three main reasons: (1) it is more economical
(it costs less than paying many teachers to deliver the same content
individualistically); (2) it is in synch with the expectations of a society that has
become dependent on computer technologies; and (3) it reflects how learning
occurs in the contemporary world, namely through screens and interactive
software. While this is certainly occurring, it is also true, as Anderson warned
(above), that the virtual classroom is unlikely to replace the important contact
between a real teacher and students in some shared physical space, given the
need for such contact in human learning. Integrating the traditional classroom
with computer technologies and the resources of the Internet is not replacing
teachers, but simply changing their role—as any innovation has implied
throughout FLT history (Hartsell and Yuen 2006). It is true that the first two
major technological revolutions in FLT—the language lab (traditional and
digital) and CALL (in any of its forms)—were critical in instilling a new view
of the teacher, as a partner with technology, leading over time to what can be
called blended pedagogy—to be discussed in the final chapter.

AI in Foreign Language Teaching

The third technological revolution in FLT is an ongoing one, again having

profound implications for everything, from the role of the teacher to models
for developing appropriate pedagogy. And, as in the case of previous
technologies, there is no turning back the clock (Gkountara and Prasad 2022).
As McLuhan and Leonard (1967: 24) predicted decades ago, changes in
technology invariably bring about changes in how teachers and students
perceive their own roles:

Tomorrow’s educator will be able to set about the exciting task of

creating a new kind of learning environment. Students will rove freely
through this place of learning, be it contained in a room, a building, a
16 Marcel Danesi

cluster of buildings or an even larger schoolhouse. There will be no

distinction between work and play in the new school, for the student will
be totally involved. Responsibility for the effectiveness of learning will
be shifted from student to teacher.

As Holmes, Persson, Chounta, Wasson, and Dimitrova (2022: 10) have

aptly noted, the advent of AI is raising profound new issues on how we
perceive education and its objectives. With its ability to analyze data on
student performance and teacher and curricular needs, AI is becoming more
and more a tool for educators to create customized lesson plans and
assessments that align with each student’s learning style, which can improve
student experience and motivation, and ultimately lead to better outcomes.

Generative AI

To grasp the magnitude of the revolution we are undergoing, it is useful to go

over a few technological aspects of Generative AI, as a backdrop to gleaning
any implications it bears for FLT. It is defined as artificial intelligence capable
of generating text, images, or other media, with the ability to learn the patterns
in novel input and then generate new data with similar characteristics (Taulli
2023). Starting in 2020, advances in so-called transformer-based deep neural
networks led to the creation of a Generative AI systems capable of accepting
natural language prompts as input, called chatbots—computer programs that
process human conversation, allowing humans to interact with digital devices
as if they were communicating with a real person—of which ChatGPT (Chat
Generative Pre-trained Transformer) is the most widely-used one (at present).
Simply put, a deep neural network is a class of Machine Learning algorithms
that mimic the information processing of the brain. ChatGPT is a large
language model-based chatbot launched on November 30, 2022, which
enables users to refine and guide a conversation towards a desired length,
format, style, level of detail, and language used. It is based on the transformer
architecture developed by Google, fine-tuned for conversational applications.
It successor, GPT-4, is one of the most adaptive of all chatbots available at the
time of writing this book.
Chatbots can produce essays, carry on interactive dialogues, and translate
from any language to another, with almost perfect grammar and appropriate
style and register. Someone enters a written prompt, to which ChatGPT
responds with an answer, as though it was a person chatting with the prompter
 Technology in Foreign Language Teaching 17

rather than an AI application. The use of plugins can expand the range of tasks
that ChatGPT can carry out, including producing simulated oral speech and
reacting to speech commands, rather than just written ones.
Like previous technologies, the chatbots are transforming the FL
classroom, serving as language-learning aids, providing learners with
opportunities for engaging in meaningful conversations in the FL, and getting
instant feedback on their grammar and vocabulary usage. And like the
language labs of the past, interacting with ChatGPT helps learners improve
their listening, speaking, reading, and writing skills through the simulation of
real-life language-based scenarios. However, as will be discussed throughout
this book, it is essential to maintain a critical stance when using chatbots. As
an AI system, it may not always provide contextually appropriate responses.
Moreover, replacing real teachers completely with AI would be
counterproductive, given that a large part of learning is an embodied process,
requiring the meaningful signals that the body emits and receives which
bolster the learning process. A key project headed by Bernd Hackl (2016) has
shown that the constant interaction between teachers and students in a physical
classroom creates a social environment that is supportive of learning,
combining bodily language with verbal signals that enhances meaning
exchanges. The study found overall that successful learning hinges on the
physical presence of teachers and the physical learning context they create in
the classroom.

General Features

The main strength of AI language models is that they can adapt to student
interactions and to teacher requirements literally on the spot. Among the
features that current chatbots make available, the following are of specific
importance to FLT:

• New information. Chatbots can either supplement or introduce the

new information related to the FL as required in a course of study.
For example, teachers of Chinese can use ChatGPT to introduce
pinyin (a phonetic notation system for representing the sounds of
Mandarin Chinese with Latin alphabetic characters). By prompting
ChatGPT with a question such as, “What are the rules for Pinyin?”
the AI can provide a detailed explanation of the essential rules for
18 Marcel Danesi

pinyin and provide illustrations and practice material tailored

according to prompts.
• Differentiated instruction. Given the diverse backgrounds and
learning styles of students, ChatGPT can quickly adapt to them via its
analysis of both the different prompts and the actual performances of
students in specific areas of the FL. As a result of its analysis,
ChatGPT can generate individualized instruction and exercise
materials at various difficulty levels, as well as providing appropriate
translation into the learners’ NL. In several tests that I myself
conducted with ChatGPT, asking it to translate an utterance into
Italian and to provide stylistic variants to it, the result was truly
amazing, given its accuracy and its ability to even provide regional
variation correctly, coming up with versions that might have escaped
the attention of even a veteran teacher.
• Materials. As I also discovered in the case of Italian, ChatGPT is able
to generate quizzes, tests, and appropriate classroom activities, with
the relevant prompts and trained on actual materials already used in a
course. ChatGPT was able to instantly produce cloze tests, multiple-
choice exercises, true-or-false questions, as well as to suggest role-
playing and pair-based activities. Moreover, as I was able to confirm,
it had the capacity to adjust to learning level as required. ChatGPT
can also generate lesson plans, word lists and flash cards that are
essential for teaching a specific theme.
• Personalized teaching. As mentioned, ChatGPT has the ability to
generate materials and activities tailored to the individual needs and
competence levels of each learner. For instance, I gave ChatGPT a
prompt to generate a dialogue on asking for coffee at an Italian bar,
which it did competently. I also asked it to show sensitivity to the
identities of the interlocutors and to create a relevant dialogue that
revolved around them, which it did as well.
• Expanded exposure. ChatGPT can provide learners with access to a
vast repertoire and diversity of learning materials in the FL, including
those that involve real-life situations and authentic language use,
since it can navigate relevant sites and platforms on the Internet.
• Immediate feedback. Learners receive immediate feedback on their
language, including the errors they make, followed by suggested
corrections, and explanations concerning the nature of the errors. As
such, the AI facilitates self-evaluation. Learners can then prompt
 Technology in Foreign Language Teaching 19

ChatGPT to generate follow-up materials and quizzes for self-

assessment so that they can monitor their progress on their own,
without fear of being wrong in a classroom environment.
• Motivation. Overall, a chatbot in the classroom can help foster a
congenial motivating learning environment, allowing learners to feel
secure and comfortable and to not be concerned about losing face.

Caveats

There are, needless to say, some caveats with chatbots that are worth
mentioning at this point, and which will be elaborated upon, along with the
advantages, in subsequent chapters. In addition to the potential threat to
academic integrity, the chatbots might provide incorrect responses, without
knowing it. For instance, they might give incorrect instructions on the number
of strokes needed in writing, say, a Chinese character. Grading may also not
be accurate, especially since the traditional handwritten tests need to be
converted into digital formats, which are not flexible to interpretive variability.
Student work that involves subjective evaluation (such as creativity, style,
etc.) might also not be suitable for the AI system to guide and effectuate.
Interestingly, it has been found that chatbots sometimes “lie,” so to speak;
that is, they come up with invented information on their own, making them
somewhat unreliable. Rather than lies, they are called ‘hallucinations.”
Technically, these are (amazingly) confident responses by an AI that do not
seem to be justified by its training data. For example, a hallucinating chatbot
might embed plausible-sounding random falsehoods within its generated
content. In the case of FLT, this makes it even more important to maintain a
human teacher in the whole process, since only humans have the ability to
interpret content, not just generate it. To students, who may lack knowledge
of the FL and its culture according to their level of competency, hallucinations
might appear plausible because the chatbot’s answer is coherent.

The Contemporary Language Classroom

Putting aside faults, such as hallucinations, and given that chatbots require
human interlocutors with all their inconsistencies, there is little doubt that they
constitute a technological revolution in FLT. The traditional way of teaching
20 Marcel Danesi

with textbooks and the usual apparatus of exercises and tests written on paper
is increasingly out of synch with the times, even though it may still have utility
in various ways. Even in the early era of the language lab, quizzes and tests,
as well as some instructional functions, took place in the laboratory via audio
(and visual) recordings. With the arrival of AI and the Internet, the situation
has changed (or is changing) once again. Knowledge-attainment now is guided
not only by listening and reading, but also by navigating cyberspace with the
help of AI. The Canadian communications theorist, Marshall McLuhan (1962,
1964) claimed that the medium used to deliver content can alter the way the
content is understood. Today, this has revealed itself to be a veritable law of
media. It certainly cannot be denied that technology has been re-shaping the
world, including the academy.
Since the ALM and the language lab, the FL classroom has evolved
according to the McLuhanian law of media. The classroom consisting of desks
in which students sit in front of an instructor who writes on a blackboard is
now a quaint picture from the past. There is little doubt, however, that the
presence of a real teacher with real students in direct contact with each other
is still vital to the learning process, as discussed briefly above. It is in the
classroom where students, along with teachers, help one another understand
the ideas, tasks, and skills reflected in a particular lesson through an interactive
format. In this type of learning environment, students can practice how to
listen to one another, how to make meaning, and how to find common ground
while participating in a conversation or even in some drill session. This is
impossible to do with a machine alone, no matter how intelligent it might be.
Simply put, humans are social beings, machines are not, and thus require
social interaction so that real learning can unfold.

Classrooms without Walls

One of the main educational changes that the Internet and chatbots (trained on
the Internet) have brought about is what McLuhan called “classrooms without
walls” (McLuhan 1960). By this, he meant that classroom procedures can
occur via electronic media anywhere on Earth, being no longer restricted by
the time and space constraints of the physical classroom. Of course, even in
previous eras, when print materials such as books were the basis of education,
teachers and students could access ideas beyond the classroom, such as in
libraries, and share ideas through letter correspondences. But print moves
slowly, since books and journals must be bought, read, and then discussed
 Technology in Foreign Language Teaching 21

through dialogue in class or through further paper-based correspondence.

Electronic media make content access and communication instant.
In a discerning and quasi-prophetic way, McLuhan (1960: 3) foresaw the
radical educational changes that current technologies would bring about long
before these came into existence:

These new developments, under quiet analytic survey, point to a basic

strategy of culture for the classroom. When the printed book first
appeared, it threatened the oral procedures of teaching and created the
classroom as we now know it. Instead of making their own text, their own
dictionary, their own grammar, students started out with these tools. They
could study not one but several languages. Today these new media
threaten, instead to merely reinforce, the procedures of this traditional
classroom. It’s customary to answer this threat with denunciations of the
unfortunate character and effect of movies and TV, just as the comic book
was feared and scorned and rejected from the classroom. Its good and
bad features in form and content, when carefully set beside other kinds
of art and narrative, could have become a major asset to the teacher.

Because people the world over can now see themselves as participants in
events going on in some other part of the world by simply looking at a
computer screen, they truly tend to feel interconnected. Real-space villages
and communities involve territorialities that imply knowledge of the same
native languages and symbolic-ritualistic systems that bind people within
them together. But in the Internet-connected global village (a term introduced
by McLuhan), made up of interactive social networks, the forms of language
and culture of real-space communities are no longer requirements for entry.
Indeed, the global village has generated new lingua francas which are
ensconced in new forms of literacy, some of which are divorced from
historical linguistic from cultural practices. Thus, a new dynamic has emerged
uniting people in non-traditional ways. And these have great socio-emotional
power. The global village has also made the classroom without walls a
concrete reality, since information and training can be gleaned
individualistically on platforms and sites such as Google, YouTube, and the
like.
22 Marcel Danesi

A Partnership

Education today is on the threshold of a veritable paradigm shift. To put it

figuratively, students now live on both sides of a computer screen. At the
threshold of the shift, linguist and political critic Noam Chomsky (2002)
sensed an inherent danger in it, because he saw technology as potentially
making us more passive and inclined to let machines do the thinking for us,
thus making us even more inclined to abrogate our responsibility to think and
act on our own as individuals, thus debilitating true democracy and meaningful
discourse. On the other hand, the same technologies have given everyone a
voice, which can be used for benefit, despite inherent dangers.
Whatever the truth, the revolution has occurred, whereby we now merge
with our technologies in a new kind of partnership in how we carry out daily
activities. This same partnership is becoming intrinsic to education and, in our
case, specifically to FLT. Above all else, it involves an ever-broadening
outward reach away from strict classroom-based learning made possible by
the technologically-shaped global environment in which people now interact
and learn. It is relevant to cite McLuhan and Leonard’s (1967: 25) prescient
assessment of a technologized world—an assessment that they put forth
decades before the Internet:

The world communications net, the all-involving linkage of electric

circuitry, will grow and become more sensitive. It will also develop new
modes of feedback so that communication can become dialogue instead
of monologue. It will breach the wall between “in” and “out” of school.
It will join all people everywhere. When this has happened, we may at
last realize that our place of learning is the world itself, the entire planet
we live on. The little red schoolhouse is already well on its way toward
becoming the little round schoolhouse. Someday, all of us will spend our
lives in our own school, the world. And education—in the sense of
learning to love, to grow, to change—can become not the woeful
preparation for some job that makes us less than we could be but the very
essence, the joyful whole of existence itself.

In effect, the isolated classroom with a single teacher instructing and

interacting with a small group of students in one specific region of the world
is becoming more and more an anachronism in the current age. The shift from
the traditional “walled-in classroom” to the “classroom without walls” has
started to gain momentum. The partnership between humans and machines is
 Technology in Foreign Language Teaching 23

defining the classroom more and more, with both, being needed to enhance
learning and teaching in ways that were unimaginable in the not-too-distant
past.

Epilogue

With the advent of the language lab, FLT had merged, for the first time, with
a technology in an effort to solve the dilemma of FL-learning in a formal
environment. This was followed by CALL-based pedagogy, which introduced
computers into FLT. With the advent of AI the third revolution is well
underway. The question of whether AI can truly make FL-learning successful
in a formal environment is ultimately a moot one. Technology is how humans
evolve socially and intellectually. As McLuhan (1968: 85) so aptly put it, we
live in “a man-made environment that transfers the evolutionary process from
biology to technology.”
There are various kinds of ways in which technology extends human
capacities: physical (the wheel extends the capacities of the foot for
locomotion), intellectual (the alphabet extends the ability to record knowledge
efficiently), symbolic (numeral systems extend the ability to count),
mechanical (the printing press extended the use of writing), mnemonic (the
computer allows for greater storage of information than single brains can ever
expect to store), and so on. As we invent new and more powerful technologies,
so too do we change our modes of knowledge-making, of understanding the
world, of transmitting information, of interacting socially, and so on. In
previous eras, the print book was the main tool for conducting knowledge
enterprises and learning. It was a tool that humans knew how to use efficiently
and effectively for various reasons. It also formed the basis of how we learned
information and grasped knowledge.
As the age of print gave way to the electronic age, and now to the AI age,
the tools for knowledge-making and communicating have also changed, as has
our concept of ourselves and especially our bodies. To cite McLuhan (1970:
180) again, “When the evolutionary process shifts from biology to software
technology the body becomes the old hardware environment. The human body
is now a probe, a laboratory for experiments.” The lesson to be learned from
McLuhan is that social evolution is guided by the forces of technological
innovations but that such innovations do not cut the chains in the historical
chronicle that we ourselves have fashioned. And this is why humans are still
crucial to social processes, including education.
Chapter 2

Foreign Language Learning

Prologue

The ultimate sign of successful FL-learning is bilingual competence—the

ability to navigate meaning and communication equally or in large part in both
the native language (NL) and the new language (FL). And the sign that this
end goal has been achieved to greater or lesser degrees is the student’s capacity
to translate accurately and in culturally-appropriate ways from the NL to the
FL and vice versa. It is intriguing to note that the new chatbots are highly
expert at translating, and for this reason their design and functions may offer
relevant insights into FL-learning itself. As mentioned briefly in the previous
chapter, translating from English to Italian using ChatGPT indicated to me that
the AI was highly expert at doing so. What was particularly impressive was
the chatbot’s ability to infer contextual meaning from my prompts. In a similar
anecdote, Maxwell Timothy (2023), asked ChatGPT to translate the Spanish
phrase, “Gracias por preguntar, pero estoy bastante seguro aquí.” The
chatbot produced the following accurate translation: “Thanks for asking, but
I’m quite safe here,” adding further information about other possibilities for
translating the expression that varied according to Spanish social context.
Anecdotes like this abound, all pointing to the expertise that ChatGPT
shows in translating, seemingly knowing what items are contextually and
semantically appropriate. Another relevant example comes from the online
site Summa Linguae (summalinguae.com/translation/how-to-use-chatgpt-for-
translation/), where even a Chinese idiom is found to be no challenge for
ChatGPT, which accurately translated “未雨绸缪” (wèi yǔ chóumóu) as
“prepare for a rainy day,” and as an option it generated, “make provision
against possible trouble.” The Chinese idiom does indeed convey the idea of
taking precautions or making preparations in advance so as to prevent or
mitigate future difficulties, emphasizing the importance of planning for
potential challenges—a cultural concept imprinted in the idiomatic
expression. Further navigations throughout the Internet have revealed that
ChatGPT has the ability to not only take cultural connotations into account in
the translation, but regional-dialectal differences as well. Although post-
editing human-based interpretation is still needed in many instances, the fact
26 Marcel Danesi

that the AI is able to translate from one language to another so accurately is a

rather remarkable achievement.
As mentioned, this has implications for FL-learning theories, since the
architecture of the chatbot may provide clues as to what language is and how
we process it. This chapter will look at the traditional FL-learning theories in
reference to chatbot architecture.
One of the main challenges that FLT has always faced is to create
conditions in a classroom that are conducive to a meaningful, personalized
experience for students, leading to varied proposals in the past of how to
synchronize teaching practices with the learning styles and emotional needs
of the students (Witkin and Goodenough 1981). This implied, above all else,
the use of pedagogical tactics matched to the particular learning situation. It is
in this area that AI seems to have particular relevance, as will be discussed,
given that as it is able to quickly analyze not only the linguistic behavior of
the students, but also their level of knowledge and their needs, suggesting the
most suitable pedagogical techniques to facilitate learning for each individual,
as well as finding a middle ground that would apply to the class in general.
Thus, AI presents itself, above all, as a support to teachers and students, not
only in terms of enhancing self-learning, but also in terms of activities to be
developed for the entire classroom. However, only the teacher can intuit the
crucial emotional needs characterizing any situation and, therefore, to interpret
more accurately what the algorithm proposes. According to research, for some
types of learners, interactive exercises with an artificial system or exercises
using a computer can be extremely stimulating (for example, Peterson 2013).
Other students learn better, however, via social interaction within a real
student work group. It is up to the teacher, therefore, to determine how AI
could be used in specific ways to maximize the learning experience and
outcomes.

Theories and Models

Since the Direct Method (chapter 1), a major objective of FLT has been to find
ways to activate the brain’s innate learning mechanisms and steer them
towards the native-like acquisition of the FL as effectively and efficiently as
possible. One of the central features of the early language lab was, in fact, to
expose students to native speech, even if the drills were controlled
pedagogically and thus somewhat artificially. Given the implications that AI
systems such as chatbots now harbor for enhancing and facilitating FL-
 Foreign Language Learning 27

learning, it is useful to go over some of the major theories and models of the
past here in a schematic fashion, so as to assess how the use of AI fits in with
them.
In many of the previous models, there is a central premise—namely, that
students must be involved directly in processing FL input via exposure to
authentic materials, speech samples, instruction, and exercises. This is based
on a view of FL-learning that goes back considerably in time, whereby
students were expected to be participants, not just observers in the learning
process. This was, in fact, the implicit goal of the reformers at the end of the
nineteenth century (chapter 1), who turned to psychology to gain insights into
how learning occurs in the brain and then translating these into classroom
pedagogical models. The belief was that the students would feel included in
the learning process psychologically, rather than being treated as passive
recipients of instruction, as was the case largely in eras before the reform
movement, when pedagogy was centered on grammatical instruction in the
student’s NL and then getting students to translate from the NL to the FL and
vice versa—a method that was designated “Grammar-Translation” by the
reformers.
Ever since, the sciences of psychology and linguistics have been enlisted
to both understand FL-learning and to derive from the relevant research
effective classroom pedagogy. So, trends in both these theoretical domains
have directly influenced trends in the pedagogical domain. The question now
becomes: How do the uses of AI transformer models in the language
classroom fit in, if at all, with the traditional theories of FL-learning? In order
to attempt an answer to this question, it is useful to outline the major models
here that may still have some resonance as to how FLs are acquired even today.

Interlanguage Theory

One of the most widely-used models of FL-learning is Interlanguage Theory,

introduced by Larry Selinker in 1972, which is a derivative of Transfer Theory
(chapter 1). Like the latter model, Interlanguage Theory is based on the
common observation of the constant emergence of systematic errors that
students make due to interference from the NL (negative transfer). To this
view, however, Interlanguage Theory adds errors that result, not only from the
NL, but from general processes of overgeneralization. The previous errors are
called interlinguistic and the latter intralinguistic. As an example of an
interlinguistic error, an English-speaking learner of Italian typically adds the
28 Marcel Danesi

preposition per (“for”) to the Italian version of “I am waiting for the bus” =
“Aspetto per l’autobus”; instead of the native Aspetto l’autobus, without the
preposition. In this case the NL syntactic structure is transferred to the FL,
producing the error (see Figure 2.1):

error source

NL I am waiting for Mary
   
FL Io aspetto per Maria

Figure 2.1. Interlinguistic Transfer.

Figure 2.2. Interlanguage Theory.

Now, an error made by students of English as a foreign language such as

“She goed to the store” (rather than “She went to the store”) is intralinguistic,
since its source is not interference from any NL form, but rather the result of
the tendency to assume regularity by the process of analogy with other forms
in the relevant verb category, where the suffix -ed is used to indicate past
tense—she played, she worked, etc. The consistency of such non-random
errors in students’ efforts to speak or write the FL produces an “interlanguage”
that defines learner speech—a kind of “student dialect” of the FL that is
specific to non-native students of the language, whose characteristics result
from an overlap between errors due to NL transfer and errors due to processes
of analogy-based overgeneralization. The errors are seen as evidence that
learners are constructing an unconscious preliminary theory of the forms and
 Foreign Language Learning 29

uses of the FL based on the linguistic data to which they have been exposed,
on their way to acquisition of the FL (see Figure 2.2 above).
These two error types give an interlanguage its unique linguistic
organization. An interlanguage can become fossilized, that is, become the
actual form of the FL that the learner takes away from the classroom.
Interlanguages can vary according to learning context, and especially the type
of experiences that students bring to the FL-learning task. As mentioned,
Interlanguage Theory is an elaborated version of Transfer Theory, claiming
that both NL transfer and “natural” learning mechanisms, such as analogy, are
involved in shaping FL-learning (Dulay, Burt, and Krashen 1982). The Direct
Method, on the other hand, did not envision any impact of the NL on the
learning process; rather it believed that FL and NL learning were essentially
isomorphic; that is, the same psychological mechanisms and processes
enlisted by the child in acquiring the NL were involved in FL-learning. In
Interlanguage Theory, these mechanisms were seen as manifesting themselves
as intralinguistic errors. Initially, as briefly discussed, the DM was a reaction
to the Grammar-Translation method that emerged as far back as the
Renaissance. The late nineteenth-century reformers saw the emphasis on
grammar training as an obstacle for the typical student, since it did not provide
access to the language itself directly, and thus posed an impediment to many
students, since it required a background knowledge of grammar itself. The DM
was tailored instead to reflect how children learn their NLs, using techniques
such as the following (Richards and Rodgers 1986: 7-8):

• All instruction and classroom activities were carried out in the FL, at
least as much as possible .
• Vocabulary was imparted not through translation, but through
gestures, pictures, etc.
• Every lesson began with listening and imitation activities revolving
around a dialogue.
• Pattern practice drills followed, from which the learner was expected
to induce the relevant rules of grammar inherent in them.
• Reading and writing activities were delayed till after the acquisition
of listening and speaking skills.

These techniques were believed to simulate the inductive and

generalization processes involved in NL acquisition (chapter 1). However,
there were doubts from the outset about the validity of this learning model, as
30 Marcel Danesi

mentioned. After the demise of the DM, language educators started to look
more closely at the effects of the students’ NL habits on the learning process.
This led, as discussed, to Transfer Theory and the Audio-Lingual Method. The
origins of Transfer Theory can be traced, as discussed, to the key work of
Charles Fries (chapter 1), who saw the persistence of regularly-occurring
errors that students typically made as traceable to differences between the NL
and the FL. So, identifying which features of pronunciation, grammar, and
vocabulary were the same or different entailed two premises: (1) teaching
should start with features that the FL shared with the NL, given that the learner
was already familiar with them, and would thus purportedly acquire them
automatically; and (2) those that differed would instead receive much more
instructional salience later on in the pedagogical process. In this way, positive
transfer could be maximized and negative transfer minimized through
pedagogical means.
Of special relevance to the discussion here is that most of the early
theories saw FL-learning as an inductive process. In the case of the theory
underlying the DM, induction was claimed to be similar to the process shown
by children in acquiring their NL; in both Transfer and Interlanguage Theory,
the errors produced by students were seen as derived in some part from the
incorrect induction of pattern on the basis of the given pedagogical input.
Now, the relevant aspect of AI language models is that they too are trained to
induce patterns from large databases. The relevant program can then extract
and use the relevant information as required by an input request. In effect, AI-
language learning matches what is known about human induction, but in an
artificial-algorithmic way. The simplest learning algorithm, in fact, receives a
set of examples drawn from the language in question, from which a pattern is
established by the model that can be applied to similar data. This is not unlike
the pattern-practice concept of both the DM and the ALM. When the AI
algorithm generates a hallucination, it reflects the same kind of
overgeneralization process characterizing interlanguages. Without going into
the technical details here, the point is that induction and generalization are
intrinsic to AI learning systems, including the ChatGPT one. In an indirect
way, the method used to train AI language models and the kinds of
hallucinations these make are consistent with theories of human foreign
language learning, suggesting that these may be inherent in any learning
system—human, animal, and machine. Specifically, in machine-learning
systems, a generalization error (hallucination) is a measure of how accurately,
or not, an algorithm is able to predict outcome values for previously unseen
 Foreign Language Learning 31

data—a process that is remarkably similar to human learner efforts in guessing

what FL forms and uses may be like without having learned them formally.
Needless to say, there is much more to learning in humans than the
foregoing discussion. Of special pedagogical relevance is that ChatGPT (and
other chatbots) can easily identify interlanguage errors and then provide
students with explanations as to the source of the errors and subsequently
design exercises and other materials that can help redirect the FL construction
process away from its interlanguage phase towards a more native-like
understanding of the language and its uses (Algaraady and Mahyoob 2023).

The Input Hypothesis

The prompts and input data that ChatGPT receives guide its ability to perform
tasks in a language. The notion of input has, actually, always been a central
one in FL-learning theories, albeit from a different perspective. The DM, for
instance, constrained the input to be similar to the kind to which children were
exposed as they acquired their NL, gradually becoming more complex as FL
learners acquired competence. There has even been an entire model developed
on the view of the crucial importance of input—namely, the Input Hypothesis,
developed by the American applied linguist Stephen Krashen (1982, 1985).
Aware of the differences between childhood and adult (adolescent) situations,
due to the different environment, ages, and background knowledge, Krashen
claimed that there existed a neurological difference between acquisition and
learning—the former characterizes both NL and FL processes in the early
stages, and is spontaneous and unconscious, while the latter involves
conscious processing of the input and is guided by instruction and other formal
pedagogical systems. Krashen maintained that it was crucial to ensure that the
input to which students are exposed with regard to any novel task is conducive
to acquisition, and thus that it activate spontaneous learning mechanisms in
the brain. Learning-based input, on the other hand, involves the activation of
conscious reflection on the language, which occurs when students know
enough about the language to be able to reflect upon it formally, which
Krashen calls monitoring.
The distinction between acquisition and learning continues to have
relevance because it encapsulates something that teachers have always felt
intuitively—namely that students pick up certain things spontaneously and
with little effort, but require much more conscious effort and focus to grasp
other things. A part of Krashen’s model is based on the ideas of psychologist
32 Marcel Danesi

Lev Vygotsky (1962), including his “i + 1” characterization of acquisition,

which corresponds to Vygotsky’s notion of “zones of proximal development.”
This means that children and early FL learners progress through zones of
development which expand on their own as they are able to understand new
input by themselves—hence “i (input) + 1.” Krashen (1985: 1) saw classroom
learning as unfolding through a natural sequence, which “does not appear to
be determined solely by formal simplicity;” nor is it dependent on “the order
in which rules are taught in language classes.” Now, for input to be able to
instantiate acquisition processes, it must be able to lower what Krashen called
the “affective filter,” a mental block that prevents students from fully
acquiring the input to which they are exposed because of either fear of making
mistakes or else of resistance to new information that seems too complex or
abstruse.
The primary strategy in getting acquisition to unfold naturally, lowering
the affective filter, is to ensure not only that the input to which learners are
exposed is understandable, and that it contains “a bit” of information that is
beyond the student’s developing competence (Krashen 1985: 2):

The Input Hypothesis claims that humans acquire language in only one
way—by understanding messages, or by receiving “comprehensible
input.” We progress along the natural order by understanding input that
contains structures at our next “stage”—structures that are a bit beyond
our current level of competence. We move from i, our current level, to i
+ 1, the next level along the natural order, by understanding input
containing i + 1. We are able to understand language containing
unacquired grammar with the help of context, which includes extra-
linguistic information, our knowledge of the world, and previously
acquired linguistic competence.

Judiciously, Krashen claimed that his views were based on anecdotal

evidence and on pedagogical tradition, rather than on strict empirical research.
Nonetheless, the notion of input resonates with FL teachers. Significantly,
Krashen’s ideas have parallels in Generative AI research, where the notion of
“attention” seems to correspond, grosso modo, to Krashen’s notion of
monitoring. Transformers, such as ChatGPT, use attention to “weigh” the
influence of different words when generating a response. For instance, while
generating a response to the sentence “The cat chased its tail,” the model
appears to understand that “cat” is the subject and is more important than
“tail.” To allow it fine-tune its attention system, the AI model is exposed to a
 Foreign Language Learning 33

large corpus of data from the Internet, allowing it to induce statistical-

probability patterns of actual language use. For instance, since a phrase such
as “The sky is …” is often followed by the word “blue” in the training data,
the model will learn to predict “blue” as the likely next word. In a way, this is
the AI’s “i + 1” mode, whereby it chooses each word based on probabilities
derived from assessing the input. Interestingly, ChatGPT sometimes
introduces randomness in its word selection process coming up with truly
creative responses, again much like students in a FL course, even inventing
things out of the blue (so to speak), as its hallucinations show.

Universal Grammar Theory

Starting with the early ideas of American linguist Noam Chomsky (1957,
1965, 1986) a view of language emerged, called Universal Grammar Theory,
that was adopted by some applied linguists and FL teachers as a model to
design instruction and classroom materials (Jakobovits 1971, Chastain 1971).
Chomsky asserted that human infants possessed a species-specific language
capacity in the brain, which he initially called the Language Acquisition
Device, that allowed them to spontaneously develop their native-language
grammars by simply being exposed to verbal input. The capacity was based
on the belief that all languages are designed with the same innate blueprint,
differing only in fine-tuning details (called parameters) that specific grammars
require. The blueprint is the Universal Grammar (UG), consisting of a set of
principles of language design wired into the brain at birth that are constrained
by the specific parameters that characterize the grammar of the particular
language to which children exposed.
UG Theory is an attempt to account for the fact that childhood language
acquisition follows distinct stages universally, no matter where a language is
acquired or what that language is. Differences are due to organizational
subprinciples (parameters) that allow a specific language grammar to take
shape from the fund of general principles in the UG. But UG Theory ignores
two crucial facts of childhood language acquisition: (1) it overlooks the critical
role played by imitation; and (2) it excludes the role of processes such as
analogy and metaphor in shaping the learning process (to be discussed
subsequently).
While the relevance of UG Theory for FLT has been debated since it came
onto the scene, it actually brought about a rethink in how to model teaching
practices on the basis of a linguistics-based learning theory. As semiotician
34 Marcel Danesi

Charles Morris (1938, 1946) argued, one cannot isolate any one dimension of
language in any scientific consideration of its nature. There are three main
intertwined dimensions that coalesce to create language as a semiotic system:
(1) syntax, the organization of words into larger structures; (2) semantics, how
words in isolation and in syntactic frames deliver meaning; and (3) pragmatics,
how language is used context and how this influences its forms and meanings.
This tripartite model is inherent, significantly, in the design of chatbots, since
a computer will be able to determine the semantics of, say, an ambiguous word
on the basis of the words surrounding it in different texts (syntax) and in their
actual pragmatic uses. One area in which this is saliently evident is in how
computers assess collocations, sequences of words that typically co-occur in
speech more often than would be anticipated by random chance. Collocations
are not idioms, which have fixed phraseology. Phrases such as crystal clear,
cosmetic surgery, and clean bill of health are collocations. Whether the
collocation is derived from some syntactic (make choices) or lexical-semantic
(clear cut) criterion, the principle underlying collocations—frequency of
usage of words in tandem—always applies. And it is this principle that
undergirds chatbot design. First, the algorithm identifies a key word in context
and then determines the frequency of the combination of other words with the
key word in order to disambiguate the meaning of a phrase.
Chatbots are constructed on the basis of the above tripartite model,
providing insights on how the internal mechanisms of a natural language may
be activated with novel inputs—hence, their theoretical value to designing or
refining FL-learning theories and models. Specifically, the work in this area
has become a highly relevant one on at least four counts:

1. It forces FL researchers to unravel the relation between structure and

meaning in the formation of even the simplest sentences and how they
relate to external information.
2. It produces machine-testable models of meaning that can then be
discussed vis-à-vis the theoretical models of linguists.
3. It brings out the relation between grammar and context of use, and
how it might be modeled.
4. It allows the linguist to relate language to knowledge, and specifically
to how knowledge is represented linguistically.
 Foreign Language Learning 35

Pragmatic Models

In the 1970s, FLT started moving away from its theory-to-practice paradigm
moving instead towards the serious consideration of how to incorporate
pragmatic-communicative features of language into classroom teaching. It
was the linguist Dell Hymes (1971) who ignited this shift when he argued
convincingly that language systems were hardly impervious to influences
from real-world social interaction, suggesting that, on the contrary, verbal
structures were even modified themselves over time by language use. He
argued overall that the ability to apply language to specific situations in a
meaningful-systematic way constituted a different kind of competence from
pure linguistic competence, calling it communicative competence.
Actually, as far back as the work and ideas by Harold Palmer (1922), the
notion of a communicative, or pragmatic, competence was already fomenting
in the domain of FLT, as witnessed by the various oral methods that surfaced
after World War I, based on the idea that oral fluency would not emerge
without teaching communicative aspects of the new language directly. By the
mid-1970s pragmatic models started becoming broadly incorporated into FLT
(Van Ek 1975, Wilkins 1976). Their organizing principle was to teach how
native speakers employed language forms variably, yet systematically, to
carry out specific types of social functions. A simple protocol such as saying
hello or good-bye, for instance, is based on an implicit social script that
requires a detailed knowledge of the appropriate words, phrases, structures,
and nonverbal cues that come together cohesively to enable a speaker to carry
out a successful interaction with another speaker. An infringement of any of
the procedural details of this script might lead to a breakdown in
communication.
The topic of communicative competence will be discussed in chapter 4.
For the present purpose, suffice it to say that it is based on the view that
knowing how to use language during conversations is as systematic as
knowing the grammar of the language being employed. Already in the early
1920s, anthropologist Bronislaw Malinowski (1922) saw language as
fulfilling and serving specific social functions. For example, he coined the
term phatic communion in reference to the exchange of words and phrases that
are important less for their dictionary meanings than for establishing social
contact. When we greet someone with “How are you?” we hardly expect a
medical report, as would a doctor. The typical function of that statement is
simply to make contact. Malinowski also showed that the type of language
used in phatic communion could be used by the linguist to identify social
36 Marcel Danesi

variables—if the speakers are adults or children, if there is a difference in

status between the speakers, if they are well acquainted with each other, and
so on. Phatic communion is norm-based behavior. As work on pragmatics
became widespread in the 1970s, Malinowski’s views were expanded to study
all contexts and situations of language use.
One area in which chatbots are of particular relevance is that they
constitute a kind of pliable language lab, having the ability to replicate real
speech and conversations with accuracy. Moreover, a chatbot can respond to
prompts by students with personalized, relevant information, allowing them to
practice conversations minus the anxiety they may feel when talking to a
person. The chatbot also has the ability to adapt to the level of the
conversation, helping students identify and fix mistakes not only in grammar
but also in pragmatic appropriateness. In effect, ChatGPT is a self-contained
pragmatic model of language teaching, serving as a conversation partner, at
any time and at any place. An example of how ChatGPT can be used to carry
out a personalized conversation in French is provided by Edwige Simon
(2023), showing how the simulation is truly verisimilar to real conversations:

Prompt: S’il vous plaît, posez moi des questions sur ma famille
personnelle et professionelle, une question à la fois (“Please
ask me a few questions on my personal and professional
family, one question at a time”).
ChatGPT: Je comprends. Voici une question concernant votre famille
personnelle: Peuvez-vous me parler un peu de votre
famille? Combien avez-vous d’enfants et quelles sont les
choses que vous appréciez le plus dans votre role de parent?
(“I understand. Here are questions concerning your
personal family: Can you talk to me a bit about your
family? How many children to you have and what things
do you appreciate the most in your role as parent?”).

The interaction is remarkable on several counts: (1) it shows accuracy in

vocabulary selection; (2) it employs appropriate grammar; and (3) it uses the
correct level of formality (showing the use of the polite vous forms). However,
as discussed in the previous chapter, while the chatbot can simulate a real-life
situation, achieving communicative goals in a socially appropriate manner, it
cannot provide the human perspective of events, which implies a true
understanding of the relative importance of the events. AI is often not
 Foreign Language Learning 37

sophisticated enough to understand cultural differences in expressing and

reading thoughts and emotions, making it harder to draw accurate conclusions.

Learning Variables

One of the most important advantages of AI in FLT is, arguably, its ability to
adapt to individualized student styles of learning and to tap into learner
motivations for studying a language. In most pedagogical models, the learning
styles of students have been assumed to be critical as predictors of the best
teaching method to use, as was their motivation for enrolling in a course. Some
learners learn best through visual materials, others through audio forms of
instruction, others still by studying alone, and so on.
People learn in different ways and at different speeds, and their incentives
to learn may differ considerably. These variables cannot be ignored in the
design of any teaching approach or curriculum—as did methods such as the
DM and ALM to varying extents. It has become obvious that it is
impracticable to expect everyone to learn in the same way with the same
textbook and materials and be equally successful. This is where chatbots come
into the picture, since they make personalized approaches practicable, while
at the same time allowing for a “common pedagogy” to be used in certain
situations. Chatbots adapt to learners’ needs and respond to teacher requests
for ideas and materials that relate to a specific situation. Teachers can also
upload content into an AI system, which can then generate textbooks tailored
to a specific curriculum, course, or primary learning style of students.

Motivational Factors

The two main motivational factors identified by research as affecting success

in FL learning are called integrative and instrumental (Gardner 1985). The
former refers to the desire on the part of students to integrate with the speakers
and culture of the language they are studying; the latter refers instead to the
desire to study a language in order to accomplish a task, such as passing a
course or enhancing career opportunities, among other practical goals. With
various modifications and elaborations, current studies confirm that these two
factors are still significant as theoretical frameworks for assessing learning
38 Marcel Danesi

outcomes and designing teaching approaches (for example, Hong and

Ganapathy 2018).
A study by Khan and Takkac (2021) found that the instrumental factor is
still a powerful one, no matter the students’ backgrounds, with the desire for
career and economic advancement constituting a major incentive for students
to enroll in a language course. Another factor identified by the researchers was
the desire on the part of students today to become global citizens, whereby the
language is seen as an entry point into such citizenry, reshaping the integrative
factor somewhat. Interestingly, the researchers also discovered that the ways
in which the language was taught would impact motivation. And this has clear
implications, since in a complex world built on ever-changing technologies, it
cannot be assumed that the historically relevant methods will be motivational
stimuli for attracting students to a language and for facilitating learning
success. The main drawback of the methods of the past (and even the present)
is that they require the teacher’s and student’s total commitment to their
specific pedagogical philosophy. The mid-1980s Proficiency Movement in the
United States led to much discussion and interesting spin-off research on how
to integrate different methods and approaches (Omaggio 1986), stressing
integration and eclecticism. It is within this integrated framework that AI can
become a pivotal aspect of FL learning and teaching.

Enhancing Motivation

A key feature of AI is, as mentioned, the personalization of the learning

experience and thus the tailoring of the teaching process to be consistent with
varying learning styles. In a typical classroom situation, it is almost impossible
for teachers to find an approach or materials that suits one and all. With AI,
on the other hand, the specific needs of each individual student can be fulfilled
outside the classroom, since the chatbot can be used as a virtual assistant.
Moreover, AI systems allow for the collection of relevant data about learners,
their abilities, and their styles. This can then be used not only to predict future
performance but also to make effective individualized (student-centered)
education truly possible. In effect, AI-supplemented FLT allows students to
work at their own pace, at the same time that it allows them to engage with
others using the same systems, without fear of embarrassment. AI can repeat
topics, provide learners with tasks that they are capable of completing
successfully, and even take factors such as student cultural background into
account. AI-based platforms can also grade tests and evaluate essays
 Foreign Language Learning 39

automatically after submission, pinpointing errors and how to fix them.

Students can thus take immediate action to correct their mistakes and avoid
them in the future. In this way, their interlanguages may be less and less of a
factor in shaping FL-learning.
A study by Zhai and Wibowo (2022) found that chatbots had the capacity
of establishing and maintaining the interest of students, given that they had the
capacity “for dealing with learners’ emotional discomfort, the impact of
humor and the consideration of learners’ cultural backgrounds.” They also
found that chatbots have the ability to provide feedback to teachers for
identifying how learners perceive and react to the learning content.

Machine Learning

The traditional theories and models of FL-learning are not mutually exclusive;
they provide complementary bits and pieces of what may be occurring in the
brain as students learn to speak another language—that is, Transfer,
Interlanguage, Input, motivation, and pragmatic theories furnish diverse
angles from which to envision how students learn in classroom environments.
Research in so-called Machine Learning (on which chatbots are built) has
come forward to both confirm some of the ideas in these models, as discussed
implicitly in this chapter, but also to suggest different angles from which to
view FL-learning and how to approach different learning styles.
Machine Learning is based on the algorithmic mining of large databases.
The idea is to simulate the human use of real-world information in resolving
polysemy (the different meanings of the same lexical items), allowing the
algorithm to infer the appropriate meaning on the basis of probability metrics.
The algorithm searches for analogous or isomorphic forms and converts them
into options for the system. The details of how this is done are rather complex;
and they need not interest us here as such. Suffice it to say that the computer
modeling of linguistic meaning involves mining data from millions of texts on
the Internet, analyzing them statistically in terms of syntactic-semantic-
pragmatic categories.
N-gram modeling is used to predict the next item in a sequence in a chain.
The idea goes back to the founder of information theory, Claude Shannon
(1948), who asked the following question: “Given a sequence of letters, such
as the sequence “for ex…,” what is the likelihood of the next letter?” A
probability distribution to answer this question can be easily derived given a
frequency history of size n: with the letters a = 0.4, b = 0.00001, …, summing
40 Marcel Danesi

to 1.0. In the model, the probability of a word is computed by determining the

presence of a certain number of previous words and their phonemic structure.
This seems to reflect what humans do when “searching” for the right word or
phraseology. So, the next likely letter is a, since the sequence “for ex…” is
found mainly in the phrase for example. Of course, it could be some other
vowel, as in for exhilaration; but the frequency is extremely low in this case,
and moreover, the algorithm is able to determine the input context on the basis
of its analysis of its probable meaning. Overall, Machine Learning gauges the
fidelity of a possible sequence on the basis of the following:

1. the position of a word or words in a large number of texts;

2. the linguistic features typically associated with the topic or theme of
the text, which involves specific kinds of grammatical and lexical
choices;
3. syntactic considerations involving the likelihood that a certain
structure will follow or precede others.

Figure 2.3. Machine Learning (Wikimedia Commons).

Incredibly, contemporary learning algorithms have the ability to

“discover” their “own” algorithms, without needing to be explicitly prompted
as to what to do by any human-developed algorithm. As a scientific endeavor,
Machine Learning grew out of the early quest for AI and became embedded
in Deep Learning technologies, which train computers to process data in a way
that mirrors the human brain. Deep Learning models can recognize complex
patterns in pictures, text, sounds, and other data to generate accurate insights
and predictions. So, Machine Learning is really a subfield of Artificial
Intelligence and Deep Learning a subfield of Machine Learning (see Figure
2.3).
 Foreign Language Learning 41

Chatbots

Chatbots are products of research in the AI-Machine Learning-Deep Learning

integrated research paradigm. It is useful here to review the main kinds of
features that they make available for FLT purposes (to be discussed in more
detail in the remaining chapters of this book):

• The chatbot can recognize the student’s-teacher’s speech patterns. It

can thus automatically create comprehension, reading, writing, and
speaking tests and activities for users as well as grade and categorize
user responses according to proficiency level.
• Chatbots use Deep Learning tools to remind users of words and
phrases that they may require and have difficulty recalling.
• With data mining from public servers, chatbots enable live, real-
world conversations with human users. Some use virtual reality (VR)
plugins to allow the students to immerse themselves simulatively into
a real-world FL cultural context.
• Chatbots provide personalized content and practice activities. They
are virtual teaching assistants helping both students and teachers
determine what activities and instructional practices are best suited to
a given situation.
• Because of Deep Learning, chatbots are “learning companions,”
having the capacity to become increasingly familiar with users’
speech and stylistic patterns over time. As such, the chatbot offers a
level of personalization that human-based teaching cannot possibly
achieve.
• Chatbots provide instant feedback, adapting to individual needs. They
can also help the teacher create customized textbooks.

Voice recognition technology can also play a role, since it enables the
chatbot to understand and analyze the learner’s oral speech. Like the language
lab, therefore, the chatbot provides opportunities for learners to develop a
more authentic pronunciation and improve their overall listening and speaking
skills, outside of any specific lab setting. Another aspect of chatbots is their
ability to generate appropriate games in which students can become involved
for learning purposes.
42 Marcel Danesi

Now, while AI has made significant progress in aiding and enhancing

language learning, there are still limitations to the technology (at the time of
writing this book). Chatbots may struggle to understand complex language
nuances, newly-formed expressions, and cultural context. Moreover, one
cannot overemphasize the importance of human-to-human interaction in the
learning process. Engaging with other human beings in conversations, and
immersing oneself in the FL culture physically (as in immersion programs),
can provide invaluable learning experiences that AI cannot fully replicate.

Conceptualization

One of the pivotal concerns of all traditional theories of language learning is

the role of conceptualization—that is, how to grasp the concepts of the FL
culture through the language—an aspect that Robert Lado was already
emphasizing as far back as the 1950s (Lado 1957, 1964). Different languages
encode concepts often in different ways. The more distant the cultural and
historical relation between languages, the greater the conceptual differences
between them. Although people may see the same rainbow, the number and
range of the rainbow’s hues they can name will depend on how many color
terms have been encoded by the languages they speak. Some languages have
everyday words for a dozen colors; others can get by with only a couple.
Cultures leave uncoded those aspects of reality that they consider unimportant
to them, encoding only those that are important. For example, given the world
they inhabit, the Canadian Inuit have had little interest in distinguishing hues
for different types of plants; what they needed, on the other hand, were words
to talk about different kinds of snow. That is why in some Inuit language more
than 50 words referring to ice and snow are still in use. Leaving aside the
controversy surrounding this view of language, suffice it to say here that it
resonates with students and teachers alike as they grapple with different
concepts (see Danesi 2021). At the very least, the question that the foregoing
discussion raises is as follows: Can chatbots interrelate the new language to
the conceptual system of the culture that uses it?
Consider, as a simple case-in-point, the difference between orologio in
Italian and watch and clock in English. The Italian and English words refer at
a literal referential level to “a mechanical device for registering the passage of
time.” But in English the two words call attention to the “portability” of the
device as well—watches are worn or carried, clocks are put on tables, hung on
walls, etc. No such attention is necessitated by the word orologio. The notion
 Foreign Language Learning 43

of “portability” is not relevant in the use of orologio, but that of “location” is

(table clock, wrist watch, etc.). In Italian this is conveyed by da + “location”:
for example, orologio da tavolo “table clock,” orologio da polso “wrist
watch,” orologio da muro “wall clock,” etc. Now, chatbots seem capable of
contrasting languages at this level. It becomes a little more problematic to get
them to conceptualize abstract concepts such as justice and love, however.
This will be discussed in detail in chapter 5.
Overall, the work in Machine Learning is yielding some truly remarkable
computational feats, which are of obvious relevance to assessing and
validating FL-learning theories and models. As Klimová and Seraj (2023)
found in their review of the relevant literature in this ever-broadening area of
research, it is becoming ever more apparent that AI has increasing potential
“in applying and integrating the existing theories and concepts used in EFL
teaching and learning.”

Epilogue

As Max Black (1962) pointed out at the start of the AI revolution, the idea of
trying to discover how a computer has been programmed in order to
extrapolate how the mind works has borne great fruit for the study of mind
itself. But Black also expressed a caveat, namely that computers can never
truly be intelligent in the human sense because the laws of nature will not
allow it. Nonetheless, whether a computer system such as a chatbot is aware
of what it is doing intelligently, it is an obvious aid in helping us understand
what the brain does when it is processing language via simulation. AI systems
now have the capacity to learn by themselves, and it is this new autonomy that
is shedding even more light on our own brains.
Since the Reform Movement of the nineteenth century, the goal of FL
research has been to examine the ways in which we learn languages naturally
and then translate the findings into relevant pedagogy. AI clearly provides a
further possibility in this area. The essential problems of FL-learning apply to
computer programs as well and can be broken down into four components: (1)
what to say or write in certain situations, (2) when to do it, (3) how to do it,
and (4) what to expect realistically at different learning stages. The central
objective of the Reform Movement was to unify pedagogical practices by
developing a standardized curriculum as the consolidating framework for
these. AI is now an emerging, different way of envisioning FL-learning—
based on adaptability to individual needs within the context of classroom
44 Marcel Danesi

pedagogy. As such, it needs serious consideration, otherwise FL teachers will

be left behind as were the teachers using Grammar-Translation as the Reform
Movement gained momentum and spread broadly.
Chapter 3

Developing Linguistic Competence

Prologue

Whatever school of teaching philosophy to which one might subscribe, there

is consensus on what the main competencies required to use any language are.
One competency is the ability to control the structural components of a
language cohesively—its phonological, grammatical, and semantic systems—
called generally linguistic competence. A second is the ability to use the
language systematically and fluidly for expressive, interactional, and other
kinds of pragmatic functions—called communicative competence. A third
involves the ability to think with a language, and to perceive events in the
world in terms of its classificatory-conceptual structure, an ability that can be
called conceptual competence (Danesi 2017).
The purpose of this chapter is to zero in on the potential and actual
pedagogical uses of Generative AI apps, such as ChatGPT, to promote the
learning of the three main skills associated with the development of linguistic
competence in the FL—namely, the ability to use its phonology, grammar, and
vocabulary correctly and effectively. This breakdown of skills is, of course,
artificial, given that the three components are interactive in all speech and
learning events. Their separation here is simply for convenience of exposition
and discussion; however, given the nature of chatbots, not to mention natural
language understanding itself, the integration of the three skills is hardly
tangential to the discussion and will come up constantly.
ChatGPT and similar chatbots are of importance not only to students as
self-learning tools, constituting essentially interactive and highly versatile
language labs on a computer, but to teachers as well, since they can be used to
generate appropriate course content in a short time. For example, ChatGPT
can produce word lists that are related to a theme, as well as tests and
assignments. It can also customize instructional notes for the course, creating
a self-contained textbook. Consider an example provided by the website
ZenGengo (2023), which is worth repeating here as an initial case-in-point.
The objective is to teach Spanish-speaking students of English aspects of
English pronunciation that they normally find difficult. In the past this could
have been done with prepared recorded material based on pre-selected word
46 Marcel Danesi

lists in a language lab (traditional or digital). This meant that the teacher had
to conform to the contents of the pre-established drills. However, with
ChatGPT, the teacher of English can produce personalized and task-relevant
word lists by simply asking it to come up with a list based on sounds that
Spanish speakers typically find problematic pronouncing. With its ability to
mine huge databases, ChatGPT can easily come up with words that pose
typical difficulties for such students, in effect, carrying out its own Contrastive
Analysis for a specific pedagogical objective. Here is the example (ZenGengo
2023). The prompt on the site asks the chatbot to produce words whose sounds
Spanish speakers have difficulty pronouncing:

ChatGPT: Here are some examples:

1. Three
2. Think
3. This
4. The
5. Voice
6. Work
7. Ship
8. Vision
9. Cat
10. Cup

Without going into the details here, suffice it to say that ChatGPT
identified (on the site) the main problematic phonemes, as for example, the
English interdentals produced by putting the tongue between the teeth (the []
in think, the [] in this), as well as vowels such as [æ] in cat and [i] in ship. I
asked ChatGPT to come up with a similar list for English-speaking students
of Italian and found a consistency with the Spanish output. The chatbot was in
fact able to recognize even the minutest differences between English and
Italian phonology that would commonly cause pronunciation difficulties for
learners. For example, it identified the pronunciation of syllable-final /l/ as a
velar sound ([ɫ]) on the part of English speakers (as in kill and bill) as an area
of potential difficulty, given that in the pronunciation of Italian words, the /l/
occurring in the same position is pronounced as non-velar ([l]). So, the chatbot
suggested exemplary words such as quella and bella, which would allow
students to hear and then practice the different articulations of the /l/ in Italian.
 Developing Linguistic Competence 47

Pronunciation and Spelling

What is remarkable about the word lists provided by ChatGPT is the accuracy
of its phonological analysis and the pedagogical value of the words.
Essentially, it can be said that there are two teachers involved here—the
human and the machine, cooperating (so to speak) in coming up with
appropriate material for a specific area of FLT. This recalls, in a strange new
way, a form of partnership teaching that goes back to the 1940s, called the
Army Specialized Training Program, which employed two teachers in its
pedagogical system. One was a native speaker of the FL who would introduce
dialogues and conduct exercises and pattern practice drills in the FL. The other
was a linguist who provided scientific explanations of the linguistic features
of new material. The chatbot is both a drill master and linguist, forming a
partnership with human teachers who will prompt it accordingly.
In the area of pronunciation training, the chatbot itself, with a voice
plugin, constitutes an adaptive, sophisticated language lab, with which
students can interact by prompting it for appropriate words and pronunciation
exercises, receiving feedback from the machine teacher in real-time as to how
they are speaking and how they can improve their pronunciation, free of the
fear of the implicit human judgment that they may sense for their incorrect
speech. As Godwin-Jones (2022: 125) observes:

Their [chatbot] use has been shown to be effective for novice learners,
who are able to practice pronunciation and basic conversations. The
systems can serve as models of expert speakers, and, compared to
humans, possess infinite patience, allowing for extensive trial and error
without judgment. In that way, they have been shown to help overcome
anxiety and encourage a willingness to communicate.

It is remarkable to note, as the word lists and related analyses produced

by ChatGPT (above) indicate, that the AI has the capacity to conduct a
Contrastive Analysis (chapter 1) on its own, which is built clearly into its
choice of words in a list for pronunciation practice purposes. As such, this
feature of chatbots is of great pedagogical relevance because, as Stern (1983:
46) aptly put it: “Contrastive Analysis was not intended to offer a new method
of teaching; but it was a form of language description which was particularly
applicable to curriculum development, the preparation and evaluation of
teaching materials, to the diagnosis of learning problems, and to testing.”
48 Marcel Danesi

Listening and Speaking

The study of the sounds of a language and how they are structured to produce
larger forms such as words falls under the rubric of phonology, as is well
known. Phonology has been extended into the training of AI and specifically
to the development of speech recognition AI. This involves devices, which are
activated by the human voice, and have the capacity to interpret voice
commands and convert them into physical or other kinds of actions,
eliminating the need to input information manually. One of the earliest speech
recognition devices was IBM’s “Shoebox” in 1962, which had the ability to
recognize 16 different words. In 1996, the company then launched Voice Type
Simply Speaking—a software that had a 42,000-word vocabulary. In recent
years due to advancements in Deep Learning and big data mining, many
speech recognition applications and devices have become available. These
allow for the customization of the technology, from nuances of speech variants
to name recognition. Among the new abilities of speech recognition systems,
a relevant one is a software that trains a device to adapt to an acoustic
environment and speaker styles (like voice pitch, volume, and pace).
The vagaries of human speech have actually made development of speech
recognition technologies challenging. Speech recognizers are made up of a
few components, such as the speech input, feature extraction, feature vectors,
a decoder, and a word output. The decoder leverages acoustic models, a
pronunciation dictionary, and language models to determine the appropriate
output. The point here is that modern speech recognition systems are
constructed on the basis of various phonological models from linguistics, and
this is relevant to the ability of a chatbot to carry out an implicit Contrastive
Analysis. As I discovered myself (above), an AI system such as ChatGPT
allows for students to interact constructively with it by prompting it to generate
relevant word lists and derivative pronunciation drills, thus allowing for a
customization of the listening-speaking aspect that is critical to the acquisition
of the FL.
As documented by the relevant research literature (for example, Alamer
and Almulhim 2021), one of the main obstacles to successful FL learning is
the anxiety that students might feel in attempting to pronounce FL words,
which hampers the learning process because of the affective filter that students
typically bring to the classroom situation (chapter 2). Chatbots can eliminate
this obstacle, given that they allow learners to listen to new speech input over
and over according to their particular needs, so that they can acquire the new
pronunciation habits at their own pace. They allow the learners to pronounce
 Developing Linguistic Competence 49

new words repetitively as well, providing comments as to how to improve

their pronunciation, without the anxiety that learners might otherwise feel with
a classroom audience of peers and teacher listening to their efforts. Students
are usually not conscious of how they listen in their own NL, unless they
encounter external interferences, such as environmental noise. The tendency
of students is to ignore the pronunciation features of the speech input of the
FL as well, relying on the phonological habits of the NL, which is the source
of typical pronunciation errors (chapter 2). We hardly realize how important
listening is to language competency, as relevant studies indicate (for example,
Lipari 2014).
What was especially remarkable about the chatbot that I used was the
accurate phonetic quality of its pronunciation, its use of native-like prosodic-
intonation patterns, the accuracy of its pauses, and the instantaneous responses
with which it processed my input. These features are difficult to reproduce in
the same way in a traditional classroom teaching situation (Brunfaut and
Révész 2015). The voice control plugins that are available for ChatGPT allow
students to speak directly to it, and it will in turn recognize and respond to the
prompter, allowing for personalized practice in listening comprehension and
pronunciation. Some plugins even provide options for adjusting the gender of
the chatbot’s voice, and the speed with which it generates content, allowing
users to adjust it to their learning styles and communicative partner
preferences.

Writing

The relation of writing systems to phonological ones falls under the rubric of
graphemics. The term orthography is used as well, but the technical difference
is that the former refers to the correspondences between phonemes and
graphemes (writing characters) and the latter to the set of conventions for
writing a language, including norms of spelling and punctuation. Most
languages today have a system of writing. While the phonology of a language
changes over time, its orthography tends to remain stable, and thus often does
not reflect phonological changes. An example in English is the initial /kn/
cluster (as in knight, knot, know), in which the /k/ has been eliminated in
pronunciation but which the orthography has not reflected, having retained the
k letter. The /kn/ cluster was a phonological structure of proto-Germanic, from
which English is derived, and which continues to be pronounced in modern-
day German (Knopf “button”). Orthography is clearly not governed by the
50 Marcel Danesi

equivalent of sound shifts in phonology; it is regulated by external factors,

including practices related to literacy. And this is why there is so much
confusion in some orthographic systems, such as the English one when taught
as a FL—for instance, [f] stands for a voiceless labiodental fricative, while in
writing it is represented in various ways:

• the letter f in words such as fish, fun, after, flow, infer;

• the digraph ph: in words such as phonetics, phrase, philosophy,
phenomenon, triumph;
• the digraph gh: in words such as enough, rough, tough, cough, laugh.

Figure 3.1. A Page from McGuffey’s Reader (1836).

While exposure to spoken language in childhood does not require any

intervention on the part of adults, since children construct the language on
their own without training (as discussed briefly in chapter 2), learning how to
read and write requires such intervention. In English, the notion of phonics as
a method to teach reading was introduced in the nineteenth century. One of
the first examples of a phonics reader is the 1836 one by American educator
William Holmes McGuffey. Above is a page from the reader (see Figure 3.1).
Given the vastly different orthographies that might be involved in FLT,
such as for example the pictographic-based Chinese ones versus the Roman
alphabet ones used in English and other languages, this is where ChatGPT can
have specific pedagogical functions. In addition to providing feedback to
students, the chatbot can also generate individualized exercises and exposure
 Developing Linguistic Competence 51

to reading materials in the FL types focusing on orthographic patterns that are

specific to the FL. As Keir Williams (2023) has noted, ChatGPT has been used
to effectively help students identify and correct errors in their written work,
including spelling, grammatical, and vocabulary errors, while providing real-
time suggestions for corrections. Again, students can express their needs
without worrying about their individual mistakes, since the chatbot corrects
them in real time. From this type of AI interaction, students can accomplish or
observe the following:

• realize what types of errors they frequently make, so that they can
work towards improving them over time, without any fear or shame
in committing the errors;
• learn directly from their own mistakes, and thus be more inclined to
utilize the spelling and grammar suggestions provided by ChatGPT;
• view the corrections as opportunities for learning;
• receive in-depth explanations of the needed corrections, aiding in the
development of writing skills.

Finally, the ability of ChatGPT to transliterate texts in languages with

different scripts (as for example, Chinese, Hebrew, Arabic, Korean, etc.),
enables learners to practice reading out loud in the FL without having mastered
the script at any given point in time, making it consistent with the Input model
of learning (i + 1), whereby a task can be just beyond present abilities, but is
still within learning reach (chapter 2).

Grammar

Grammar differences between the NL and FL are the source of many

interlanguage errors and learning difficulties. Asking ChatGPT to write, say,
a paragraph using a specific grammar point, such as the subjunctive in French,
requesting at the same time that it highlight the relevant forms by writing them
in capital letters, as did Simon (2023), is an example of how personalized
chatbot grammar instruction and exemplification can unfold outside the
classroom. Here is Simon’s example, which constitutes an exercise providing
students with a native-like paragraph that aims to make understanding the
subjunctive much more pointed and focused:
52 Marcel Danesi

Prompt: Write a paragraph using the subjunctive in French, and

write all the subjunctive forms in capital letters.
ChatGPT: Il est essential que nous AGISSIONS maintenant pour
protéger l’environnmement. Il faut que chaque individu
PRENNE conscience de l’urgence de la situation. Il est
crucial que les gouvernements PRENNENT des mesures
concrètes pour réduire les émissions de carbone. Il est
nécessaire que les entreprises ADAPTENT leurs pratiques
pour être plus respectueuses de l’environnement. Il est
important que nous COLLABORIONS tous ensemble afin
de préserver notre planète pour les générations futures. Il
est temps que chacun CONTRIBUE à un avenir durable.

Whatever theory of grammar one espouses, it is clear that the chatbot has
the capacity to focus on specific needs and produce relevant output. As such,
it has great value in grammar teaching, given that it can compare NL and FL
grammatical points so as to bring out similarities and differences. When asked
(by myself) to explain the reasons why the subjunctive was required in the
paragraph above, ChatGPT put forth explanations that indicated how a verb
or expression in the main clause triggered the subjunctive in the dependent
clause. When asked to explain the difference between the two clauses, it gave
a truly concrete explanation with further examples. It indicated in its responses
that the main clause must have the relative pronoun que or qui to introduce the
dependent clause (“Il est essential que nous agissions maintenant pour
protéger l’environnmement”) and that the subject of the main clause and of
the dependent clause must be different.

Words and Sentences

One area that is highly relevant in FL-learning is word-structure recognition

in the new language, that is, the ability to recognize how words are constructed
with morphemes or lexemes—the former refer to any type of meaning-bearing
unit with grammatical function, and the latter with lexical function, although
the term morpheme often covers both. A way to conceive a lexeme is as a
minimal free form—a form that can stand on its own. For example, the word
logic is a minimal free form, because it conveys a single piece of meaning and
cannot be broken down further. The form illogical is also a word, but it is
 Developing Linguistic Competence 53

composed of the same lexeme and two bound morphemes—the prefix il- and
the suffix -al.
The connection between words, how they are formed, and how they are
used to create or insert into larger structures such as sentences was studied by
philosophers already in antiquity. The Greek philosopher Plato wrote the
following relevant comment in his Cratylus (Plato, Project Gutenberg 1999):
“sentences are, I conceive, a combination of verbs and nouns.” Plato’s pupil,
Aristotle, added another class of words to sentence construction, namely
conjunctions. These are among the first ways to classify words in terms of
syntactic categories, that is, in terms of how they function in sentences—thus
laying the foundations for the notion of grammar, defined essentially as a set
of rules that describe how words are organized in phrases and sentences. FL
learners are faced with a daunting task, which Aristotle himself identified: how
to organize words according to rules of combination into phrases and
sentences. The term word class is used commonly to refer to the category in
which a part of speech belongs. Word classes may be open or closed: open
classes, which typically include nouns, verbs and adjectives, acquire new
members constantly, while closed classes, such as pronouns and conjunctions,
rarely do so, if at all. All languages have similar classes of nouns and verbs,
with the same syntactic functions, but beyond these there are often challenging
specific grammatical differences for learners to recognize: Japanese has three
classes of adjectives; Chinese, Korean, Japanese and Vietnamese have a class
of nominal classifiers (a word or affix that accompanies nouns classifying
them according to referent); and so on.
Grammar teaching and practice is an area in which chatbots can be used
effectively, since they can easily segment words into their morphemes and
lexemes and, with relevant prompts, can highlight which aspect of FL
grammar requires more study and attention on the part of the learner. A chatbot
can also easily explain what functions words play in sentences. The main
pedagogical prompts that may be of utility in this area of FLT are the
following:

• morphology: prompting the chatbot to identify FL lexemes and

morphemes for students in terms of how they are used to construct FL
words;
• syntax: prompting the chatbot to assign FL words to syntactic
categories, so as to indicate to students how they function in FL
sentences and larger structures (such as paragraphs);
54 Marcel Danesi

• categorization: prompting the chatbot to identify for the students how

a specific type of word is categorized according to its structure, such
as the endings that might identify a word as a noun;
• semantics: prompting the chatbot to indicate to the students how
meaning is assigned to words and larger structures, based on a large
corpus of FL words in context;
• coreference: prompting the chatbot to explain coreference systems to
students, such as how pronouns replace nouns in specific ways.

The latter point is an important one, since coreference is one of the hardest
features to learn about the grammar of the FL. To get a sense of the task at
hand, consider coreference in English, as realized with the following two types
of grammatical devices: (1) anaphoric devices, which refer back to some word
or phrase, as the use of she referring back to Sophia, in a sentence such as
“Sophia went to the mall, where she wanted to do some shopping;” and (2)
cataphoric devices, which anticipate some word or phrase, as for example, in
the sentence “Even though she will deny it, I tell you that Mary did it.” Subject
and object pronouns, locative particles, demonstratives, adverbs, and other
kinds of grammatical classes, can function as anaphoric and cataphoric
devices in conversations. There are other kinds of coreferential structures,
which need not concern us here. The point is that an aspect of grammar such
as coreference in the FL might take a lot of time and effort to teach and, on the
part of the students, to acquire. The chatbot can be used in such areas to
maximize teaching and learning time, as well as providing opportunities for
practice and reinforcement, given that students can work with the chatbot to
get relevant practice in this area and to receive concrete explanations of how
coreference works in the FL.

Grammar and Usage

As the foregoing discussion implied, ChatGPT is highly useful as a grammar-

teaching partner to both the teacher and student. Simon (2023), for instance,
provides the following example of how a chatbot can provide appropriate and
immediate feedback related to grammar in the case of someone trying to learn
English:

Prompt: Provide feedback on my English: “Yesterday, I go to the

pool and had fun swim with all the friends.”
 Developing Linguistic Competence 55

ChatGPT: Here’s a revised version of your sentence with some

corrections: “Yesterday, I went to the pool and had fun
swimming with all my friends.” Overall, your sentence is
understandable, but there are a few grammatical errors.
“Go” should be changed to “went” to match the past tense
of the sentence. Additionally, “swim” should be changed to
“swimming” to use the correct verb form. Finally, it’s more
common to say “all my friends” instead of “the friends.”

As this example shows, and as I myself confirmed with similar prompts,

the chatbot is an expert grammarian. I asked it to evaluate the sentence “I goed
to the store yesterday,” to which it responded: “The correct form is ‘went’
instead of ‘goed.’ The correct sentence is ‘I went to the store yesterday’.” It
then informed me that “went” is an irregular form of the verb that breaks away
from the normal verb conjugation paradigm. ChatGPT is clearly able to tackle
any grammar query and provide personalized examples and drills responding
to prompts such as: “What is the correct verb tense in this sentence?” “Why is
this tense required?” or “How can I make this sentence more understandable
and native-like?.” Here are a few more illustrative examples (from:
fhttps://polyglotclub.com/wiki/Language/Multiple-languages/Culture/How-
to-learn-a-language-with-ChatGPT):

Prompt: What is the verb tense in this sentence? “She will be visiting
her grandparents next week.”
ChatGPT: The verb tense used in this sentence is future continuous.
Prompt: How can I make this sentence more concise? “I am hoping
that maybe in the future, I can possibly attend the
university.”
ChatGPT: You can make this sentence more concise by saying “I hope
to attend the university in the future.”
Prompt: I’m writing an email to a potential employer. Can you
check if my sentence structure and grammar are correct?:
“Dear hiring manager, I is excited to apply for the position
at your company.”
ChatGPT: The correct form is “am” instead of “is.” The correct
sentence is “I am excited to apply for the position at your
company.”

ChatGPT’s responses are never abstruse, as can happen sometimes in

traditional textbook explanations of grammar; they are pertinent to the user’s
56 Marcel Danesi

specific requests and learning requirements. To use another example, this time
in the case of learners of Spanish as a FL (Caulfield 2023), let us suppose that
a student does not grasp the reason why a pronoun subject may be omitted in
Spanish sentences:

Prompt: In the sentence “Viajo a Mexico,” what is the subject?

ChatGPT: In the sentence “Viajo a Mexico,” the subject is the pronoun
“yo” (I). The verb “viajo” (I travel) is conjugated to match
the subject.
Prompt: Why is the subject not included in the sentence?
ChatGPT: I apologize for the confusion in my previous response. In
the sentence “Viajo a México,” the subject is actually
implied and not explicitly stated. The verb “viajo” (I travel)
is conjugated in the first-person singular form, which
indicates that the subject is “yo” (I). In Spanish, it is
common to omit the subject pronoun when it can be
inferred from the verb conjugation and the context.
Therefore, even though the subject pronoun “yo” is not
present in the sentence, it is understood that the speaker is
referring to themselves as the subject.

Not only is this explanation accurate, but it is articulated in response to

the specific prompts used, making it much more effective for the individual
learner. Another important function of ChatGPT is that it can create
personalized grammar exercises on a topic of one’s choice, as illustrated below
(Caulfield 2023):

Prompt: Create a Spanish exercise with 5 incomplete Spanish

sentences to practice subject-verb agreement. Example
exercise sentence: Ella _______ como mecánica. Example
solution: Ella trabaja como mecánica. Please don’t present
the solutions in the exercise itself. Instead, I would like you
to include a separate section below the exercise, which
shows the correct solutions.
ChatGPT: Exercises
1. Nosotros _______ al parque todos los domingos.
2. Tú _______ muy bien en el examen.
3. Ellos _______ libros en la biblioteca.
4. Mi hermana y yo _______ al cine esta noche.
5. La profesora _______ muy amable con sus alumnos.
 Developing Linguistic Competence 57

Correct Solutions:
1. Nosotros vamos al parque todos los domingos.
2. Tú te desempeñaste muy bien en el examen.
3. Ellos leen libros en la biblioteca.
4. Mi hermana y yo iremos al cine esta noche.
5. La profesora es muy amable con sus alumnos.

Such exercises can then be followed by translations into English and

comparisons and contrasts made by the chatbot between the NL and FL, as
prompted. Without going into further examples and details here, suffice it to
say that chatbots can easily tailor the FL grammar to specific learner needs
and can also provide teachers and students alike with analyses, examples, and
review materials that cannot possibly be predicted by pre-fixed print materials
or established curricular plans. All this raises the possibility of truly
transforming traditional FLT, as Richard Campbell (2023) has aptly observed:

Artificial intelligence has the potential to transform conventional

teaching strategies in the field of language learning. Language learners
can practice their language skills through interactive conversations in a
more personalized and interesting way with AI-powered languages
learning tools like chatbots and virtual assistants. These tools use
algorithms for natural language processing to identify speech patterns
and offer immediate feedback on pronunciation and grammar issues.

Needless to say, there are some caveats. While chatbots have a truly
remarkable ability to detect and correct a large array of grammatical errors,
they cannot cover all of a language’s grammar (at present), as Fang, Yang,
Lan, Wong, Hu, Chao, and Zhang (2023) found, a fact that learners and
teachers should take into account:

Our experimental results and human evaluations demonstrate that

ChatGPT has excellent error detection capabilities and can freely correct
errors to make the corrected sentences very fluent, possibly due to its
over-correction tendencies and not adhering to the principle of minimal
edits. Additionally, its performance in non-English and low-resource
settings highlights its potential in multilingual tasks. However, further
analysis of various types of errors at the document-level has shown that
ChatGPT cannot effectively correct agreement, coreference, tense errors
across sentences, and cross-sentence boundary errors.
58 Marcel Danesi

Vocabulary

The type of pedagogical support that ChatGPT offers in the domains of

pronunciation and grammar applies as well to the teaching and learning of FL
vocabulary. The chatbot can generate a vocabulary list on a specific topic with
just a few prompts, as discussed and illustrated at the start of this chapter. Once
the vocabulary list has been produced, ChatGPT can assist in testing the
knowledge gained by creating various types of related exercises and drills,
including illustrations of how words are used with differing semantic nuances
in sentences or within larger texts, thus providing to the student the possibility
to form a deeper understanding of their meaning and usage. For example,
ChatGPT can be asked to incorporate FL words into sentences, paragraphs,
reading passages, or even dialogues, as Simon (2023) has shown in the case
of an advanced French course:

Prompt: Create a text using high frequency vocabulary on the topic

of politics in French. Highlight the vocabulary words.
ChatGPT: La politique est un domaine complexe et crucial de la
société. Les élections, partis politiques, et
gouvernements sont des éléments centraux de ce système.
Les citoyens jouent un rôle important dans la démocratie,
où ils expriment leur voix par le biais du vote. Les
politiciens, élus par le peuple, assument des responsabilités
telles que légiférer, gouverner, et représenter les intérêts
de leurs électeurs.

ChatGPT can also assist learners by responding concretely to a wide range

of vocabulary questions, such as “What is the meaning of this word?,” “Can
you give me synonyms for this word?” and so on. Follow-up practice material
generated by the AI can then help learners practice different aspects of word
and phrasal or idiomatic meaning.

Polysemy

Note the differential meanings of cat in these four sentences:

1. “My cat has a long tail”

2. “My cat is a wonderful companion.”
 Developing Linguistic Competence 59

3. “Jazz musicians are cool cats.”

4. “They let the cat out of the bag.”

The first sentence has literal, or more technically, denotative meaning,

whereby the word cat refers to a specific kind of feline mammal. The second
uses the word cat with a specific cultural connotative meaning, namely as a
pet. The third sentence shows the use of cat in a figurative way, associating its
perceived sleek movements with certain types of musicians (discussed further
in chapter 4). And the meaning of cat in the fourth sentence is also figurative,
but in a different way. In this case, it uses the animal as a metaphor for secrecy,
likely associated with how cats are able to hide from danger. In effect,
deciphering the meaning of words entails knowing the differences among
denotative, connotative, and figurative dimensions of meaning and the kinds
of relations that hold among words and cultural knowledge.
The various meanings of a word such as cat are called polysemic. Can a
chatbot make these distinctions? It is in this area that chatbots show some
difficulty, although with neural networks and Deep Learning, they can still
come up with an appropriate meaning by deciphering the contextual cues that
are present in a prompt. Nevertheless, polysemy is tough to compute for a
chatbot, hence the need for a human teacher in this domain of vocabulary
learning. Chatbots rely on the programmed information and data on which
they have been trained. If the chatbot’s training does not include the specific
data on FL polysemic variability, it may not be able to provide a relevant
response.

Thematic Vocabulary

As in the case of pronunciation and grammar, a chatbot can be prompted to

generate theme-specific vocabulary for learning or instruction. For instance, I
asked ChatGPT the following:

Prompt: Create an Italian–English vocabulary list on the theme of

cooking.
ChatGPT: Sure! Here’s an Italian-English vocabulary list on the
theme of cooking:
1. Cucinare: Cook
2. Ricetta: Recipe
3. Forno: Oven
60 Marcel Danesi

4. Stufa/Fornello: Stove
5. Padella: Pan
6. Pentola: Pot
7. Coltello: Knife
8. Forchetta: Fork
9. Cucchiaio: Spoon

After that, I asked ChatGPT if the word padella can be used in idiomatic
expressions, and it responded with one example, dalla padella nella brace
(“from the pan into the fire”) and, as far as can be told, this is certainly the
most frequent idiomatic use of the word, if not the only one. I then prompted
ChatGPT to find at least one idiomatic expression for each of the other words,
and it came up with accurate suggestions as well.
Clearly, ChatGPT is a useful tool for thematic vocabulary learning or
teaching. The chatbot can suggest words and phrases that are thematically-
focused and explain their range of uses, helping learners construct sentences
using those words. With an appropriate plugin it can also provide metaphorical
ways of saying something—a topic to be discussed in chapter 5. As Halder
(2023) has shown, with the appropriate plugin, ChatGPT can generate
utterance-appropriate metaphors in English, such as the following:

Prompt: Generate a metaphor for overcoming life’s obstacles.

ChatGPT: “steep hill to climb” (among others)
Prompt: Generate a metaphor for love.
ChatGPT: “love is magic” (again among others)
Prompt: Generate a metaphor for a strong friendship.
ChatGPT: “solid friendship” (once again among many others)
Prompt: Generate a metaphor for hope.
ChatGPT: “a journey with no end” (along with many other
suggestions)

The prompts can be followed-up with requests for ChatGPT to use the
metaphors in relevant examples and to indicate any synonymous expressions.
Given its ability to navigate large amounts of data the chatbot can come up
with all kinds of relevant conversation-appropriate thematic expressions. Here
are some examples from Spanish (Caulfield 2023):

Prompt: How do I say the following sentence in Spanish? “I often

travel to Mexico for work and go on vacation there.”
 Developing Linguistic Competence 61

ChatGPT: To say “I often travel to Mexico for work and go on

vacation there” in Spanish, you can say: ”Viajo a menudo
a México por trabajo y también voy de vacaciones allí.”
Please note that translations can vary depending on the
context, and this translation assumes a general meaning.
Prompt: Are there other ways to express the same information in
Spanish?
ChatGPT Certainly! There are multiple ways to express the same
information in Spanish. Here are a couple of alternative
ways to convey the sentence: “Con frecuencia viajo a
México por trabajo y también voy de vacaciones allá.”
“Regularmente viajo a México por motivos de trabajo y
también aprovecho para vacacionar allí.” “A menudo viajo
a México por cuestiones laborales y también me tomo
vacaciones allí.” These variations maintain the same
meaning but present slight differences in wording and
sentence structure.

As in other areas of the FL, ChatGPT can follow-up with interactive

vocabulary games and exercises, such as word matching or fill-in-the-blank
sentences. And it can be used by teachers to create flashcards with the required
type of vocabulary words and phrases in the FL, along with their translations
and pertinent illustrative images, so as to help language learners to memorize
new words effectively.

Epilogue

The ability of chatbots to analyze learner data, adapt to the learner’s needs,
and provide personalized feedback and guidance in the primary areas of
linguistic competence development (pronunciation, grammar, vocabulary) is
what makes the AI highly useful as a complementary teaching-learning tool,
at the very least. AI-generated content can also incorporate interactive features
such as speech recognition and pronunciation feedback to help learners
practice and improve even these psycho-motor skills. Overall, AI can greatly
help students develop linguistic competence in the FL, in a way that is tailored
to meet their specific needs and adapt to their learning styles.
Traditional FL learning materials, technologies, and methodologies,
including textbooks and audio CDs, and even digital language labs, are static
62 Marcel Danesi

to varying degrees, providing an established sequence of linguistic items to be

taught according to a pre-established curricular script. Ever since the Direct
Method, the underlying objective has been for FLT to provide a “one-size-fits-
all” approach that consisted of limited opportunities for the student to interact
with the material individually or receive personalized feedback without
anxiety. On the other hand, AI systems such as ChatGPT use Machine
Learning technologies that allow them to carry out specific or task-based
analyses of learner data, which can then be used to make the teaching-learning
experience a responsive one to actual needs. Another key difference between
AI-assisted FLT and traditional FLT is the ability of chatbots to incorporate
real-world examples and current events into the learning experience. AI
algorithms can process vast amounts of data to identify relevant and up-to-
date examples of the FL used in practical situations.
The primary benefits of an AI-shaped pedagogy of FL pronunciation,
grammar, and vocabulary can be summarized as follows:

• Personalization: ChatGPT can identify learner problems and learner

styles, and thus adapt the pedagogical materials and instructions it has
the ability to generate according to their individual needs.
• Real-time feedback: As discussed throughout this chapter, chatbots
can provide learners with real-time feedback on their pronunciation,
grammar, and vocabulary usage. This can help them identify areas for
improvement and allow them to make corrections as they learn.
• Affective filter: As Krashen emphasized (chapter 2), students often
have an affective filter that prevents them from fully acquiring the
input to which they are exposed, and they may uneasy as to what to
say in a classroom environment. Interaction with a machine not only
lowers the filter, but eventually may eliminate it completely.
• Interactivity: Chatbots can include interactive features such as speech
recognition, thus providing learners with the opportunity to practice
their language skills in a sophisticated language lab type context.
• Real-world examples: AI algorithms can analyze vast amounts of data
to identify relevant and up-to-date examples to help learners
understand how the FL language is used in (changing) real-world
contexts.
• Gamification: AI has the ability to create games on the spot, and adapt
them to the learning level and preferences of the learners.
 Developing Linguistic Competence 63

The last point requires some elaboration. Among the relevant findings
related to the use of puzzles and games in FLT the following stand out (for a
summary of the research, see Danesi 2022):

• Puzzles stimulate creative thinking, thus helping students unpack

hidden information in them on their own via the problem-solving
format.
• They motivate students to explore the new notions on their own.
• Most puzzles reflect content that is easily understandable, no matter
what the student’s cultural background, which makes them useful as
classroom devices. Since the chatbot can adapt to the student’s level
of knowledge, it can produce puzzles in the FL that can greatly
enhance the acquisition process as per the Input Hypothesis (chapter
2).
• Puzzles can be tailored to reinforce specific learning objectives and
chatbots are clearly expert at doing so, creating puzzles in terms of
level grammar and vocabulary achieved by a student, and according
to the individual student needs.
• When combined with other pedagogical materials, puzzles thus
complement the learning process in a supportive way.
• Games and puzzles that are found commonly in the FL culture can be
adapted profitably for the enhancement of learning the language.
Since chatbots can extract these from the Internet and adapt them to
the learning situation at hand, they are particularly useful in creating
such culturally-based ludic content.

As mentioned throughout this book, putting the enthusiasm for using

chatbots aside, given their versatility and adaptability, there are caveats. The
main one is that the best way to practice a new language is to speak it with
another human, and hence the importance of a human teacher in facilitating
and bolstering the learning process (discussed further in chapter 6). ChatGPT
may not, in effect, always be the best conversation partner since it might not
be able to pick up on contextual clues, as Richard Campbell (2023) has aptly
pointed out:

AI is a tool that is gaining popularity among language learners. The

absence of interpersonal interaction is one of the main negatives, though.
While AI can give students access to a variety of knowledge and
64 Marcel Danesi

resources, it cannot take the place of the advantages of speaking with a

human teacher or conversation partner. The importance of human
interaction in learning a new language cannot be overstated, despite the
fact that AI can clearly improve language learning experiences in many
ways.
Chapter 4

Promoting Communicative Competence

Prologue

A common technique in FLT, used in most pedagogical approaches, is the

dialogue, which is typically designed to present new information in simulated
conversational ways. Traditional print textbooks generally include a dialogue
per chapter based on the new vocabulary and grammar introduced within it.
Because print is static, the dialogue in the textbook cannot be immediately
updated or adapted to specific needs, and thus has diminishing value as a
pedagogical tool over time. With ChatGPT, on the other hand, dialogues can
be constantly updated according to changing conversational styles in the FL
and the content within them chosen according to changing lexical trends in the
culture. Moreover, unlike the print dialogue, a chatbot-generated one can be
tailored to meet specific learner and course needs. Consider the following
example (from the website: https://sanako.com/how-will-chatgpt-influence-
and-impact-language-education), whereby ChatGPT has been prompted to
provide a beginner-level dialogue in Spanish. It presents thematically-relevant
vocabulary and grammar, based on a specific communicative theme—making
contact:

Prompt: Create a beginner-level dialogue in Spanish between two

friends discussing their favorite hobbies.
ChatGPT: Personaje 1: Ana
Personaje 2: Luis
Ana: ¡Hola, Luis! ¿Cómo estás?
Luis: Hola, Ana. Estoy bien, ¿y tú?
Ana: Estoy muy bien, gracias. Oye, ¿cuáles son tus pasatiempos
favoritos?
Luis: Me gusta mucho leer libros y escuchar música. ¿Y tú, Ana?
¿Qué te gusta hacer en tu tiempo libre?
Ana: A mí me encanta pintar y cocinar. Además, disfruto mucho
de ver películas.
Luis: ¡Qué interesante! ¿Qué tipo de películas te gusta ver?
66 Marcel Danesi

Ana: Me gustan las películas de comedia y aventuras. Siempre

me divierten mucho. ¿Y a ti, qué tipo de música te gusta
escuchar?
Luis: Me gusta escuchar música pop y rock. Me ayuda a
relajarme y a pasar un buen rato.
Ana: ¡Eso suena genial! Tal vez algún día podríamos cocinar
juntos y escuchar música mientras lo hacemos.
Luis: Claro, me parece una idea estupenda. Será divertido
compartir nuestros pasatiempos.
Ana: ¡Perfecto! Hasta entonces, ¡nos vemos!
Luis: ¡Hasta luego, Ana!

Note that ChatGPT created appropriate characters, with common Spanish

names, who spoke in an informal style, with the relevant idiomatic
expressions, conveying joviality and familiarity. Now, the details of the
dialogue can be changed by different prompts, such as asking ChatGPT to
utilize the topic of different hobbies that may be more interesting to a specific
learner or even class, making it even more thematic or individualized
according to specific student or course needs.
In effect, AI is transforming the dialogue into a truly adaptive and
constantly updated interactional form, ensuring that its conversational
structure and content are always relevant to what is occurring in the FL culture.
As such, it is possible to make the dialogue a non-static pliable technique
designed to help students develop communicative competence in the FL. As
discussed in previous chapters, communicative competence is a term that Dell
Hymes coined originally in 1966 to complement the idea of linguistic
competence, implying that there is a dynamic interaction between language
structure and language use. This notion had immediate impacts on FLT, also
as discussed, changing its focus away from the strict concept of method, as in
the DM and ALM, to that of classroom interaction, so that students can learn
by conversing with each other and with the instructor. Learners are expected
to converse about personal experiences, and instructors to deal with
conversational topics directly. However, even in such an open pedagogical
environment, the dialogue concept was retained to shape derivative
interactions, and it continues to be a significant pedagogical tool in the current
AI age, as the above example shows.
This chapter will deal with the topic of how AI can help promote
communicative competence in the FL. It is remarkable that ChatGPT can
understand and respond to human speech in a natural way, and can thus be
 Promoting Communicative Competence 67

used to create increasingly advanced dialogical-conversational materials. It is

a “talking language lab,” so to speak, that teachers and students can take with
them anywhere and at any time. It is also a “virtual assistant” to both learners
and instructors, given its the ability to learn from its interactions with users,
and to respond to inquiries in a relevant way.

Communicating in a Foreign Language

As mentioned in chapter 2, it was during the 1970s that interest in converting

psychological and linguistic theories into pedagogical practices, which started
with the Direct Method, had waned considerably in FLT. Teachers started
questioning the implicit goal of all the methods at the time—the development
of FL skills systematically and sequentially, from listening and speaking to
reading and writing. Rarely, they emphasized, did this view of FLT lead to
learners acquiring the ability to use the FL by the end of a course of study to
do things with it in the real world, as, for example, how to be polite, how to
make social contact, how to express oneself in specific situations, etc.
Imparting this kind of know-how was not taken into account in any systematic
way by any of the methods, with a few exceptions. Language educators at the
time started even to look at history to see if it could offer any insights into
solving this dilemma. From this interest, works by educators such as the Czech
teacher Comenius in the seventeenth century, came to the forefront of the new
debate in FLT. Comenius had claimed that students learned best when they
had to decipher and produce dialogues reflecting real life interactions. From
these, he suggested, students could easily induce the appropriate linguistic and
cultural rules on their own.
The overall result of this new Zeitgeist was a focus on communication in
the FL. It is interesting to note that chatbots are designed to carry out
dialogues, as if they were themselves learners of the language spoken
according to communicative-teaching principles. For this reason, among
others, they constitute effective tools for promoting communicative
competence, given their own design. Made-to-order dialogues such as the
Spanish one above are arguably effective as communication-promoters
because they are prompted by teachers and students alike, and they are based
on the most updated language and conversational styles that are occurring in
the FL culture.
68 Marcel Danesi

Communicative Competence

To reiterate, it was the linguist Dell Hymes (1966, 1971) who indirectly
brought about the end of the method notion in FLT when he argued
convincingly that languages were hardly impervious to influences from real-
world social interaction. He suggested that, on the contrary, verbal structures
were shaped, and over time even modified, by language use. Moreover, he
argued that the ability to apply language to specific social situations in a
meaningful way constituted a different kind of competence, calling it
communicative competence. Hymes’s term held an instant appeal to teachers,
who at the time were also beginning to suspect that the method notion was a
flawed one. The follow-up research on communicative competence led, as a
consequence, to a paradigm shift in FLT—away from the focus on linguistic
competence to a focus on communicative competence and, a little later, to an
amalgam of both competencies, known by the mid-1980s as the Proficiency
Movement, a standards-based system of instruction and assessment that
requires learners to demonstrate competence in targeted knowledge, skills, and
abilities before moving onto the next level or challenge, no matter what
pedagogical approach was used to get them to that level.
The first steps to develop communicative syllabi were taken by the
Council of Europe (Van Ek 1975, Wilkins 1976). Their organizing principle
was the concept of speech act, defined as a communicative strategy that native
speakers employ unconsciously to carry out specific types of social
functions—a notion that goes back to John Austin’s 1962 book, How to Do
Things with Words. A simple social protocol such as saying hello or good-bye,
for instance, constitutes a speech act, because it is based on an implicit detailed
knowledge of the appropriate words, phrases, structures, and nonverbal cues
that come together cohesively in a script-like fashion to enable a speaker to
carry out a successful speech interaction with another speaker. An
infringement of any of the procedural details of this script might lead to a
breakdown in communication.
Now, the question becomes: Can chatbots be used specifically for the
development of communicative competence, assuming that they have the
ability to differentiate between the functions of specific speech acts, as the
example in Spanish above suggests? Leaving aside anomalies that I myself
noted when prompting ChatGPT to create dialogues in different situations in
Italian, the chatbot was able to generate acceptable dialogues used in typical
social situations such as ordering a meal at a restaurant, carrying our affairs at
a bank, or simply chatting with an acquaintance. When I compared these with
 Promoting Communicative Competence 69

textbook dialogues on the same themes, I found that the chatbot had updated
the language and speech strategies considerably to reflect contemporary talks.
The same pattern of findings are discerned in similar uses of chatbots in other
languages. Simon (2023) provides an example of a restaurant-based
interaction with ChatGPT in French:

Prompt: Je vais prendre la salade méditerranéenne.

ChatGPT: Parfait, la salade méditerranéenne est un excellent choix!
Voulez-vous une vinaigrette spécifique avec votre salade?
Nous avons une vinaigrette à l’huile d’olive et au citron, ou
une vinaigrette balsamique. Ou peut-être préférez-vous une
autre option?
Prompt: Je vais prendre une vinaigrette balsamique.
ChatGPT: Très bien, je vais vous apporter une salade méditerranéenne
avec une vinaigrette balsamique. C’est un choix délicieux!

Not only did the chatbot simulate what an actual server would say in a
French restaurant, providing the same kind of optional paths that a
conversation might take, which led the prompter to respond according to the
options, but it also used the appropriate formal register (with the relevant vous
pronoun and related verb forms) as well as the formal social tone required in
such a situation. Now, while a chatbot can provide cultural and
communication-appropriate conversation practice, it is always important to
keep in mind that it is an AI system and may not be able to fully replicate the
experience of interacting with a human interlocutor, such as a waiter. Human
conversation partners bring cultural, contextual, and bodily cues and nuances
that an AI model does not possess—a topic discussed further in chapter 6.

Role of Context

The role of context in promoting communicative-conceptual accuracy cannot

be overestimated. Context shapes meaning and interpretation. The human
mind always interprets information, not just processes it, in contextualized
ways. For this reason, it can unpack any ambiguity in the use of words.
Linguist Yehoshua Bar-Hillel (1960) was among the first to use an example
of linguistic ambiguity that came to be known as the Bar-Hillel Paradox to
show that the early versions of Machine Learning programs were incapable of
deciphering contextualized constraints on meaning. As Bar-Hillel showed,
70 Marcel Danesi

humans use extra-linguistic (contextualized) information to make sense of

messages that computers at the time (the late 1950s) were not be capable of
accessing in the same way. In other words, context was identified as a
determinant in how humans understand words and interpret their meanings.
His example is as follows:

1. The pen is in the box (= the writing instrument is in the container)

2. The box is in the pen (= the container is inside another container
[playpen])

Humans can distinguish meaningfully between the two utterances because

they have access to outside information about the meaning of the word pen.
As this example shows, and as discussed in the previous chapter, polysemy is
an inbuilt feature of human language, which produces ambiguity that is
resolved by real-world knowledge. Bar-Hillel’s Paradox led shortly thereafter
to the serious study of extra-linguistic inferences in human discourse. Indeed,
it can be argued that it was the starting point for the growth of pragmatics and
conversation analysis as major branches of linguistics. Overall, Bar-Hillel’s
Paradox brought out the importance of real-world context in determining the
human meaning of forms such as words. In order for a fully-automatic
Machine Learning system to resolve Bar-Hillel’s sentences correctly, it would
have to have some rule subsystem that would indicate that: (a) if pens refers
to writing instruments, then they are (typically) smaller than boxes; (b) if pens
refers to large containers, then boxes can be put into them (if smaller); and (3)
that it is impossible for a bigger object to be contained by a smaller one.
Actually, chatbots can now resolve context-based ambiguity almost
systematically, given the Deep Learning and Natural Language Programming
systems on which they are based. These allow the chatbots to conduct both an
“internal linguistic analysis” of the grammatical and lexical items requested
by prompts and then an “external linguistic analysis” of the real world contexts
that constrain the selection and concatenation of the items. Nevertheless, the
guesswork involved in AI systems is vastly different than the one involved in
human inferences. Humans make their hunches on real contextualized
experiences in the world, algorithms involve probability models of such
experiences. The two are isomorphic systems, not identical ones.
In effect, Machine Learning has made many advances in this area that
have led to AI systems today which deal with polysemy and other ambiguities
to high degrees of accuracy. The algorithm starts by making inferences about
the appropriate meaning of a word in terms of frequency distribution measures
 Promoting Communicative Competence 71

at the so-called “first layer” in the neural network. The algorithm then searches
for contextualized options at the “second layer.” Frequency again determines
the appropriate option for its output at the “third layer.” The details of how
this is done are rather complex; and need not interest us here as such. Suffice
it to say that the chatbot’s grasp of the correct meaning involves mining data
from millions of texts on the Internet, analyzing the words within them
statistically in terms of contextualized patterns, and then selecting the required
meaning according to the prompt. The reason why the chatbot is so accurate
is that the neural networks on which it is trained can access a billion to a trillion
examples of real conversations. The chatbot then uses its knowledge about
syntax in order to generate appropriate responses.
A key aspect of human communication is the ability of interlocutors to
make a conversation flow logically and coherently, taking contextual variables
into account, such as social register and content shifts. ChatGPT approximates
the flow by using a combination of techniques, which appear to mirror how
humans keep conversations flowing in a meaningful way. These include the
following:

• Dialogue state tracking: ChatGPT can keep track of the context and
previous content of the conversation, including previous queries and
responses, and then uses this information to understand the current
state of the conversation and generate responses that are relevant to
the flow. In the French restaurant scene above, the chatbot understood
how to add to the content of the customer’s order, suggesting
balsamic vinegar and then, remembering this, was able to connect the
suggestion to the request.
• Response selection: ChatGPT can select the most appropriate
response from a repertory of possible responses based on the context
and content flow of the conversation, using advanced Machine
Learning algorithms to do so, which enable it to take into account
factors such as the relevance and coherence of the response. The
whole restaurant scene above is grounded on the response selection
system built into chatbots.
• Response generation: ChatGPT has the ability to generate natural
language responses that are grammatically and semantically correct.
It achieves this by using a neural network architecture that is capable
of generating responses that are both relevant and verisimilar to real-
72 Marcel Danesi

life situations. Aspects of the architecture were described

schematically above.

Conversations

Conversations occur when people meet and interact in some (usually)

purposeful way. They involve the ability to communicate coherently
according to context. Teaching this in a FL involves keeping in mind the fact
that even in promoting communicative competence there is bound to be NL
transfer, whereby the conversational scripts between the NL and the FL could
be isomorphic (positive transfer) or different (negative transfer). Passing on
the ability to converse in the FL, however, involves a different type of
Contrastive Analysis—one that looks at the speech act forms used in the NL
and the FL in order to determine potential areas of negative transfer. If two
people decide to have coffee together at a coffee bar, then it would be
anomalous or bizarre if they did so in silence in American culture; rather, small
talk (at the very least) is expected in this situation. But this is not so in other
cultures, where small talk may be seen as anomalous or even insulting. The
point to be emphasized is that a conversation is not realized necessarily in the
same way across languages. Weather is a common topic in regions where the
climate is unpredictable. Questions about the family are usual in many Asian
and Middle Eastern countries. In some cultures, small talk between new
acquaintances may feature exchanges based on the differential social class
structure of the interlocutors. Interestingly, some countries, such as Finland
and Sweden, utilize small talk sporadically, being more comfortable with
silence.
An overarching law of pragmatic interactional behavior is, in fact, that
human conversation is highly variable along different cultural axes. ChatGPT
has the ability to come up with variants that are culture-specific and even
dialectal variants in a FL, if asked to do so, as can be seen, for instance, in the
following example (from Caulfield 2023):

Prompt: I am learning Spanish because I am planning to spend some

time in Peru. Can you give me a list of common idioms that
are unique to Peruvian Spanish, with English translations?
ChatGPT: Certainly! Here are some common idioms that are unique
to Peruvian Spanish, along with their English translations:
“Estar en la luna”—To be on the moon (meaning to be
 Promoting Communicative Competence 73

absent-minded or daydreaming); “Estar en la papa”—To be

in the potato (meaning to be in a good situation or feeling
great); “Hablar hasta por los codos”—To talk through one’s
elbows (meaning to talk a lot). These idioms are specific to
Peruvian Spanish and may not be widely used in other
Spanish-speaking countries. Learning these expressions
will help you understand and communicate better with
locals in Peru.

In normal social conversations, we expect speech to be relevant to the

topic at hand, and often we expect it to be scripted, following some agreed-
upon “internal script” between interlocutors in specific situations, which
routinizes speech acts in typical social situations. Script theory was described
decades ago by computer scientist by Roger Schank (1984: 125) as follows:

When we read a story, we try to evaluate the reasoning processes of the

main character. We try to determine why he does what he does and what
he will do next. We examine what we would do in a similar situation, and
we try to make the same connections that the main character seems to be
making. We ask ourselves, What is he trying to do? What’s his plan?
Why did he do what he just did? Any understanding system has to be able
to decipher the reasoning processes that actors in stories go
through…[this means] understanding how people formulate goals and
plans to achieve those goals. Sometimes people achieve their goals by
resorting to a script. When a script is unavailable, that is, when the
situation is in some way novel, people are able to make up new plans.

Conversational scripts exist across the social spectrum. Making contact

with a stranger, for instance, requires access to both the appropriate speech
protocol, its contextualized variants, and the specific category of verbal
structures that encode it. The enactment of agreements, disagreements,
flirtations, and so on can be seen to unfold in a script-like fashion, within
conceptualized constrains such as the type of relation the interlocutors have
with each other. Scripts are among the most common tools, in the form of
dialogues, used to impart communicative competence in the FL. The most
common ones can be generated by ChatGPT via prompts such as “How do
you greet someone in the FL formally?” “How do you make small talk in the
FL at a coffee bar”? and so on. These can be followed up by asking ChatGPT
to explain the differences between the NL and the FL, so as to complete the
pedagogical process.
74 Marcel Danesi

Needless to say, there is more to human interaction than script theory and
routinized conversations. But the reliance on systematic ways to speak in
different situations is still a common factor in human dialogical behavior. The
speeches and interactions that typically take place during social rituals and
rites, such as sermons, rallies, political debates, and other ceremonial types of
speech, for example, show that speech involves different kinds of script-like
behavior, either traditionally coded or specifically composed for the occasion.
Speech in ritualistic situations is not intended to create new meanings, but to
assert communal sense-making and to ensure cultural cohesion. Among the
questions that are involved in training chatbots to learn how to carry out speech
acts, the following are of special relevance:

• What are the objectives and effects of different types of speech in

certain FL situations? What keywords characterize them and what
styles are appropriate in the FL?
• What FL cultural rules and conventions are built into the speech
patterns?
• How do these patterns reveal FL cultural values, beliefs, assumptions,
and worldviews? What specific words indicate these?
• How does the speech used in the FL culture relate to social, political,
cultural, and historical processes and systems?

Structure of Conversations

The ability to carry out conversations in the FL is a core part of developing

communicative competence. However, no generally accepted definition of
conversation exists, beyond the fact that it is a communicative act that
involves at least two people talking together. As Thornbury and Slade (2006:
15) characterize it: “Conversation is the kind of speech that happens
informally, symmetrically, and for the purposes of establishing and
maintaining social ties.” If someone asked an interlocutor, “How are you?,”
and the interlocutor answered illogically, “Today is a fine day,” then there is
no meaningful exchange possible, unless the interlocutors know each other
and know as well that irony or humor is sometimes used in their interactions
via such non-sequential responses. The expected answer is a formulaic one:
“Not bad, and you.” Conversations reveal a lot about how we interact with
 Promoting Communicative Competence 75

each other, what our intents are, and even who we are or think we are. So far,
chatbots seem able to differentiate between normal and ironic or conflictual
speech, tending to answer anomalous communication in prompts with
expressions such as “That does not make sense,” and similar responses.
Among the most common types of conversations that chatbots can easily
generate, the following are worth mentioning here:

• Small talk, in which people exchange information, feelings, and

thoughts in a perfunctory fashion in the FL.
• Discussion, in which the conversation in the FL is perceived to be
about something important.
• Chat, which is an informal friendly conversation in the FL.
• Gossip, which is a conversation in which people talk about things they
have heard, which may well not be true.
• Chitchat, which is polite friendly conversation about unimportant
subjects, often for reasons of psychological appeasement.
• Banter, which is a friendly conversation in which people joke or make
fun of each other.

The in-depth study of computer-generated conversation goes back to the

work of Terry Winograd in the late 1960s. One of his computer experiments
involved what he called SHRDLU—a name that reflects the moderately high
frequency of the letters S, H, R, D, L, U in conversational texts (Winograd
1971). SHRDLU was capable of carrying out a seemingly natural conversation
with a user who inputted commands on a keyboard. The machine would
produce observations such as “I don’t understand which pyramid you mean.”
The following stretch of dialogue between a person and SHRDLU has now
become a classic one in the annals of Artificial Intelligence (http://hci.
stanford.edu/winograd/shrdlu/):

Person: Pick up a big red block.

Computer: OK.
Person: Grasp the pyramid.
Computer: I don’t understand which pyramid you mean.
Person: Find a block which is taller than the one you are holding
and put it into the box.
Computer: By “it,” I assume you mean the block which is taller than
the one I am holding. OK.
Person: What does the box contain?
76 Marcel Danesi

Computer: The blue pyramid and the blue block.

Person: What is the pyramid supported by?
Computer: The box.
Person: How many blocks are not in the box?
Computer: Four of them.
Person: Is at least one of them narrower than the one which I told
you to pick up?
Computer: Yes, the red cube.
Person: Is it supported?
Computer: Yes, by the table.
Person: Can the table pick up blocks?
Computer: No.

Clearly, SHRDLU appears to know how to approximate how humans

think when they talk. The dialogue contains, in fact, sequences that are logical
sequiturs and sensitive to the relevance and quality of the content flow. The
computer could also remember the names given to previously-mentioned
objects, and to their arrangements, and it could answer questions about the
blocks and their relevance to the situation at hand. In effect, the SHRDLU
dialogue is an example of what a conversation is essentially about—a set of
statements that cohere into a dialogical unity, containing features such as the
following:

• Turn-taking: when people are involved in a conversation, they do not

talk all at the same time (usually)—one person speaks, and then
another person can follow. A turn becomes evident through certain
cues that interlocutors recognize and count on, called turn
construction units. These allow speakers to predict when someone is
finished so that others can speak, thus minimizing both overlap and
silence gaps. In the SHRDLU conversation, which consists mainly of
a question-and-answer structure, the computer’s turn comes after a
question. This is the simplest form of turn-taking, since it is logically
expected.
• Sequence organization: The point above brings out another intrinsic
feature of conversations—namely, they show sequence organization.
The most common type is seen in question-and-answer sequences,
known as adjacency pairs—sequences that tend to occur in pairs.
However, the pairs may be split over various turns.
 Promoting Communicative Competence 77

• Sequence expansion: This allows a conversation made up of more

than a topic or theme to be constructed and understood as still
relevant, given additional elements introduced within it.
• Repair: This describes how interlocutors deal with emerging
problems in the flow of the conversation. Participants seek to correct
any trouble source by initiating a repair, such as admitting that a
wrong word was chosen.
• Phases: Interactions may involve various phases, including opening
and closing a conversation, assessing something said, complaining
about it, and so on.

Using ChatGPT as a conversation partner will allow students and teachers

to zero in on these aspects of conversation in the FL, even if not explained
explicitly at first. The following example, put forth by Álvarez-Morillo
(2023), is a case-in-point of how a student can use the chatbot to engage in a
meaningful and information-seeking Spanish conversation—an interaction in
which turn-taking, sequence organization, phase structure, and other aspects
are clearly evident in it:

Prompt: Let’s pretend you are a high school student from Uruguay.
I’m going to ask you a series of questions in Spanish, and
I’d like you to respond as if you were human.
ChatGPT: ¡Claro! Estoy listo para responder tus preguntas si fuera un
estudiante de secundaria de Uruguay. Adelante,
ipregúntame lo que quieras!
Prompt: ¿Cómo te llamas?
ChatGPT: ¡Hola! Me llamo Juan. ¿Y tú, cómo te llamas?
Prompt: Me llamo Natalia. Soy estudiante de 4to de secundaria.
¿Tú? ¿En qué grado estás?

As can be seen, the contact protocol between prompter and chatbot

unfolds in a Spanish native-like way. After making “contact” with the
machine, the interaction can become even more theme-specific, projecting the
whole exercise into relevant semantic and cultural domains (from Álvarez-
Morillo 2023):

Prompt: ¿Te puedo hacer unas preguntas?

ChatGPT: ¡Claro, Natalia! Estoy aquí para ayudarte. Adelante, puedes
hacerme las preguntas que desees.
78 Marcel Danesi

Prompt: ¿Cuál es tu materia favorita?

ChatGPT: Mi materia favorita es la historia. Me encanta aprender
sobre eventos pasados y cómo han moldeado el mundo en
el que vivimos hoy. Además, me fascina investigar
diferentes culturas y civilizaciones.

The aspect that sets a chatbot-generated dialogue from static textbook

ones is that it adapts personally to the prompter, using her name, and reacting
directly to her questions—something that could not possibly be achieved in a
pre-written print dialogue.

Conversation Analysis

Perhaps in no other domain of research in pragmatics has the study of

communicative competence borne such fertile fruit as in the study of actual
conversations. Conversation Analysis (CA) has become a broad area of
investigation within various branches of linguistics, anthropology, sociology,
and psychology since at least the 1980s. A major finding of CA is that
conversations are not random; rather, they are governed by social rules, as
discussed and illustrated above. Even a simple protocol such as calling
someone on the phone requires a detailed knowledge of the appropriate words
to start and end to call. Note that the protocol changes if an email or text
message is used in lieu of an oral phone conversation.
In CA, words, phrases, and sentences are studied as units within
conversational sequences, frames, or texts. Personal pronouns, for instance,
are not viewed exclusively as parts of speech, but as anaphoric or cataphoric
trace devices (chapter 3), serving conversational needs, such as maintaining
the smooth flow of conversation by connecting its parts like an electric
network of wires, and eliminating repetitive items. Consider the following two
utterances, which have the same content but deliver it in different ways:

1. Mary is a friend. Mary likes to talk about Mary. But I still like Mary.
2. Mary is a friend. She likes to talk about herself. But I still like her.

The first one is perceived normally as stilted because of its repetition of

the noun Mary; the second one, on the other hand, comes across as more
“conversationally natural,” given its use of pronouns as trace devices. In this
 Promoting Communicative Competence 79

view of language, pronouns are designed to keep the flow of the conversation
smooth and economical. The choice of pronouns is thus not due to a rule of
grammar; but to a rule of conversation. Conversational texts show, therefore,
that we are sensitive to textual rules, more so than strictly grammatical ones—
that is, we anticipate how the forms in a text or utterance relate to each other
and cohere sequentially into a message-making system.
Another major finding of CA concerns what is called the social framing
of speech. Conversations can be framed to be aggressive or subdued,
competitive or cooperative, depending on situation. In the case of competitive
speech, the language used is typically adversarial, whereas in the case of
cooperative speech, the language indicates that the speakers are inclined to
work together to produce shared meanings. In other words, cooperative or
competitive speech acts have identifiable characteristics. The former are
marked by features such as the following:

1. Speakers build upon each other’s comments (“That’s true,” “I

agree”).
2. They use hedges to indicate consent (“Uh-huh,” “Yeah,” “Sure,”
“Right”).
3. When disagreement surfaces, it is negotiated with various formulas
(“Yeah, but, maybe”).
4. Tag questions are used to ensure consent (“You agree, don’t you?”)

This implicit script imparts a sense of togetherness among speakers. As

Robin Lakoff (1975) found in her research, speakers regularly refrain from
saying what they mean in many situations in the service of the higher goal of
politeness or cooperation in its broadest sense; that is, to fulfill the function of
collaboration. Competitive conversations are also marked by specific
strategies, which stand in direct opposition to how they are used in
cooperation:

1. Interlocutors tend to contradict one another’s comments (“That’s

really not true,” “I wouldn’t say that”).
2. They use hedges to indicate dissent (“No-no,” “No way,” “Not true”).
3. Difference of opinion is indicated with expressions that convey doubt
or uncertainty (“Sure, but, maybe”).
4. Tag questions are used as challenges (“You don’t mean that, do
you?”).
80 Marcel Danesi

Clearly, knowing how to frame speech in conversations entails knowledge

of how language, intent, social relations, and situation intersect. The sequences
that occur in conversations follow a kind of implicit script that interlocutors in
such settings agree upon—asking relevant questions, giving support or
expressing opposition, discussing relevant topics in the given situation, and so
on. As we saw with the ChatGPT dialogues above, it is clear that AI is able to
mine from the huge network of texts in its database the appropriate
conversation protocols and scripts that are relevant to the situation at hand in
terms of the appropriate textual structures. With suitable prompts, teachers and
students can in fact get the chatbot to generate correct up-to-date
conversational content. Examples of very useful prompts are given by
Álvarez-Morillo (2023), which can be paraphrased here for convenience,
using Italian as the FL. From these, it is easy to imagine the kinds of
conversational scripts that will be generated by ChatGPT:

Prompt: Let’s pretend you are a tour company in Italy called La

Bella Italian. I will pretend to be a tourist. I will ask you a
series of questions in Italian, and I’d like you to respond as
if you were a human tour organizer.
Prompt: Let’s pretend you are at a restaurant in Pisa called Al Dente.
I will pretend to be someone who wants to dine at the
restaurant. I will ask you a series of questions in Italian.
Prompt: Pretend that you are my new Italian friend, Claudia. I will
be myself. You will speak to me in Italian as if we were
classmates.
Prompt: A minor traffic accident happened in the street named Via
Appia. I’m a local news reporter and want to interview you
about what you saw. I will be asking questions in Italian.
Prompt: Let’s pretend you are my Italian teacher, and you want to
plan a school trip for the class. I will ask you questions
about the trip in Italian.

Speech varies according to the age, occupation, socioeconomic status, and

other variables of the interlocutors, being marked accordingly by so-called
registers—modes of speaking or writing that are designed to match the
situation, the medium used (face-to-face conversation, online interaction, or
writing), the relations between the speakers, and the nature of the topic
involved. Take, for example, saying good-bye to another person in English.
This will vary typically as follows:
 Promoting Communicative Competence 81

Highly Formal: Good-bye!

Mid Formal: Bye!
Informal: See ya’!

The choice of one or the other expressions is not random or optional; it is

a matter of the enactment of politeness criteria present in English society. It
would be considered rude to address, say, a judge during a court proceeding
with an informal mode of speech (unless the judge permits it); and it would be
considered to be aberrant or strange to address a close friend with a highly
formal register. This type of systematic speech variation is found throughout
the world. For example, in the traditional, historical Javanese society of
Indonesia, members of the different classes were expected to utilize a distinct
register of speech. At the top of the social hierarchy were the aristocrats; in
the middle the townsfolk; and at the bottom rural denizens. The formal register
was used by aristocrats who did not know one another very well, but could
also be used by a member of the townsfolk who happened to be addressing a
high government official. The middle register was used by townsfolk who
were not friends, and by rural people when addressing their social superiors.
The informal register was used by the latter among themselves, or by an
aristocrat or town person talking to a rural denizen, and among friends on any
social level. It was also the register used to speak to children of any class.
Now, in modern Javanese society some of these registers have been
recalibrated, as the society has changed. But they have nonetheless left
residues in how people address each other to this day (Harwati 2018).
Relevant prompts for ChatGPT in the domain of register appropriateness
might include the following in the case of, say, Korean as the FL:

Prompt: Let’s pretend you are my Korean teacher, how would I say
hello politely to a young person?
Prompt: Help me carry out an actual contact protocol in Korean with
that person.
Prompt: What kinds of words will help me show politeness in a
commercial store situation?
Prompt: How should I greet someone in authority in Korea?

ChatGPT can handle different registers of language use in response to user

input, adjusting the tone, lexicon, grammar, and style of its responses
accordingly, because it has been trained on a broad range of texts, based on
formal and informal speech samples, and can thus generate responses in
82 Marcel Danesi

different registers. ChatGPT can thus use appropriate greetings and

salutations, and incorporate polite language when necessary. And it can take
into account the context and purpose of the conversation to determine the
appropriate level of formality.

Interactional Verisimilitude

One of the most difficult aspects of traditional classroom and print textbook-
based FLT is its ability to recreate conversational activities in the classroom
or in the language lab that are verisimilar to real-life conversations in the FL.
With AI systems this has become a more achievable goal, as the conversations
generated by ChatGPT in this and previous chapters clearly indicate.
Using chatbots does not completely solve the problem of recreating real
communication in FL situations, but it certainly expands the ability of previous
technologies, such as language labs, and materials such as films and DVDs, to
create more realistic conversational simulations. ChatGPT makes relevant
pragmatic content creation effortless through its ability to respond in a
communicatively natural way to prompts in the FL. It can thus help provide
learners with personalized instruction and practice and teachers with updated
conversational materials and interactions. In sum, the main value of ChatGPT
in the domain of communicative competence teaching is that it is a realistic
FL conversational partner, allowing learners to practice real-life dialogues in
the FL in a continually updated way.

Focused Interactions

The examples used in this (and previous) chapters of human-chatbot

conversations showed the ability of ChatGPT to focus on specific themes,
based on real-life situations. With suitable speech recognition and production
plugins, the FL conversations can also be practiced orally. Moreover, the
chatbot never tires of being asked to repeat or go over something. The
algorithm has what is called a spaced repetition capacity, which allows it to
move words in a conversation from its short-term memory to its long-term
memory through time intervals set by users for purposes of repetition and
review.
 Promoting Communicative Competence 83

Chatbots also allow pictures of different objects as prompts, for which

they can provide the matching words as well as explanations as to their
meaning or cultural relevance in the FL. There are now also virtual reality apps
that can be incorporated into chatbots allowing students to interact in a
simulated FL environment. With focused text, images, sounds, and simulated
virtual reality (VR) functions and components chatbots offer a truly expanded
multidimensional humanoid language lab experience that can be adapted
constantly to specific learning objectives. VR apps are designed specifically
for synchronous language teaching and learning that allows students to interact
in different FL locations (a restaurant, a doctor’s office, the airport, in a
schoolroom, etc.). Users can thus converse with a virtual passenger on an
airplane, order cultural dishes in a restaurant, or check into a hotel. VR
technology is an effective learning tool, providing the immersive experience
that traditional FLT has been incapable of providing in the typical classroom;
it clearly fits in with McLuhan’s classroom without walls concept (chapter 2).
The question that has not been asked so far is whether or not students
themselves favor or, at least, appreciate the use of chatbots in FL-learning. A
study conducted in 2019 by Yin and Satar (2020) presents relevant results.
The participants were Chinese-speaking learners of English, who differed in
their level of linguistic and communicative knowledge (elementary versus
advanced) and who were given the task of interacting with two types of
chatbots: (1) the pedagogical chatbot “Tutor Mike” and (2) the conversational
chatbot Mitsuku.” Tutor Mike would teach formal aspects of the English
language (pronunciation, grammar, vocabulary), while Mitsuku was a chatbot
with which learners could converse in English. The conclusions of the study
are as follows:

• Mitsuku was preferred by participants with advanced knowledge of

English, since it offered an opportunity for conversation that was
verisimilar to person-to-person interaction.
• The interactions with Mitsuku also proved relaxing for students with
poor English skills, who said that they were not afraid of being
laughed at during such interactions (as could occur in a classroom
with an audience of peers), emphasizing that in this case they did not
suffer from performance anxiety.
• Tutor Mike was most helpful for students with lower language levels.
• However, after some time, Mitsuku's speeches were perceived as
predictable, redundant, and unemotional by many students.
84 Marcel Danesi

Despite a few critiques and shortcomings, such as the last one above, a
commonly-perceived benefit by all the students was the immediate feedback
that both chatbots offered, patiently repeating the information that students
asked of them. The ability of both chatbots to learn from the interactions with
users, adapting to different learning styles and utilize the appropriate registers
of communication, was also seen as highly important.
The study also showed that chatbot-based training can be made to focus
on FL culture. To give a practical example, let us assume that a student of
Italian is doing a project on Dante’s Divine Comedy. The learner can ask
ChatGPT to clarify the meaning of some words and various passages in the
text, since they come from a different era (the medieval ages). The learner can
then ask for information about the author of the text as well as information on
his other texts. The student could obviously get the same information through
a search engine; but with ChatGPT, it gets more immediate and contextualized
responses. Moreover, ChatGPT understands what students want to say, even
if they do not say it explicitly, responding consistently to what is implied in a
prompt, even if not stated overtly. In effect, chatbots can answer questions
related to culture to which a teacher may not have immediate access, or may
not even know, making realistic cultural training a more focused goal.
Examples of relevant prompts that a chatbot can easily answer in culture-
specific terms (paraphrased from Álvarez-Morillo 2023):

Prompt: Tell me about traditional celebrations and festivals in

Mexico.
Prompt: What are some popular traditional dishes or cuisine in
Japan?
Prompt: Can you explain the importance of family in Brazil?
Prompt: What are famous landmarks or historical sites in Germany?
Prompt: What customs or etiquette practices are essential in France?

On the whole, the interactional possibilities that in the past could only be
realized with great effort outside the classroom through immersion or
programs abroad in the foreign country are now within reach in the classroom
with chatbots. Of course, virtual worlds cannot replace real worlds, but they
certainly can help students grasp aspects of the FL culture that would have
been difficult to grasp via traditional classroom pedagogy, the use of a print
textbook, and the language lab of conventional FLT. As Alessandro Lenci
(2022) states [translation mine]:
 Promoting Communicative Competence 85

Why is this “new AI” important for the study of language? Because it is
closer to the way humans learn and use language. Cognition like
communication is multimodal, and language learning takes place in a
context that is inherently interactional. In addition, language is not
learned in a supervised manner, using data explicitly labeled by others.
Therefore, research on AI that focuses on multimodality, on learning
models closer to human ones and able to learn from realistic and
contained amounts of data, is certainly something that we linguists can
and, I would add, must look at with interest.

Interactional Pedagogy

In a comprehensive review of AI chatbot systems for EFL (English as a

Foreign Language), Zhai and Wibowo (2022) found that they improved
significantly the ability of students to use English in communicatively accurate
ways. Similarly, Sabzalieva and Valentini (2023: 7) found that AI-assisted
FLT made interactional pedagogy a reality, thus helping teachers enhance
students’ overall communicative competence in the FL. The real-life accuracy
of the dialogues that a chatbot creates is what makes them valuable tools for
gaining access to cultural reality and ongoing shifts in FL culture.
When ChatGPT is integrated with other classroom activities, the relevant
research further indicates that the blend of classroom-based and personalized
AI dialogue practice allows for the effective development of communicative
competence in the FL. As Skjuve and Brandtzaeg (2019) observe:

The increasing usage of chatbots is fundamentally changing the way

people interact with new technology. Instead of clicking buttons to
functionally navigate on a web page, people can access content and
services by the use of natural language in interaction with an artificial
agent (e.g., chatbot). This change toward human–chatbot interaction is
typically manifested through a social and natural conversational style.
This shift of how to interact with data and services has major
repercussions for how to explore and measure conversational user
experience with chatbots.

As I discovered with ChatGPT, the chatbot can be fine-tuned to

understand and respond to specific inquiries in a way that feels natural in
Italian (the FL in this case). This can clearly help to reduce the workload of
86 Marcel Danesi

the teacher by handling a large volume of simple queries by students, as a

virtual native speaker. ChatGPT thus allows teachers to establish an
interactional pedagogical system in the classroom that would have been a
dream in the past. The chatbot can also schedule briefing sessions with
individual students, set reminders as to when tests or other activities are
scheduled, and even produce the course syllabus according to given criteria.
There are, of course, some limitations and caveats that emerge with such
chatbot-oriented pedagogy, some of which will be discussed in more detail in
the final chapter. One of these is the potential for the chatbot to perpetuate or
even amplify biases that exist in the data on which it is trained. This can lead
to the generation or reinforcement of harmful stereotypes that are often
associated with the FL culture and its people. Since chatbots now have the
incredible ability to reprogram (recode) themselves, they can, however, be
asked to eliminate prejudices and biases from their dialogues, even though it
is hard for the chatbots to do so without substantial retraining.

Epilogue

By asking ChatGPT to decompose and explain a simple script-like

conversation into its pragmatic, linguistic, and conceptual components, it will
do so in a generic way. If more pointed and focused questions are involved,
then it can be asked more specific research questions. Some of these are listed
below:

1. What aspects of FL dialogues are different from corresponding NL

ones in terms of register and appropriate level of speech?
2. Which words in a FL dialogue refer to the same objects in the NL one
differentially? How can this be resolved?
3. How can FL dialogues be classified in terms of their social function?
Can you give examples of each one in the FL?
4. How can the FL parts of speech be tagged for social function?
5. How do FL speakers identify relationships among named entities in a
dialogue?

The main point of this chapter has been that conversations that take place
between chatbots, students, and teachers can enhance the acquisition of
communicative competence. As such, the chatbots are highly useful as
pedagogical tools, both for the learner and the teacher, streamlining, at the
 Promoting Communicative Competence 87

very least, the process of FL-learning. They have made the language lab, print
textbooks, and dictionaries virtually obsolete. Conversations with a chatbot
can also be tailored according to personalized needs. For instance, one might
chat about food and recipes, choosing to interact virtually in the FL with a
“chef” bot; one can talk about healthcare by asking the chatbot to take the role
of a FL “doctor;” and so on. The question becomes not when to integrate AI
technologies into FLT but how—a question that will be taken up in more detail
in the remaining two chapters.
Chapter 5

Enhancing Conceptual Competence

Prologue

A common persistent problem manifested by FL learners is the tendency to

put together expressions or messages with FL words and phrases that reflect
NL concepts, not FL concepts. In other words, they are conveying the thoughts
they form in the NL through the FL structures they are learning. For example,
the present author would often hear his beginning students of Italian say things
such as “Io sono caduto in amore” which was their Italian version of the
English concept “I fell in love.” The conceptually-appropriate expression in
Italian would be “Mi sono innamorato.” The interference process in this case
inheres in the transfer of a concept that is articulated in a specific way in the
NL to the FL via a literal word-for-word mental translation; as a result the
phrase “Io sono caduto in amore” is grammatically correct in Italian, but it
reflects English, not Italian, meaning.
The ability to form appropriate FL concepts with the new language is a
type of competence that is different from purely linguistic or communicative
competence, since it involves the ability to think about something directly in
the FL without recourse to the ways in which the student thinks about it in the
NL. It can be called conceptual competence. When the conceptual systems
between the NL and FL overlap, the students’ FL utterance is accurate
conceptually; when they do not, it is anomalous. As argued in previous work,
aberrant utterances such as the one above lack “conceptual fluency” (Danesi
2017). Psychologically, they result from a process that can be called
“conceptual calquing.” This is the tendency to assume that the structures of
the FL bear the same meanings and grammatical features that the concepts
carried by the structures in the NL do.
Conceptual fluency is intertwined with the related notion of metaphorical
competence, put forward for the first time in the mid-1980s (Danesi 1986) as
a pedagogical response to the research findings and insights that were
crystallizing at the time in the fledgling field of cognitive linguistics, and
specifically, in the field of Conceptual Metaphor Theory (Lakoff and Johnson
1980, Lakoff 1987, Johnson 1987). The premise behind the coinage of the
term was as follows: if metaphorical concepts formed the backbone of abstract
90 Marcel Danesi

thinking, as the relevant research was revealing, then metaphor could no

longer be ignored within FLT.
This chapter deals with conceptual competence—an ability that AI now
seems to be have, as can be gleaned by the various ChatGPT responses and
dialogues in previous chapters. As I also found out in the context of an AI
research project in text analysis (Neuman, Danesi, and Vilenchik (2023), the
new Transformers do indeed appear to have the ability to negotiate
metaphorical speech because of extensive training on large datasets in which
metaphor is abundantly present. Given that Conceptual Metaphor Theory has
shown that metaphor use is systematic, not just idiomatic, then it is a small
step to see why AI can tap into the relevant system via inductive learning
(chapter 1). Chatbots can thus engage students in helping them enhance
conceptual competence in the FL, in ways that are consistent with their use in
helping learners develop linguistic and communicative competencies.
Evidence that ChatGPT is now adept at processing metaphorical language
is its use as a tool in literary translation (chapter 1), showing an ability to
preserve the nuances and the verbal images that literary works contain,
especially poetic ones. The task of literary translation has always been a
complex and challenging one, entailing not only a profound understanding of
the source and target languages, but also an appreciation of the cultural context
in which a work is created and the stylistic idiosyncrasies of authors. Given
the chatbot’s training on a vast amount of linguistic-literary data, it develops
a nuanced understanding of the source language, and is thus able to produce
translations that are not only highly accurate but also faithful in style and tone
to the original work. But it should always be kept in mind that ChatGPT is an
AI tool trained to recognize patterns in large quantities of text, and
occasionally will come up with anomalies that are conceptually aberrant.

Conceptual Fluency

Conceptual fluency theory, as mentioned, was motivated by the common

observation that, although classroom-based student discourse often manifests
a high degree of linguistic and even communicative fluency—that is, the
ability to manipulate the formal grammatical and communicative structures of
the FL—it seems to lack the expressive (conceptually-based) fluency that
characterizes the discourse of native speakers. Students tend to use FL
structures as “carriers” of their native language concepts. When the NL and
FL conceptual systems coincide in some area of discourse, then the student
 Enhancing Conceptual Competence 91

utterance is serendipitously native-like; when they do not, it manifests

anomalous conceptual content. What student discourse often lacks, in other
words, is conceptual fluency. From the extensive research in cognitive
linguistics since the late 1970s, it has become obvious that many concepts
reveal metaphoric, metonymic, and ironic forms, and it is now an established
principle that these are hardly subsidiary or decorative forms of speech. They
are at the core of how we conceptualize the world (Danesi 2022). The
pedagogical implications in this area cannot be ignored. Indeed, it can be
claimed that the incorporation of conceptual competence training into an
integrated pedagogical approach to FLT will likely make true proficiency in
the FL a more realizable outcome of the student experience.
The question of how to teach literal concepts in the FL has always had a
fairly straightforward answer, from the Direct Method onwards. It has been
normally assumed that they are best taught directly—namely, by
“demonstrating” them through ancillary devices that allow their meanings to
be illustrated in some concrete way (through pictures, simple dialogues, etc.).
By and large, this type of pedagogy has always produced, and will continue to
produce, good results in all kinds of classroom situations. The dilemma, it
would seem, has always inhered in how to teach non-literal concepts. How
does one teach students to use the concept of “love” appropriately when
discussing it in the FL, given its abstract nature and many non-literal
referential points? As discussed above, it is in such domains of meaning that
NL conceptual transfer can be seen to occur. It is in this rea that ChatGPT
presents itself as a potentially useful tool, because it tends to locate and use
the conceptually-appropriate FL form and then explain how and why it is
different from the NL form. So, when English-speaking students of Italian
produce a phrase such as cadere in amore for “falling in love” (above) they
are using their NL image schema of love as a trap, as it is called in cognitive
linguistics, and transferring it to the conceptualization of Italian amore
(“love”), which is based on a different image schema. This type of “conceptual
interference” is more destructive of meaning flow than other kinds of errors
present in student interlanguages, since it impugns meaning, not only
grammatical or communicative structure.
With advanced versions of ChatGPT, and as I myself discovered by
prompting the chatbot to generate and explain metaphors of love in Italian, it
is potentially a useful enhancer of conceptual competence, as can be
ascertained in the translations it is able to cope up with, as Frąckiewicz (2023)
has noted:
92 Marcel Danesi

The secret behind ChatGPT-4’s success lies in its use of deep learning
techniques, which enable it to analyze vast amounts of text data and learn
patterns and structures that are common across different languages. This
allows the model to develop a sophisticated understanding of the nuances
and subtleties that are unique to each language, making it better equipped
to handle the challenges of literary translation. Moreover, ChatGPT-4’s
ability to generate multiple translations for a given input allows
translators to choose the option that best captures the spirit and tone of
the original text, ensuring that the final result is both accurate and
engaging.

Learning a language goes beyond grammar, vocabulary, and

communicative scripts; it involves understanding the conceptual-cultural
nuances that are imprinted in the FL, many of which are revealed via metaphor
and other figurative forms. ChatGPT is of great significance in this area
because it exposes learners to authentic language patterns and expressions,
enhancing students’ ability to use the FL conceptually in real-world situations.

Cultural Meanings

Proficiency in the FL cannot be pinned down simply to developing high levels

of linguistic and communicative competence, as discussed. The presence of
figurative language in common discourse, not as a decorative communicative
tool, but as a basic form of conceptualization, makes the notion of conceptual
competence a central one in any modeling of FLT.
The gist of the research in Conceptual Metaphor Theory indicates that
metaphor in particular is the vehicle that embeds and then expresses social and
cultural phenomena of all kinds; in other words, the so-called conceptual
metaphors that undergird a language, systematically reflect social emphases,
perceptions, and are often a key to understanding specific social symbols and
rituals. A conceptual metaphor is a thought formula that is imprinted and
delivered via specific linguistic metaphors. For instance, individual
metaphorical expressions such as “She’s a tiger,” “He’s an eagle,” and so on,
are not random creative verbal artifacts; rather they are all based on the same
conceptual metaphor—people are animals, which is used to depict variable
human personalities via this formula. The same concept is also part of cultural
representations, including names given to sports teams (Chicago Bears,
 Enhancing Conceptual Competence 93

Toronto Blue Jays), the use of animal stories to impart moral and ethical
notions to children, and so on.
Take, for instance, the love is a sweet taste conceptual metaphor in
English, which can be seen in such common expressions as “She’s my
sweetheart,” “They went on a honeymoon,” and so on. The instantiations of
this conceptual metaphor do not stop at the level of language. Giving sweets
to a loved one on Valentine’s day, symbolizing matrimonial love at a wedding
ceremony with a cake, sweetening the breath with candy before kissing a
paramour, and so on are all symbolic-ritualistic correlates of the same
conceptual metaphor. Incidentally, in Chagga, a Bantu language of Tanzania,
similar cultural practices exist. It is no coincidence that the language possesses
the same conceptual metaphor. In Chagga the man is perceived to be the eater
and the woman his sweet food, as can be detected in expressions that mean, in
translated form, “Does she taste sweet?” “She tastes sweet as sugar honey”
(Emantian 1995: 168). The research has shown, overall, that material,
symbolic, and ritualistic culture is itself a mirror of metaphorical thinking.
Conceptual metaphors are also a trace to the historical processes that
characterize a society’s fund of unconscious knowledge. A common proverb
like “He has fallen from grace” refers historically to early religious narratives.
Today, we continue to use it with only a dim awareness (if any) of its origins.
Conceptual metaphors that portray life as a journey—“I’m still a long way
from my goal,” “There is no end in sight”—are similarly rooted in such
narratives. As the Canadian literary critic Northrop Frye (1981) pointed out,
one cannot penetrate such expressions without having been exposed, directly
or indirectly, to the original stories. These are the source domains for many of
the conceptual metaphors we use today in English for judging human actions
and offering advice, bestowing upon everyday life an unconscious
metaphysical meaning and value. When we say “An eye for an eye and a tooth
for a tooth” we are invoking imagery that reverberates with religious meaning
in a largely unconscious way. Every culture has similar proverbs, aphorisms,
and sayings. They constitute a remarkable code of ethics and of practical
knowledge that anthropologists call “folk wisdom.” Indeed, the very concept
of wisdom implies the ability to apply proverbial language insightfully to a
situation.
Clearly, the kind of competence that involves thought patterns based on
embedded cultural ideas and traditions goes beyond the rules of grammar and
communication. The intent of the notion of conceptual competence is to stress
how cultural knowledge is expressed through metaphor and other forms of
verbal figuration. In any interaction, conceptual competence provides the
94 Marcel Danesi

means (words, phrases, nonverbal cues, etc.) through which interlocutors can
feel a shared sense of cultural meanings, while communicative competence
provides the means (protocols, scripts, speech acts, etc.) for concretely
realizing these meanings in a specific context. Proficiency can thus be defined
as the ability to control the relevant conceptual and communicative systems of
a language. As Andreou and Galantomos (2009: 2) aptly observe: “Conceptual
competence is neither an isolated nor a simplified phenomenon. Rather, it is
an indispensable component of the wider notion of communicative
competence and a complex set of specific sub-competencies.”
To further explore what conceptual competence entails, even at a
superficial lexical level, consider the following utterances:

1. I apologize for not coming to the party yesterday.

2. I promise I will make it up to you later.
3. I thank you for understanding.
4. I’ll bet you that there will be another occasion.
5. I warn you to beware of their advice.

These utterances illustrate the ways in which distinct kinds of speech

acts—apologizing, promising, thanking, betting, and warning—are expressed
in English. Now, an interlocutor must know not only the actual expressive
forms, but also their different conceptual nuances. Apologizing for not
attending a party is different from apologizing for stepping on someone’s foot.
Similarly, promising that you will make something up later is quite different
from promising that you will never have another glass of beer. Thanking
someone for understanding is different from thanking a server for providing a
cup of coffee that has been requested. Betting that there will be another
occasion is vastly different from betting one’s entire wages on a horse race.
And, finally, warning someone about the danger of someone’s advice is rather
different from warning someone that they are in trouble with the law. In effect,
the pragmatic dimension in utterances subsumes the conceptual-cultural
dimension and, in its implicature for joint action.
The founder of speech act theory, John Austin (1962), observed that, when
used in the first person of the present indicative, verbs like promise, apologize,
thank, bet and warn can be used performatively to realize a speech act.
Languages make available a vast array of performative devices for attaining
joint actions in verbal communication. For instance, in English (1) below is a
more appropriate way of making a hospitable offer than the blunt utterance in
(2):
 Enhancing Conceptual Competence 95

1. How about a beer?

2. Have a beer!

But in other cultures this may not be the case. Wierzbicka (2003: 62)
points out that those who may not be acquainted with the conceptual-semantic-
pragmatic conventions of English speaking-cultures, might well interpret (1)
as a genuine question rather than an invitation, and would certainly consider
(2) a more appropriate type of offer expression. Now, this may seem to be a
problem in lexical choice and it pragmatic consequences, but a closer
consideration will show that each type of expression has a conceptual history
of usage within a culture that entails different modes of understanding the
world (which need not concern us here, see Danesi 2017).
Using the concepts of a different culture through the learning of its
language involves what psychologists call a Theory of Mind, the capacity to
understand other people by ascribing mental states to them and, by extension,
the meaning of their culture-specific expressions. This includes the knowledge
that others’ beliefs and thoughts may be different from one’s own. This comes
out in the fact that the appropriate use of metaphor in a language, which is
largely unconscious, is based on knowing how the other person sees the world.
As Tonini, Bischetti, Del Sette, Tosi, Lecce, and Bambini (2023) have noted,
“metaphor resolution across tasks” implies a Theory of Mind, since it provides
better access to the psychological lexicon (i.e., terms referring to mental states)
and better context processing, serving as a springboard to achieve
sophisticated pragmatic skills.” Now does ChatGPT have a Theory of Mind
(Holterman and Deemter 2023)? Of course is does not, in any human sense.
But the research shows that the chatbot can arrive at correct answers to a
conceptual dilemma more often than would be expected based on chance:
“although correct answers were often arrived at on the basis of false
assumptions or invalid reasoning.” Whatever the case, the process of guessing
based on informed metaphorical knowledge, via inductive training, brings the
usefulness of the chatbot tool closer to the task of training students to develop
conceptual competence through exposure to the system of conceptual
metaphors present in the FL.
On the technical side, Wachowiak and Gromann (2023) have shown how
language models, such as GPT-3, have the ability to “detect metaphoric
language and predict the metaphor’s source domain without any pre-set
domains,” which provides the technical reason why a chatbot can converse in
conceptually-fluent ways in a FL. Significantly, the researchers found that
“GPT-3 generates the correct source domain for a new sample with an
96 Marcel Danesi

accuracy of 65.15% in English and 34.65% in Spanish.” While this may seem
to be a low rate for conceptual competence training, it is actually rather high
when compared to the kinds of errors that students make in this area. However,
the caveat issued above must be reiterated here—AI thinking is not identical
to human thinking and thus will come up with anomalies, of which teachers
and students must be constantly on the lookout for. As Memisevic (2022) aptly
observes:

Neural networks lack the kind of body and grounding that human
concepts rely on. A neural network’s representation of concepts like
“pain,” “embarrassment,” or “joy” will not bear even the slightest
resemblance to our human representations of those concepts. A neural
network’s representation of concepts like “and,” “seven,” or “up” will be
more aligned albeit still vastly different in many ways. Nevertheless, one
crucial aspect of human cognition, which neural networks seem to master
increasingly well, is the ability to uncover deep and hidden connections
between seemingly unrelated concepts and to leverage these in creative
and original ways. As the level of abstraction rises at which we train our
networks, so does the level of capability they surprise and amaze us with.

Calquing

The transfer of NL concepts to FL forms can be called calquing (Danesi 2017).

In the case of a positive transfer, the conceptual calque from the NL onto the
FL produces an isomorphism. So, the English sentence, “They are going to
school” can be calqued onto the Italian corresponding sentence, word-per-
word; “Loro vanno alla scuola.” So, “going to school” in English coincides
conceptually with andare a scuola in Italian and thus the two utterances
coincide in meaning and syntactic structure. Now, the concept of “skipping
school” requires a conceptual adjustment, given the tendency of English
speakers to calque this NL concept onto Italian as saltare la scuola, whereas
the actual concept in the FL is formulated with the expression marinare la
scuola.
There have been attempts to identify these particular types of errors
outside of a conceptual competence framework, as, for example, simple NL
lexical transfers (Laufer and Waldman 2011, Peters 2016). While the end
result is indeed a lexical error, the process may well involve different
conceptualizations, which involve different image schemas (“skipping” versus
 Enhancing Conceptual Competence 97

“marinating”). Conceptual competence is not a synonym for semantic and

lexical knowledge. It involves the ability to navigate figurative referential
source domains that are critical in a conceptual-cultural sense.
An example of how ChatGPT can be used in this area is provided by
Caulfield (2023).

Prompt: Please be my Spanish conversation partner. Only speak to

me in Spanish. Keep your answers short (maximum 50
words). Ask me questions. Let’s start the conversation with
the following topic: la cultura argentina.
ChatGPT: ¡Claro! Estoy listo para hablar contigo en español.
Comencemos hablando sobre la cultura argentina. ¿Cuál es
tu aspecto favorito de la cultura argentina? ¿Te gusta el
tango, el fútbol o la comida?
Prompt: Disfruto de la música de tango, pero no soy muy buen
bailarín.
ChatGPT: ¡Qué genial que te gusta la música de tango! No es
necesario ser un buen bailarín para disfrutarla. ¿Tienes
alguna canción de tango favorita o algún intérprete que te
gusta escuchar?
Prompt: No soy un experto ¿Tienes sugerencias de músicos de tango
populares?
ChatGPT: ¡Claro! Aquí tienes algunas sugerencias de músicos de
tango populares para que puedas explorar su música: Carlos
Gardel, Astor Piazzolla, Aníbal Troilo, Osvaldo Pugliese y
Juan D’Arienzo. Estos son solo algunos, ¡hay muchos más
para descubrir!

The conversation was, clearly, conceptually fluent, with no calques from

negative transfer evident in it. One could explain the use of tener (“to have”)
in tienes algunas sugerencias as part of an idiomatic expression, but even so,
its selection in terms of the pragmatic context is remarkably accurate
conceptually. In fact, throughout the conversation tener became a
metaphorical fulcrum on which the whole dialogue revolved conceptually. So
too were conceptualizations verbalized with disfrutar. It is interesting to note
that the notion of calquing in FL-learning goes as far back as the nineteenth
century. In an 1894 article, linguist Louis Duvau made the following revealing
comment in this regard:
98 Marcel Danesi

Another phenomenon of hybridization is the creation in a language of a

new word, derived or composed with the help of elements already
existing in that language, and which is not distinguished in any way by
the external aspect of the older words, but which, in fact, is only the copy
(calque) of a word existing in the mother tongue of the one who tries out
a new language.

Metaphorical Competence

Conceptual Metaphor Theory has been the target of extensive research since
the publication of George Lakoff and Mark Johnson’s book, Metaphors We
Live By. In it, the two scholars differentiate among metaphor, metonymy, and
irony correctly. However. As Sebeok and Danesi (2000) have suggested these
can be subsumed under a general conceptual competence, whereby meaning
is encoded by different mapping processes: (1) from one domain to another
(metaphor), (2) from an element in the domain onto the whole domain
(metonymy), and (3) from a contrastive domain onto another so as to highlight
it (irony). In this way, abstract concepts can be treated as based on mappings
of different kinds: in metaphor the mapping occurs from a source domain
(animals) to a target domain (people) to produce a conceptual metaphor
(people are animals); in metonymy, a part of the domain (White House) is
mapped onto the entire domain (American government); and in irony, a
contrast (nice) is mapped onto a real world event (hurricane) so as to highlight
its negative impact (“Nice hurricane, isn’t it?”). Together, these three forms
of meaning-making have been called part of metaphorical competence (Danesi
1986), although a better term might be mapping competence.
Metaphorical competence makes up a large chunk of conceptual
competence, and is a major factor in the development of conceptual fluency in
the FL. Take, for example the metaphorical statement “The professor is a
snake,” which is an instantiation of the conceptual metaphor people are
animals. In this case, perceived animal features are mapped onto the concept
of human personality, and this is the reason why specific linguistic metaphors
such as “He is a snake,” “She is a tiger,” My friend is a puppy,” and so on,
will make sense. The meaning of snake that the statement encodes, however,
is not its concrete, denotative one. Rather, it encodes the culture-specific
connotations perceived in snakes; namely “slyness,” “danger,” “slipperiness,”
etc. It is this complex of connotations that is implied in the utterance “The
professor is a snake.” Each different instantiation of the conceptual metaphor
 Enhancing Conceptual Competence 99

changes the view we get of the person: for example, in “The professor is a
gorilla,” the professor is portrayed instead as someone “aggressive,”
“combative,” “rude,” etc.—a complex of connotations that are implicit in the
new mapping of the animal source domain (gorilla). Now, if the learner of
English speaks a NL that does not have this conceptual metaphor, such
expressions will literally make no sense.
Once a conceptual metaphor has been introduced into the system of
concepts that makes up a culture, then it becomes itself a source for providing
further descriptive detail to evaluations of, say, human personality, if such a
need should arise. Thus, for instance, the specific utilization of snake as the
source domain can itself become a subdomain (made up of types of snakes),
allowing one to zero in on specific details of the personality being described:

1. He’s a cobra.
2. She’s a viper.
3. Your friend is a boa constrictor.

Within each source domain, therefore, there are subdomains that provide
the user with an array of connotative nuances that can be utilized to project
subtle detail onto the description of a certain personality. This type of
knowledge is what metaphorical competence is essentially about.

Conceptual Metaphor

Conceptual fluency theory has led to two key research findings that are highly
relevant to FLT with implications for the use of AI for the development of FL
conceptual competence. First, metaphorically-trained groups of students have
no difficulty accessing the FL conceptual system; second, students show a
remarkable ability to apply conceptual metaphors to novel communicative
tasks in native-like ways. The ChatGPT dialogues in this and previous
chapters show a high degree of conceptual fluency, if examined carefully. In
the dialogue above the use of tener as a metaphor of “having” or “possessing”
something as if it were a substance is a key image schema in Spanish. By
simply interacting with ChatGPT the student can thus see how this metaphor
can be used literally in “communicative action.” This implies that input factors
are critical to the unconscious acquisition of concepts, as Krashen and others
have shown outside of AI-based training (chapter 2). In a relevant study worth
mentioning here, Shirazi and Talebinezhad (2013) exposed a group of Iranian
100 Marcel Danesi

learners of English to metaphorical utterances systematically and routinely,

rather than exceptionally (as idioms). At the end of the instruction period they
found statistical significance in the learners’ use of metaphor in a native-like
fashion.
To reiterate, the importance of metaphor to language and thought came to
the forefront in the 1980s, after the publication of Lakoff and Johnson’s
groundbreaking book (above), which documented the fact that abstract
meaning emerges typically via mappings from concrete source domains to
abstract target domains. The relevant research has shown that the human mind
seeks to understand reality by blending domains of meaning through
experiential-imaginative processes. For example, by linking animals to human
personality, we are seeking to understand the latter in terms of the former. This
is why we interpret sentences such as “He’s a fox,” “She’s an eagle,” and so
on, as personality constructs.
It is essential to clarify what the notion of metaphor implies in this
framework. In traditional rhetoric there are many kinds of figures of speech,
and metaphor is considered to be simply one of them. But within Conceptual
Metaphor Theory many of these are seen as manifestations of the same type
of reasoning (mapping) that is exemplified by conceptual metaphors. Thus, for
example, personification (“My cat speaks Italian,” “Mystery resides here,”
etc.) is viewed as having the reverse mapping structure of the usual conceptual
metaphor—namely, mapping human qualities onto the animal target domain.
However, as discussed, metonymy and irony are viewed as separate cognitive
phenomena and thus treated separately (see below). The psychological source
of conceptual metaphors is traced to a mental mechanism called the image
schema, mentioned several times above (Lakoff 1987, Johnson 1987, Lakoff
and Johnson 1999). This is a mental percept that extracts common elements of
an action, object, event, etc., becoming the guide to the metaphorical mapping.
Image schemas such as up-versus-down, back-versus-front, near-versus-far,
etc., are seen as undergirding the formation of concepts such as happiness
(“Lately my spirits are up”), responsibility (“You have to face up to your
problems”), life (“You have a long life ahead of you”), love (“They are very
close to each other”), among many others. An image schema is something that
gives mental outline form to experience. For instance, the common experience
of how containers work and what they allow us to do underlies such concepts
as mind (“My mind is full of good memories”), emotions (“My heart is filled
with hope”), and so on. Obviously, it is impossible to determine which came
first—the metaphor or the schema. Perhaps this is a moot question, since the
occurrence of a metaphor implies a specific image schema and vice versa.
 Enhancing Conceptual Competence 101

Now, the point is that this kind of thinking, as expressed through

conceptual metaphors, is not random, irregular, or something derivative of
more basic literal speech. It is the core of abstract cognition, and is thus
systematic. This implies that metaphorical competence can be taught as can
any other competence in FL-learning environments, even if just via exposure
to metaphorically-rich input. An extensive research study by Norafkan (2013)
investigated the effect of exposure to authentic English language materials on
learners’ metaphorical competence on the part of Iranian students. She divided
53 learners into experimental and control groups. Standard textbook materials
were used with the control group while culture-based, metaphorically-rich
materials and instruction by trained native speakers were employed with the
experimental group. Both groups improved in their English language
proficiency, but the results of the study at the post-test stage showed that the
metaphorically-trained learners showed a significantly higher level of
conceptual competence.
The pedagogical problem comes down to setting up input and devising
instruction that takes conceptual metaphors into account, which is something
that ChatGPT can do instantly, as I myself found out. For instance, the chatbot
can be prompted to do the following:

• select source domains (sweetness, animals, colors, and so on) and

then provide primary utterances in which these are used in Italian;
• select target domains (emotions, personality, perception, and so on)
that are representative of specific types of speech acts or
conversational themes based on the relevant source domains;
• search for common texts with these expressions on the Internet and
then organize them according to image schema categories so as to
make them ready for teaching.

Problems did emerge with ChatGPT’s responses to these prompts, but

they were not intractable, with the chatbot easily revising its responses on the
basis of further prompts. The main problem was that it was difficult for the
chatbot to come up with the themes required for a specific teaching unit. But
then these can be easily devised by a human teacher.
102 Marcel Danesi

Metonymy and Irony

As mentioned, there are two figures of speech that are treated differentially
from metaphor within Conceptual Metaphor Theory—metonymy and irony.
Metonymy reveals a part-whole reasoning process, whereby an element in a
domain is used to stand for the entire domain:

1. She loves Austen (= the writings of Jane Austen).

2. There are many new faces around here (= people).
3. My dad doesn’t like nose rings (= people with nose rings).
4. They bought a FIAT (= car named FIAT).
5. The buses are on strike (= bus drivers).
6. The Church does not condone infidelity (= theologians, priests, etc.).
7. The White House made another announcement today (= the president,
the American government).

Irony constitutes a discourse strategy whereby words are used to convey

a meaning contrary to their literal sense; it is realized by mapping source
domains (love, enjoyment) onto target domains (torture, torment)
incongruously, so as to emphasize meaning by contrast:

1. I love being tortured.

2. They love getting hurt.
3. He enjoys torment.

There is, of course, much more to irony than this assessment. But the
discussion would be beyond the present objective. As I found out with the use
of ChatGPT for Italian pedagogy, AI does indeed detect patterns and
relationships between words and phrases based on irony or sarcasm, but finds
it challenging, which actually mirrors the human decipherment of such
language. That said, ChatGPT is programmed to learn from its interactions
with users, and, as I kept getting it to detect irony, it seemed to improve its
ability to do so.
 Enhancing Conceptual Competence 103

Contrastive Conceptual Analysis

Within the framework of conceptual competence, it is a straightforward

pedagogical task to identify the relevant image schemas and source domains
that are central to its development in the FL. This can be done initially by
getting the chatbot to identify the isomorphic and contrastive source domains
between the NL and the FL. If NL and FL concepts use identical source
domains (in part or in whole), and similar image schematic mappings, then
this would normally entail positive transfer. For example, friendship and love
tend to be delivered in terms of identical (or similar) source domains, based
on specific image schemas, in English and Italian, as corroborated,
incidentally, by the chatbot I used:

depth
English: Theirs is a profound friendship/Theirs is a profound love.
Italian: La loro amicizia è profonda/Il loro amore è profondo.

duration
English: Theirs was a brief friendship/Theirs was a brief love affair.
Italian: La loro amicizia fu breve/La loro storia fu breve.

directionality
English: Their friendship is continuing on/Their love affair is
continuing on.
Italian: La loro amicizia sta continuando/La loro storia sta
continuando.

taste
English: Theirs is a sweet friendship/Their love is sweet.
Italian: La loro amicizia è dolce/Il loro amore è dolce.

Contrasts in conceptualization between the NL and the FL, however, can

lead to negative transfer, as in other areas of FL-learning. Consider the use of
avere (“to have”), fare (“to do, make”), and essere (“to be”) in expressions
that involve the states of “coldness” and “hotness” in Italian (the FL) and
English (the NL):

Fa freddo/caldo fuori. = It is cold/hot outside.

Lui ha freddo/caldo. = He is cold/hot.
104 Marcel Danesi

Il caffè è freddo/caldo. = The coffee is cold/hot.

The different verbal selections, together with the underlying conceptual

patterns that occasioned them are common sources of negative transfer. In
Italian, the verb fare is used with respect to a weather situation—fa freddo
(literally) “it makes cold,” fa caldo (literally) “it makes hot.” The physical
state of “coldness” or “hotness” is conveyed instead by the verb essere when
referring to objects (è freddo “it is cold,” è caldo “it is hot”); and by avere “to
have” when referring to people: lui ha freddo (literally) “he has cold,” lei ha
caldo (literally) “she has hot”. The use of one verb or the other—fare, essere,
or avere—is motivated by an underlying image schematic conceptualization
of bodies, objects, and the environment as containers and the states of
“coldness” and “hotness” as entities. The container and entity schemas are the
sources of the differential verbal selections. If the container is the
environment, then the entities (“coldness” and “hotness”) are made by Nature
(fa freddo, fa caldo); if it is a human being, then the individual’s body has
them (ha freddo, ha caldo); and if it is an object, then the object is their
container (è freddo, è caldo). No such conceptual distinctions are required in
English.
This type of conceptual contrastive analysis can thus be employed to
explain an error of the type Lui è freddo, which can be traced to a negative
conceptual transfer (“He is cold”). Suffice it to say here that ChatGPT
produced conversations about hot and cold states in Italian that involved the
correct image schemas and their relevant verbs. When asked to compare these
with English expressions, it came up with an explanation that was based on
idiomaticity rather than on metaphorical conceptualization, which was
anticipated. It is up to the teacher to continue from where the chatbot left off,
as will be discussed in the next chapter.

Translation Analysis

Perhaps the best way to unravel if ChatGPT is able to negotiate and process
conceptual systems in the FL is to see how it translates texts from one language
to another, which is an indirect form of contrastive conceptual analysis. A
case-in-point is the following Spanish-English translation task asked of
ChatGPT (from https://www.translateplus.com/blog/chatgpt-translating-
content/):
 Enhancing Conceptual Competence 105

Prompt: Translate “Descubre todas las actuaciones musicales,

horarios, estrellas que entregan galardones, favoritos y un
concurso: una guía de la 37ª edición de los Goya” into
English.
ChatGPT: “Discover all the musical performances, schedules, stars
that award prizes, favorites, and a contest: a guide to the
37th edition of the Goya Awards.”

The response was conceptually correct, since ChatGPT translated the verb
entregar as “to award” rather than as “to win” or “to receive.” The text that
was translated into English by ChatGPT was a tweet by the newspaper El País,
which was posted before the event so that the winners were not known at that
point in time, as ChatGPT correctly interpreted. So, the translation “award-
winning stars” would not have been conceptually correct.
In an excellent study of translation with ChatGPT, Claire Frances and
Giovanni Zimotti (2023) looked at how the chatbot handles queries regarding
translation, as for example the following Spanish example, based on an actual
course (Módulo 1):

Prompt: Escribe una reflexión sobre tu experiencia con el español

en tu vida. Para el primer proyecto, que entregarás al final
del Módulo 1, vas a reflexionar sobre tu experiencia con el
español.
ChatGPT: Mi experiencia con el español ha sido muy importante en
mi vida. Desde pequeña, he estado rodeada de la lengua
española ya que mis padres son de América Latina y
siempre hablamos español en casa. Además, tomé clases de
español en la escuela durante toda mi educación primaria y
secundaria.

Aunque he tenido la oportunidad de hablar español con

frecuencia, siempre he sentido que hay mucho más que
puedo aprender y mejorar. He tenido la oportunidad de
viajar a algunos países de habla hispana y siempre me ha
sorprendido la riqueza y la diversidad de la lengua
española. Cada vez que viajo a un nuevo lugar, me encanta
escuchar cómo se habla el español de manera diferente y
aprender nuevas palabras y expresiones.
106 Marcel Danesi

En general, me siento muy afortunada de haber crecido

rodeada del español y estoy emocionada de seguir
aprendiendo y mejorando mi habilidad en esta lengua. Creo
que el español es un idioma muy importante en el mundo
de hoy y estoy agradecida de tener la oportunidad de
practicarlo y mejorarlo constantemente.

In order to assess the conceptual correctness of the translation, the two

researchers asked a Spanish instructor for feedback. The teacher graded it as
perfect, grammatically, lexically, and conceptually. Another instructor stated
that the essay was written with a native-like command of the language.
Clearly, the chatbot passed the translation test with flying colors. Similar
experiments using translation as a means for contrastive analysis have shown
a consistency of results. Of primary importance, pedagogically, is that the
chatbot provides a model to the learner of how to negotiate a type of writing
that is conceptually fluent, which in itself is a remarkable feat.

Metaphor Plugin

Significantly, ChatGPT offers the possibility of installing a so-called

“metaphor plugin” (Halder 2023), a software component that adds
metaphorical abilities to the chatbot’s repertory. As I discovered in the case of
Italian, the plugin was able to generate appropriate conceptual metaphors,
based on relevant image schemas, such as those discussed above in this
chapter.
The plugin also allowed me to customize the generated metaphors
according to specific requirements, such as when I prompted ChatGPT to
speak about love in Italian, which it did successfully. Suffice it to say that for
the teacher of a FL, the chatbot with a metaphor plugin makes conceptual
language teaching much more concrete and systematic, given the ability of the
chatbot to navigate the Internet and come up with the image schemas that are
closely tied to a pedagogical request. Some metaphor-based prompts are
provided by Halder (2023), and are worth repeating here for the sake of
illustration, since they can be easily adapted to FL pedagogy more generally:

Prompt: Generate a metaphor for a difficult journey in Spanish.

Prompt: I need a metaphor to describe a peaceful morning in Italian.
 Enhancing Conceptual Competence 107

Prompt: Can you provide a metaphor for an intense competition in

French?
Prompt: Generate a metaphor for love in Farsi.
Prompt: I need a metaphor to describe a chaotic situation in
Mandarin.
Prompt: Can you provide a metaphor for growing old in Arabic?
Prompt: Generate a metaphor for a strong friendship in Greek.
Prompt: Generate a metaphor for hope in Polish.
Prompt: Can you provide a metaphor for the passage of time in
Hungarian?

Continuous research on chatbots is showing more and more how their

specific type of training contributes to making them achieve a relatively high
level of conceptual fluency. A study by Zheng, Song, Hu, Fu, and Zhou (2020)
showed how work in this area is expanding and leading to significant results;
they describe their research as follows:

Our work first designs a metaphor generation framework, which

generates topic-aware and novel figurative sentences. By embedding the
framework into a chatbot system, we then enable the chatbot to
communicate with users using figurative language. Human annotators
validate the novelty and properness of the generated metaphors. More
importantly, we evaluate the effects of employing metaphors in human-
chatbot conversations. Experiments indicate that our system effectively
arouses user interests in communicating with our chatbot, resulting in
significantly longer human-chatbot conversations.

Epilogue

Because ChatGPT has the ability to access conceptually fluent language on

the Internet, it is a truly useful pedagogical tool for imparting conceptual
competence via specific input prompting. As discussed various times, a
chatbot cannot replace human teachers, for the simple reason that the latter are
ultimately the interpreters of what is conceptually correct in the FL (discussed
further in the next chapter).
The development of conceptual competence in the FL is not something
that comes about only spontaneously; it requires exposure and pedagogical
treatment in the classroom, as any other FL skill also requires. Linguist Ronald
108 Marcel Danesi

Langacker (1987, 1990, 1999) has argued that even the parts of speech
themselves are the result of image schemas working unconsciously below the
surface of grammatical structure. Nouns, for instance, encode the image
schema of a region. Thus, a count noun such as baseball is envisioned as
referring to something that involves a bounded region, and a mass noun such
as rice, a non-bounded region. Now, this difference in image schematic
structure induces grammatical distinctions. So, because bounded referents can
be counted, the form baseball has a corresponding plural form baseballs, but
rice does not. Moreover, baseball can be preceded by an indefinite article (a
baseball), rice cannot. The gist of the research suggests that literal meaning
occurs when words are used in isolation; but when they are combined into
utterances, they tend to accrue metaphorical meanings, and that these
influence grammar and communication.
As Christiansen and Kallens (2022) note, unlike the carefully scripted
dialogues found in textbooks, the language of everyday conversation tends to
be based on variable figurative meanings. As the two scholars state, chatbots
have the capacity to tap into these meanings and, as they remark,
“astonishingly, they all do it without the help of grammar.” However,
grammar does play a role in conceptualization, as Langacker showed. A
chatbot is trained to predict the next word in a sentence given what came
before across hundreds of billions of words from the Internet, print books, and
Wikis (such as Wikipedia). When it makes a wrong prediction, it has the
ability to adjust its parameters with an automatic learning algorithm. So, it
does have its own grammar—one based on network connectivity rather than
any sequencing of words in terms of parts of speech.
We hardly realize how prevalent metaphorical language is in everyday
speech. When asked how we feel about something, we might say that we are
“cool about it,” or that we are “lukewarm about it.” These expressions reveal
that we perceive feeling in terms of the sensations we experience in response
to physical conditions that we have named cool and lukewarm. As
psycholinguist Roger Brown (1958: 154) aptly puts it:

The quality is first detected in one sense modality and is named at that
stage. Afterward the quality is detected in many other phenomena that
register with other senses. The original name tends to be extended to any
experience manifesting the criterial quality.

Now, the chatbot does not have a body, so it cannot register anything via
the senses. But it simulates (artificially) how this might occur via a process
 Enhancing Conceptual Competence 109

called conceptual blending in the human brain (Fauconnier and Turner 2002,
Turner 2015). A blend is formed when the brain identifies distinct entities in
different neural regions as the same entity in another neural region. Together
they constitute the blend. In the metaphor fighting a war on cancer, the two
distinct entities, located in different regions, are “cancer” and “fighting.” The
blending process is guided by the inference that disease is a war, constituting
the final touch to the blend—a touch that keeps the two entities distinct in
different neural regions, while identifying them simultaneously as a single
entity in the third region. Getting the brain to activate blending in the FL is the
ultimate objective of conceptual fluency theory.
Chapter 6

Blended Foreign Language Teaching

Prologue

In the previous chapter, a relevant article by Frances and Zimotti (2023) was
cited in reference to a “translation test” aimed at assessing the level of
conceptual fluency of ChatGPT in Spanish. In the same article, the two
scholars ask ChatGPT a question, which zeroes in on the main question
addressed throughout this book—namely, what is the role of chatbots in FL?
ChatGPT’s answer is remarkably noteworthy, as can be seen below:

Prompt: Write an introduction to the debate surrounding the use of

ChatGPT in the foreign language classroom, citing
academic articles.
ChatGPT: The use of ChatGPT (Generative Pre-trained Transformer)
in foreign language teaching has been a topic of debate in
recent years. Some proponents argue that ChatGPT can
provide language learners with personalized,
conversational practice that is more engaging and
immersive than traditional methods. They claim that
ChatGPT can help learners improve their speaking and
listening skills, as well as their grammar and vocabulary.
However, critics argue that ChatGPT may not be able to
fully replicate the complexity and nuance of human
communication, and that it may not adequately prepare
learners for real-world interactions in the target language.
Additionally, some have raised concerns about the ethical
implications of using ChatGPT in foreign language
teaching, such as the potential for biased language output
or the potential for ChatGPT to replace human language
teachers. (Citations: “Chatbots in Language Learning: A
Review,” “The Ethics of Chatbot Language Teaching,”
“The Role of Chatbots in Language Learning:
Opportunities and Challenges”).
112 Marcel Danesi

As the two researchers note: “The fact that ChatGPT could generate this
text and cite three articles on this topic in less than forty seconds was nothing
short of astounding.” What is similarly amazing is the fact the chatbot did not
repeat robotically what it found on the Internet, as a search that I myself carried
out indicated, but that it put together a technical answer as would a human, in
academic style, as if it were a linguist or language teaching theorist. The two
researchers then followed this up with a more specific question, that is also of
obvious relevance to the present discussion:

Prompt: Can AI trick a foreign language professor into thinking that

a response was written by a student?
ChatGPT: It is possible that AI could produce responses in a foreign
language that are convincing enough to trick a professor
into thinking that they were written by a student. AI is a
broad term that encompasses a wide range of technologies
and techniques, including natural language processing
(NLP), machine learning, and deep learning. These
technologies can be used to generate human-like text based
on a given prompt or input, and it is possible that a
professor could be deceived by the responses if they are not
paying close attention or if the responses are particularly
sophisticated. However, it is worth noting that AI is not
capable of fully replicating the complexity and nuance of
human communication, and it is likely that a professor who
is familiar with the language and the abilities of their
students would be able to recognize the responses as being
generated by a machine.

What is again remarkable is how the chatbot imitated the style of a

professional researcher in language pedagogy, using academic explanations
appropriately. Overall, ChatGPT’s own answers to the two prompts
summarize, in a nutshell, the main points put forward and illustrated
practically in this book—namely that chatbots have come forward to signal
the third major technological revolution in FLT, after the language lab and
CALL, emerging as virtual assistants to both learners and teachers,
constituting a portable “intelligent language lab.” This chapter will tie together
the themes discussed in previous chapters, integrating them into a blended
model of FLT, implying a form of classroom pedagogy that involves humans
in collaboration with an AI. It will specifically assess the role and effects of
this partnership on learners, teachers, and courses.
 Blended Foreign Language Teaching 113

Needless to say, this revolution raises many serious questions, which will
need to be addressed, especially since, as McLuhan often commented about
technology, once a new technology comes into wide use, there is no turning
back the clock. As Rose Luckin (2020) has aptly remarked, echoing
McLuhan’s ideas indirectly, “The real power that AI brings to education is
connecting our learning intelligently to make us smarter in the way we
understand ourselves, the world and how we teach and learn.” As AI installs
itself into education, including FLT, the result will not be easy or
uncontroversial. But, to cite McLuhan (1962: 47) again, change is inevitable:
“When technology extends one of our senses [capacities], a new translation of
culture occurs as swiftly as the new technology is interiorized.”

The Learner

One of the aspects of human learning that is prized above all others is its inbuilt
creativity, a feature that is related to so-called autopoiesis (Maturana and
Varela 1973). For the purposes of the present discussion, this implies that
human thought organizes itself creatively on the basis of the input received
through the body and the brain’s perceptual filters. The result is a mode of
thought that is highly imaginative and referentially purposeful. As Michael
Mumford (2003: 110) has pointed out insightfully, creativity in humans
“involves the production of novel, useful products.” Every act of human
learning is creative in some way, and to some degree, new products of the
mind result from the act of learning.
Is AI creative (autopoietic) in the human sense? As discussed throughout
this book, the new AI technologies seem to have attained a high degree of
creativity, as illustrated with the responses by ChatGPT throughout this book.
Now, it is clear that the chatbot is not aware of what its thoughts are about, as
are humans. This is beside the point in the context of the present book, since
what is of relevance is the fact that the intellectual products of chatbots are
useful in relation to what learners and teachers require. In a blended teaching-
learning system, there will be phases during which the learner receives input
and interacts with a teacher creatively, and others with a machine, just as
creatively. The point is, as Marr (2023) aptly points out, “Tools like ChatGPT
and Dall-E give the appearance of being able to carry out creative tasks—such
as writing a poem or painting a picture—in a way that’s often indistinguishable
from what we can do ourselves.” It does not matter that the AI is sentient or
conscious of its creativity; just that it can produce creative products (as
114 Marcel Danesi

interpreted to be so by humans themselves). As Marr (2023) goes on to

observe, “without humans to create the data that’s been used to train seemingly
highly creative machine intelligence like Dall-E and GPT-4, they wouldn’t be
capable of ‘creating’ anything more impressive than random word-soup or a
toddler’s scribblings.”
The point here is that, for the FL-learner, the blended system is a very
effective one, provided that there is motivation to engage meaningfully with
it. As psychologist-semiotician Yair Neuman (2014: 61) has argued, the use
of Machine Translation, for instance, is powerful cognitively, because it
allows us to truly understand our culture-based peculiarities of meaning-
making:

The reason for using Machine Translation is twofold. First, there is no

better way to understand the loss accompanying translation than by
examining the most structured and formal attempt of translation known
today. Second, instead of pointing at the problems and errors of Machine
Translation, I suggest using it in order to better understand cultural
peculiarities and discrepancies. The second suggestion is somewhat
counterintuitive as we positively think of eliminating errors and solving
problems. Sometimes, however, errors can be used for the better.

The Main Competencies

As Alissa Ludanskaa (2023) has shown with regard to teaching Spanish in

terms of the different competencies (linguistic, communicative, conceptual),
ChatGPT can become a true virtual assistant, allowing individual learners to
develop the competencies at their own pace and according to their specific
learning styles, by providing theme-specific conversational practice, grammar
assistance, vocabulary training, and cultural-conceptual insights. Below are
relevant examples:

Pronunciation Practice
Prompt: ¿Puedes pronunciar la palabra ‘reloj'’en español?
ChatGPT: [Audio Pronunciation]

Conversational Practice (Saying hello)

Prompt: ¡Hola!, ¿cómo estás hoy?
ChatGPT: ¡Hola! Estoy bien, ¿y tú? ¿En qué puedo ayudarte hoy?
 Blended Foreign Language Teaching 115

Grammar Assistance
Prompt: ¿Cuál es la diferencia entre “ser” y “estar”?
ChatGPT: “Ser” y “estar” son dos verbos en español que se usan para
expresar diferentes estados o características.
“Ser” se utiliza para describir características permanentes o
esenciales, mientras que “estar” se utiliza para estados o
condiciones temporales. Por ejemplo, “soy alto” (I am tall)
usa “ser” porque es una característica permanente, mientras
que “estoy cansado” (I am tired) usa “estar” porque es una
condición temporal.
Prompt: ¿Cómo puedo formar una pregunta en pasado?
ChatGPT: Para formar una pregunta en pasado, puedes invertir el
orden del sujeto y el verbo auxiliar. Por ejemplo, “¿Hiciste
la tarea?” o “¿Fuiste al cine ayer?”

Vocabulary Help
Prompt: ¿Cuál es la traducción de “amigo” en inglés?
ChatGPT: La traducción de “amigo” en inglés es “friend.”
Prompt: ¿Qué significa “echar una mano”?
ChatGPT: Echar una mano es una expresión que significa “ayudar” o
“colaborar.” Por ejemplo, “Voy a echarte una mano con la
mudanza” significa “Voy a ayudarte con la mudanza.”

Cultural Insights
Prompt: Cuéntame sobre la celebración de “Día de los Muertos” en
México.
ChatGPT: El “Día de los Muertos” es una festividad mexicana que
honra y recuerda a los seres queridos que han fallecido. Se
celebra el 1 y 2 de noviembre, y las familias construyen
altares en sus hogares con fotos, comida, flores y objetos
que solían pertenecer a los difuntos. Es una ocasión para
recordar y celebrar la vida de los seres queridos que ya no
están con nosotros.
Prompt: ¿Cuál es la forma apropiada de saludar en España?
ChatGPT: En España, es común saludar con dos besos en las mejillas
cuando te encuentras con amigos o conocidos. También
puedes usar un apretón de manos en contextos más
formales.
116 Marcel Danesi

These examples showcase how students can interact with ChatGPT in a

concrete way, in order to seek grammar explanations, expand vocabulary, gain
cultural insights, and get conversational practice on demand. Perhaps the key
aspect of using chatbots in the FL classroom is, in fact, their ability to be
tailored to meet individual learning needs and thus to enhance motivation to
learn (chapter 2). However, it could be claimed that the opposite is true—
namely, that students who rely on chatbots may be actually less motivated than
others. Interestingly, this very issue was posed by Frances and Zimotti (2023)
to ChatGPT itself, which gave the remarkable following assessment, going
contrary to many of the positions taken in this book:

ChatGPT’s responses
1. Decreased motivation: Students who rely on ChatGPT to complete
writing assignments may be less motivated to learn the language and
improve their writing skills. This could lead to a lack of progress and
a decrease in overall language proficiency.
2. Inaccurate translations: ChatGPT may not always produce accurate
translations, particularly when it comes to idiomatic expressions,
slang, or more complex language structures. This could result in
misunderstandings or incorrect language usage in writing
assignments.
3. Reduced cultural competency: Students who rely on ChatGPT for
writing assignments may have a reduced understanding of cultural
context and may be more likely to make cultural mistakes in their
writing. This could have negative consequences for their ability to
communicate effectively with native speakers.
4. Decreased critical thinking skills: ChatGPT does not require students
to engage in critical thinking or analysis, as it simply provides pre-
written responses. This could result in a decrease in critical thinking
skills and a lack of depth in written assignments.

ChatGPT’s responses about being used by students constituted an

unexpected caveat for using a chatbot. Now, as if it were a sentient creature,
aware of the implications of its own criticism, a week after generating this
negative response Frances and Zimotti (2023) prompted ChatGPT again. This
time, the AI provided reasons for using it to the benefit of the learner, which
it stipulated as follows:
 Blended Foreign Language Teaching 117

ChatGPT’s responses
1. Increased efficiency: Chatbots can provide immediate feedback to
students on their writing, potentially allowing for a faster learning
process.
2. Personalization: Chatbots can be customized to provide specific
feedback to individual students based on their needs and learning
goals.
3. Availability: Chatbots can be available to students at any time,
potentially allowing for more flexibility in terms of when students can
receive feedback on their writing.
4. Scaling: Chatbots can potentially be used to provide feedback to a
large number of students at once, making it easier for teachers to
manage a large class.
5. Reduced workload: Chatbots can help to reduce the workload of
teachers by automating the process of providing feedback on writing
assignments.
6. Mixed results: However, it is important to note that the effectiveness
of chatbots in teaching second language writing may vary, as some
students may prefer more traditional methods of instruction and
feedback.

The irony here is that the AI outlined both the advantages and potential
disadvantages of using AI in the classroom. Albeit “artificial,” the responses
by ChatGPT are truly remarkable. Without belaboring the point, the chatbot is
truly a valuable assistant to the learner and the teacher, at the same time that it
is apparently aware of the caveats of relying totally on itself. As such, the
responses above constitute an overall argument in favor of blended FL-
learning, in which humans and machines are interactive entities.

Learner-Chatbot Interactions

As discussed in chapter 2, Krashen’s distinction between acquisition and

learning implies essentially that one or the other process is activated by
relevant input materials and activities (Krashen 1985). Acquisition is triggered
by providing input that stimulates the unconscious aspects of the process;
learning occurs when students work formally with the categories of the FL that
they have acquired. Acquisition is clearly crucial in getting the whole process
into motion. But it can be blocked by fear of making errors or by the inability
118 Marcel Danesi

to interact socially in a classroom environment, which Krashen called the

affective filter (as discussed)—a mental block that prevents students from
acquiring the input to which they are exposed because of anxiety of not
performing adequately in the FL in front of the teacher or classmates. When
the filter is “up,” students may understand what they hear and read, but they
will not be able to acquire it. When the filter is “down,” however, they are not
hampered emotionally by the possibility of failure. Krashen suggests that the
filter will go down on its own when acquisition is set in motion, after which
learning the rules of the language will either occur via what he calls the
student’s inner “Monitor” or else, if necessary, by instruction. He put it as
follows (1985: 1-2):

Acquired competence comes from our subconscious knowledge.

Learning, or conscious knowledge, serves only as an editor, or Monitor.
We appeal to learning to make corrections, to change the output of the
acquired system before we speak or write (or sometimes we speak or
write, as in self-correction).

As discussed throughout this book, one of the main reasons for

incorporating learner-chatbot interactions in the FL classroom is the fact that
these tend to lower the affective filter, facilitating the absorption of new
material much more readily and effectively, and thus enhancing the motivation
to learn. This was corroborated by a key study conducted by Yin, Goh, Yang,
and Xiaobin (2020) with regard to a course in computer science, whose
findings can be seen to apply generally to any educational context. The study
found that “students in the chatbot learning group attained significantly higher
intrinsic motivation than the traditional learning group with perceived choice
and perceived value as core predictors of intrinsic motivation.” In another
relevant study, Holmes, Persson, Chounta, Wasson, and Dimitrova (2022: 6)
summarize this whole situation aptly as follows:

Exploratory learning environments [involve] AI-supported tools in

which learners are encouraged to actively construct their own knowledge
by exploring and manipulating elements of the learning environment.
Typically, these systems use AI to provide feedback to support what
otherwise can be a challenging approach to learning.

Overall, as the research has been showing, and as discussed throughout

this book, in addition to making the acquisition of new knowledge
 Blended Foreign Language Teaching 119

personalized and non-threatening, for the FL-learner the main advantages

resulting from interactions with chatbots include the following:

• Accelerated learning. Language learners need varying time periods

to learn different aspects of the FL. Because chatbots can attend to the
needs of a learner immediately, they can help accelerate learning.
• Tailor-made learning. Unlike traditional approaches using prepared
lessons and print materials, chatbots can easily tailor the teaching
material to the specific level of the learner. Users can select the topics
that matter most to them and work at a comfortable pace.
• Time efficiency. Chatbots allow students to focus on areas in which
they need special help, saving valuable time in the classroom for other
activities.
• Blending. As emphasized here, when used in combination with
human teachers and even with traditional pedagogical methods and
materials, AI can greatly help language learners overall.

The Teacher

In addition to allowing students to learn at their own pace and to provide

teachers with ideas as to how to present new material, AI-aided FLT presents
a number of other benefits to the teacher regarding the usual routines and
preparatory aspects of pedagogy, including grading efficiency, a reduced
workload, the generation of appropriate syllabi, creating lesson plans, and
composing lesson units. Some overly-enthusiastic educators now believe that
AI will not just save teacher time, but at some point make teachers de facto
redundant, or at least reconfigure their role to be classroom technology
facilitators. But this might be the same type of miscalculation that saw the
language lab as not only an aid to learners but a substitute for basic teaching
(chapter 2). While AI technology is certainly much more sophisticated than
past technologies, the human teacher is still crucial to the learner. The reason
is actually common-sensical, as Kavika Roy (2022) has insightfully noted:

Teaching is a one-on-one and many-to-one relationship that works in

many unique ways. Not only does a teacher impart worldly knowledge
to the students, but they also learn from them. Teachers and students form
a symbiotic and synergistic ecosystem that helps in the mutual
enhancement of knowledge (various flavors). Further, what makes
120 Marcel Danesi

humanized teaching so special, is the fact that teaching happens outside

the classrooms as well. It happens throughout life, and that too in various
forms.

While AI tools can do many things automatically, quickly, and accurately,

such as assess learner assignments, they are not capable of the depth of
interpretation or accuracy of analysis that a human teacher can bring to the
tasks at hand. Of course, those who are better at learning by themselves may
not gain as much from the human teacher; but this type of learner is actually
less common than might seem. Overall, in a blended teaching system, both the
human and computer work in tandem to help learners gain proficiency
according to their varying needs and learning styles.

Learner-Centric Pedagogy

As mentioned throughout this book, chatbots diminish the risk of students

developing anxieties; as such, they make learner-centric pedagogy more viable
than at any other time in the history of FLT. However, learner-centric content
still needs to be shaped according to course goals, and this is where the teacher
comes into play—helping establish the learning goals to be pursued
systematically by the students.
Given its ability to help students with self-learning activities according to
need, AI can remove teachers from having to carry out repetitive teaching,
since the AI can take over all kinds of repetitive tasks, allowing more time for
teachers and students to interact meaningfully in the classroom. Rote teaching
can be thus taken away and delegated to technology, as was the case with some
uses of the language lab. With AI, teachers are thus in the process of
transitioning from being general pedagogical practitioners to facilitators. They
do not have to continue to be the sole purveyors of new knowledge, but can
form a partnership with AI and the students to ensure that FL-learning is
maximized.
In effect, learner-centric approaches are now greatly facilitated by AI,
with the teacher overseeing the overall pedagogical process from the outset,
as Holmes, Persson, Chounta, Wasson, and Dimitrova (2020: 36) aptly point
out: “Since Dewey, a learner-centric approach to teaching and learning has
been a recurring theme in education research and practice. This approach gives
children significant control over the learning processes, thereby maximising
learner agency.” However, as the researchers also go on to emphasize, this
 Blended Foreign Language Teaching 121

does not exclude the teacher from the scenario, since, not having the same
capacity as the human teacher to understand issues of bias and fairness, a
chatbot cannot give “genuinely informed consent, or to understand or contest
the effects of AI-based recommendations and predictions on their lives.”

A Teaching Assistant

As Nghi, Phuc, and Tat (2019) found in a significant study, chatbots are not
only “effective and useful for enhancing student performance and engagement
in learning a specific point of a foreign language,” but also have the ability to
generate “excitement and fun.” This combination makes AI an ideal teaching
assistant, which can take over many of the routine and laborious aspects of
pedagogy and tailor them according to learner needs and desires, and making
even those “exciting and fun.”
To reiterate, ChatGPT can serve as a companion to the student and
valuable assistant (grader, supplementary exercise maker, and so on) to the
teacher. It can provide learners with tailored opportunities for practicing
conversations, receiving feedback on grammar and vocabulary usage, and
exploring different semantic nuances involved in conceptually fluent
discourse. In combination with other language learning resources, such as
traditional textbooks and live classes, ChatGPT is the “assistant par
excellence,” who will never complain or become tired, but always trudge on
as requested to do.
One of ChatGPT’s most valuable skills, as also discussed throughout this
book, is giving immediate feedback to students on their speaking or writing,
identifying and correcting grammatical, lexical, and conceptual errors, as well
as suggesting alternative phrasing. As Alejandro del Carpio (2023) has aptly
pointed out, there are several main ways in which ChatGPT can be used as the
maximal assistant, which are worth paraphrasing here for the sake of
illustration:

• Language practice. ChatGPT can help students practice grammar and

vocabulary in context. It can provide examples of how to use words
and grammatical structures in real-life situations, updating these
situations constantly because of Internet updates.
• Conversation practice. ChatGPT can produce thematic conversations
based on real-life situations (shopping, eating at a restaurant, etc.),
122 Marcel Danesi

tailored to meet student needs and in response to teacher prompts for

relevant conversational material at a specific point in the curricular
sequence.
• Proofreading and editing. ChatGPT can proofread and edit FL
writing expertly. It can catch common grammar mistakes and, if
asked, break the writing down into its stylistic and conceptual
components.
• Textual materials. ChatGPT can search out textual realia on the
Internet according to teacher specifications, locating and extracting
FL texts that are appropriate at different levels of competency, from
elementary to advanced.
• Pedagogical materials. ChatGPT can help teachers create all kinds of
teaching materials, from drills to exams, as well as grading them and
providing feedback to students.
• Emotional intelligence. While ChatGPT can simulate real-life
conversations, it lacks the emotional intelligence of a human
conversation partner. This means that a human tutor or language
exchange partner is always required to complete any pedagogical
process.

The Course

As mentioned, a chatbot can be prompted to generate an entire coursebook,

based on specific curricular requirements and according to a given syllabus.
Chatbots can also create an informal, collegial environment for sharing
concerns and successes in a course, which can be rather helpful when asking
for feedback about more delicate topics, such as the biases that a FL culture
might present. In other words, chatbots are useful not only to learners and
teachers, but to course layout and design. As Sabzalieva and Valentini, (2023:
14) observe, this entails assessing chatbot use according to specific
pedagogical questions or requirements, such as the following:

• Which aspects of course structure would benefit from using AI

(student learning, assessment, planning, etc.)?
• Which AI technology should be chosen? As discussed throughout this
book ChatGPT is a very useful system, but there are others that have
different devices and plugins that may be required for a specific
 Blended Foreign Language Teaching 123

course. Some of these are described in the Appendix at the end of this
book.
• What would be the criteria for choosing a specific chatbot? What
value does the technology add?
• How effective is the AI technology in meeting the needs that an
instructor or course designer has identified?

The traditional course, consisting of a main textbook (print or electronic),

a set of practice materials (print or audio-oral as in a language lab), and a
teacher at the front of the class, directing and instructing the entire FL-learning
process at the same time may no longer be a viable model. The purpose of this
book has not been to present an exhaustive manual of AI-based techniques,
nor to be innovative in showing how chatbots and digital technologies can be
incorporated into FLT. Its goal has been simply to highlight the adaptability
and versatility of integrating chatbots into FLT and how they fit in with FL-
learning models, in a world where technology has become a key component
in people’s lives. Blended pedagogy in FLT can be used to kill the proverbial
two birds with one stone—enhancing learning in the FL and motivating
students to envision the FL classroom as a truly contemporary one and useful
to them beyond just language learning.
As Henry Widdowson (1978: 75) put it decades ago “effective teaching
materials and classroom procedures depend on principles deriving from an
understanding of what language is and how it is used.” These principles are
found in a small-scale version in the case of an integrated interaction between
learner, teacher, and computer. Nonetheless, no matter how scientific or
theoretically sound a particular teaching proposal might appear to be, it is
always susceptible to the vagaries of its human congener. Proposals such as
the blended teaching one here is no different. It is not a solution to the dilemma
of how languages are best learned in formal environments; it constitutes a
suggestion for enhancing the learning process in the modern world today,
where technology plays such a crucial role in everything we do.

Blended Pedagogy

As Suchi Rudra (2021) has aptly noted, chatbots should be enlisted to provide
tutoring on demand, not to replace human teachers. This is a key function,
since it is becoming less and less possible to support every student having
124 Marcel Danesi

some learning need with a human tutor. As a result, AI-driven tutor chatbots
are becoming an important addition to the classroom.
One of the most fascinating ideas to come out of cognitive science is, as
mentioned throughout this and the previous chapter, that of blending. The
main manifestation of this neural process is figurative language, as discussed
in chapter 5, and the claim can be made that blending is constantly at work in
FL-learning. The claim here is that blending is not only applies to learning
itself (Fauconnier and Turner 2002), but is also a broader one that involves the
formation of social structures (including education) through the amalgamation
of humans and their technologies, as McLuhan often pointed out, into unitary
social systems. Specifically, the claim here is that the pedagogical model that
best mirrors neural blending it itself a blended one. However, as Raymond
Gibbs (2000) has argued, one cannot consider blending theory as a single
theory, but rather as a framework. And that is how it is intended in this book.
Unlike traditional views of FL-learning (chapter 2), the blending
framework does not envision the process in a linear way, orchestrated through
a single teacher and a textbook, but concentrically through activities and
processes that blend student needs and teacher requisites as a course unfolds.
As a psychological notion, blending refers to how language concepts are
formed in the brain from linkages among various inputs (chapter 5). As a
pedagogical notion, it refers to how different intelligences (human and
artificial) along with human needs are linked together into a framework that
goes back and forth between humans and machines. This entails a kind of
continual navigation among the different contents, contexts, media, and
technologies to carry out the same or different tasks the individual teacher and
group of students in a classroom once did in a prescribed sequence. This means
that no one individual can control the learning process entirely; the process
organizes itself from the interactions between its human and artificial
components.

A Paradigm Shift

Technologies invariably bring about paradigm shifts in how people think,

behave, and learn. The advent of AI as a learning tool is a major part in the
paradigm shift in education today. It has changed how the roles the roles and
expectations of learners and teachers are perceived. It has also made possible
the type of “wall-less” classroom that McLuhan had predicted in the 1960s
(chapter 1). The blended model of FLT described here is a product of the
 Blended Foreign Language Teaching 125

paradigm shift, exemplifying the following three main features: (1) it provides
conditions for FL-learning through an apprenticeship format, given that
students can see themselves as apprentices through chatbot interactions; (2) it
promotes virtual connectivity to the ever-changing world of the FL culture,
made possible by the chatbot’s ability to navigate the Internet instantly and in
a thematically-focused way; and (3) it creates conditions whereby learning is
felt to be concentric (learner-centric and interactive) rather than as linear.
So, what do we language teachers do practically in this new age of
Artificial Intelligence? What kinds of curricula or methodologies are suitable?
The previous chapters have attempted to discuss the bits and pieces of a
potential set of answers to these questions. The main advantages of AI-based
pedagogy can now be summarized in point form here:

• The new technologies have not eliminated the need to use traditional
materials and the need for students to gain print literacy in the FL.
But the exclusive or primary use of a traditional textbook and written
activities is no longer the hub of pedagogy because students (and
many teachers) have not been reared in the previous age of print-
based pedagogy, but rather in an electronic-digital age.
• AI technologies make a new language interesting and connected to
the social-technological milieu in which we all live. They extend the
classroom considerably, connecting the FL to the real world in which
we live.
• Through the ability of ChatGPT to navigate large masses of
information on the Internet, the contents of a course can be extended
to the outside world. This does not obliterate the traditional materials;
it connects them to the real world and its ongoing changes.
• AI extends the classroom considerably beyond its temporal and
spatial limitations. It allows students and teachers to form a blended
learning community with the technology itself.
• Chatbots have become a kind of “teacher-by-proxy” system, able to
reinforce classroom teaching considerably. They are also teaching
assistants to the human instructor.
• In the past, the classroom isolated itself from the outside world—a
situation that McLuhan called the world of the “walled-in” classroom,
as mentioned in chapter 1.
• Above all else, chatbots can bring this world of FL culture continually
into the classroom.
126 Marcel Danesi

In 1958, McLuhan was invited to be the keynote speaker at the annual

convention of the National Association of Educational Broadcasters in
Omaha, Nebraska. It is reported that he gave a dazzling speech, and afterwards
the Association invited him to prepare a syllabus for a year-long grade eleven
course devoted entirely to the study of new media, which at the time meant
basically television. They wanted American students to be conscious of the
contradictory effects of media, as expressed by his own laws of media,
believing fully that McLuhan was the appropriate person to explain them.
McLuhan gladly accepted the task and set out immediately to work on it. But
the syllabus and accompanying textbook that he devised shortly thereafter
baffled the executive of the Association. They did see his materials as
constituting an ambitious and intellectually challenging project, but they had
no idea how they could be used in the classroom of that era. One exercise that
McLuhan included was reflective of how he envisioned the whole project:
“Speech as organized stutter is based on time. What does speech do to space?”
It is now part of McLuhanian lore that McLuhan saw the rejection as inevitable
because the times were not right. So, he went on to expand and modify his
report into his classic 1964 book, Understanding Media: The Extensions of
Man. But the public was also not ready for his ideas. Reviews of the book were
generally negative. But the reviewers did notice the potential importance of
the work and some were even enthusiastic. McLuhan had understood the
power of new technologies perfectly, as bringers of paradigm shifts.
The Reform Movement envisioned its pedagogical models as based on
standardized curricula and materials that purportedly reflected a common
denominator in learning, regardless of learner background. Education for
“one-and-all” was an Industrial Age concept and the classroom was turned
into a kind of factory where students “grew up along with the production line”
(McLuhan and Leonard 1967: 23). Mass education came about at the historical
moment when western society had mastered mass production technology, and
education adapted this model. The first FLT methods were designed with a
one-size-fits-all philosophy. Teaching called for very little student
involvement. The teacher was the center of attraction. But the teacher was also
controlled from above through imposed standardized models of teaching and
testing. Evaluation through uniform testing was the main tool for determining
learning outcomes. This model has broken down with the advent of AI. The
blended model proposed here is meant to correspond to how the world has
evolved since the Reform Movement and how much technology shapes
everyday life today.
 Blended Foreign Language Teaching 127

Epilogue

In their report to the Council of Europe regarding the impact of AI on

education, Holmes, Persson, Chounta, Wasson, and Dimitrova (2022: 11)
make the following relevant statement, paraphrasing the concerns of the
Council:

AI, like any other tool, offers many opportunities but also carries with it
many threats, which make it necessary to take human rights principles
into account in the early design of its application. Educators must be
aware of the strengths and weaknesses of AI in learning, so as to be
empowered—not overpowered—by technology in their digital
citizenship education practices. AI, via machine learning and deep
learning, can enrich education. By the same token, developments in the
AI field can deeply impact interactions between educators and learners
and among citizens at large, which may undermine the very core of
education, that is, the fostering of free will and independent and critical
thinking via learning opportunities.

The core argument that the report goes on to make is that an AI system
cannot replace the human teacher entirely. Built upon a neural network
architecture and trained with massive amounts of data, it operates through
complex algorithms that guide its responses. Its focus lies more on
performance and accuracy than on explaining the rationale behind its
decisions, which is always needed in classrooms.
There is little doubt that a paradigm shift has occurred in FLT and
education generally with AI technologies, and fundamental questions
regarding it are crucial, perhaps reflecting the same kind of questions that
emerged during the Reform Movement in the late nineteenth century. As
teachers, we need to truly recalibrate our understanding about who we are and
rethink the role of our ever-broadening partnership with technology in the
classroom. Where does the human teacher end and the artificial teacher begin?
What is the role of the physical classroom today? Does it need to exist as it
has for centuries and even millennia? Is McLuhan’s wall-less classroom
already here? If a student spends time asking ChatGPT questions, and
engaging with it on all matters involved in learning the FL, can we say that
ChatGPT is fulfilling the role of a teacher?
There really is no way to turn the back the clock, nor is it even useful to
do so, to paraphrase McLuhan one more time. A tool such as ChatGPT in FLT
128 Marcel Danesi

is not unproblematic. It challenges us to consider what we do as teachers and

what FL learning is fundamentally all about—psychologically, socially, and
emotionally. Like the faith put into the language lab in the past, however, AI
is not the magic pill that teachers have been looking for, nor is it the dangerous
thing that will eliminate them from the classroom. The work required to adapt
is never easy, but the potential is enormous if we choose to embrace the
change, as discussed throughout this book. As writer J. G. Ballard so aptly put
it decades ago (1984: vii): “Science and technology multiply around us. To an
increasing extent they dictate the languages in which we speak and think.
Either we use those languages, or we remain mute.”
Appendix
Language Learning Apps

For the purposes of this book, I have used or referred to uses of ChatGPT,
which is, in my view, a truly powerful tool in the blended model of FLT
proposed in this book. It enables users to refine and steer a conversation or a
response according to specific needs and requirements. Successive prompts
and replies form part of a conversational flow that imitates almost to perfection
how a human verbal interaction might unfold. OpenAI operates on a freemium
model, allowing users to access the GPT-3.5-based version without fees. In
contrast, its more advanced GPT-4 based version involves payment.
At the time of writing this book, an array of various useful language
learning apps were available for use in the FL classroom, some of which can
be accessed for a fee, and most of which have a chatbot option. Of these, the
following seem to me to be useful either as part of blended teaching or as
ancillary outside-the-course devices. Brief descriptions of unique features
follow below. Note that these are selections; there are others that teachers may
already be using.

Babbel

Babbel is a language-learning platform that uses AI-powered chatbots to

provide personalized language-learning recommendations to students.
Lessons include polysemic variants of words or phrases, illustrated with
pictures and video, as well as information on social register. The new material
is used in common conversations, to which users can listen and then repeat, as
well as interact with.

DALL-E

DALL-E is developed by OpenAI, constituting a version of GPT-3 modified

to generate realistic images and with the ability to combine concepts, formal
structures, and styles into learning units. Its distinguishing feature is its ability
to combine verbal and visual material. OpenAI makes a combination of
130 Marcel Danesi

ChatGPT and DALL-E available as a unitary system at the time of writing this
book.

Duolingo

This app has a so-called “streak” feature, intended to motivate users to keep
going by tracking the number of days used to reach a learning goal. It also
makes available short audio stories that allow users to check their
comprehension skills. It is meant for beginners and thus its range of grammar,
vocabulary, and conversation is constrained to this level. Duolingo also creates
characters such as a chef, a police agent, or a taxi driver, which will interact
with users, responding to their prompts in a contextually-appropriate way.
When a user gives a wrong answer, the characters will correct the prompter so
as to make the learning process more personalized.

Rosetta Stone

This app uses primarily auditory material with images, which can be
customized according to learning preferences. It also incorporates augmented
(virtual) reality which enables users to point a phone camera at an object and
get a word for it in the FL.

Memrise

This is an app-chatbot that uses short videos to show how different phrases are
used in real-life conversation in the FL. It allows for abundant pronunciation
practice as well as conversational know-how. Above all else, it makes learners
feel that they are interacting with a real teacher.

Busuu

This app helps users determine how advanced they are, allowing them to set a
daily study goal, according to a study plan it generates that presumably allows
them to reach their learning goal by a set date. The app also provides
 Appendix 131

reminders, such as the vocabulary or grammar that a user needs to review so

as to improve in these areas. Each new word or phrase is paired with an image
of its meaning.

Drops

With this app, users can check out their personal learning statistics after
completing a task (percentage of correct versus wrong answers). As such,
Drops offers specialized training in FL words used in isolation as well as a
system of review exercises to reinforce vocabulary.

Mondly

The app uses images, translations, and auditory aids to help with specific
learning requests. It also speaks the FL words and phrases in a melodic way,
perhaps based on the view that musical style enhances memory. The app also
has augmented and virtual reality features, allowing users to participate by
simulation in informal conversations. It also has chatbots with whom a user
can interact by either typing or speaking the prompts.
References

Adams, Douglas (2002). The Salmon of Doubt: Hitchhiking the Galaxy One Last Time.
New York: Del Rey Books.
Alamer, Abdullah and Almulhim, Fahad (2021). The Interrelation Between Language
Anxiety and Self-Determined Motivation; A Mixed Methods Approach. Frontiers in
Education 6: https://doi.org/10.3389/feduc.2021.618655.
Algaraady, Jeehaan and Mahyoob, Mohammad (2023). ChatGPT’s Capabilities in Spotting
and Analyzing Writing Errors Experienced by EFL Learners. Arab World English
Journal 9: 3-17.
Álvarez-Morillo, Natalia (2023). 30 ChatGPT Prompts For Language Learners. How You
Can Use AI in the World Language Classroom. Carnegie Learning,
https://www.carnegielearning.com/blog/30-chatgpt-prompts-language-learning/.
Anderson, Myrdene (2013). How Qualification and Quantification Meet, Or Don’t, in
Ethnograph. In: Mariana Bockarova, Marcel Danesi, and Rafael Núñez (eds.),
Semiotic and Cognitive Science Essays on the Nature of Mathematics, 296-329.
Munich: Lincom Europa.
Andreou, Georgia and Galantomos, Ioannis (2008). Designing a Conceptual Syllabus for
Teaching Metaphors and Idioms in a Foreign Language Context. Porta Linguarum 9:
69-77.
Austin, John L. (1962). How to Do Things with Words. Cambridge, MA: Harvard
University Press.
Ballard, J. G. (1984). Crash. London: Jonathan Cape.
Bar-Hillel, Yehoshua (1960). The Present Status of Automatic Translation of Languages.
Advances in Computers 1: 91-163.
Bibauw, Serge, Van den Noortgate, Wilm, François, Thomas, and Desmet, Piet (2022).
Dialogue Systems for language Learning: A Meta-analysis. Language Learning &
Technology 26(1): 1-24.
Black, Max (1962). Models and Metaphors. Ithaca: Cornell University Press.
Brunfaut, Tineke and Révész, Andrea (2015). The Role of Task and Listener Characteristics
in Second Language Listening. TESOL Quarterly 49 (1): 141-168.
Campbell, Richard (2023). Language Learning in the Age of AI: Challenges and
Opportunities. Medium: https://medium.com/@iPadEFLTeacher/language-learning-
in-the-age-of-ai-challenges-and-opportunities-a5a8353bedd4.
Carpio, Alejandro del (2023). 5 Ways to Learn Spanish with ChatGPT. Digital Polyglot,
https://www.digitalpolyglot.com/5-ways-to-learn-spanish-with-chatgpt/.
134 Marcel Danesi

Castagnaro, Peter J. (2006). Audiolingual Method and Behaviorism: From

Misunderstanding to Myth. Applied Linguistics 27 (3): 519-526.
Caulfield, Jack (2023). 9 Ways to Use ChatGPT for Language Learning. Scribbr,
https://www.scribbr.com/ai-tools/chatgpt-language-learning/.
Chastain, Kenneth (1971). The Development of Modern Language Skills: Theory to
Practice. Philadelphia: Center for Curriculum Development.
Chomsky, Noam (1957). Syntactic Structures. The Hague: Mouton.
Chomsky, Noam (1965). Aspects of the Theory of Syntax. Cambridge, Mass.: MIT Press.
Chomsky, Noam (1986). Knowledge of Language: Its Nature, Origin, and Use. New York:
Praeger.
Christiansen, Morten H. and Kallens, Pablo Contreras (2022). AI Is Changing Scientists’
Understanding of Language Learning—and Raising Questions about an Innate
Grammar. The Conversation, https://theconversation.com/ai-is-changing-scientists-
understanding-of-language-learning-and-raising-questions-about-an-innate-
grammar-190594.
Danesi, Marcel (1986). The Role of Metaphor in Second Language Pedagogy. Rassegna
Italiana di Linguistica Applicata [Italian Review of Applied Linguistics] 18: 1-10.
Danesi, Marcel (2017). Conceptual Fluency Theory and the Teaching of Foreign
Languages. Hauppauge: Nova Science.
Danesi, Marcel (2021). Linguistic Relativity Today: Language, Mind, Society, and the
Foundations of Linguistic Anthropology. London: Routledge.
Danesi, Marcel (2022). Metaphor, Riddles, and the Origin of Language: The Sphinx’s
Legacy. Lanham: Lexington Books.
Danesi, Marcel (2023). Puzzles and Games in Second Language Teaching. Padova:
Libreria Universitaria.
Dokukina, Irina e Gumanova, Julia (2020). The Rise of Chatbots—New Personal Assistants
in Foreign Language Learning. Procedia Computer Science 169: 542-546.
Dulay, Heidi, Burt, Marina K., and Krashen, Stephen (1982). Language Two. Oxford:
Oxford University Press.
Duvau, Louis (1894). Expressions hybrides. Mémoires de la Société de Linguistique de
Paris 8 [Hybrid expressions. Memoirs of the Linguistic Society of Paris 8]: 191.
Emantian, Michele (1995). Metaphor and the Expression of Emotion: The Value of Cross-
Cultural Perspectives. Metaphor and Symbolic Activity 10: 163-182.
Fang, Tao, Yang, Shu, Lan, Kaixin, Wong, Derek F., Hu, Jinpeng, Chao, Lidia S., and
Zhang, Yue (2023). Is ChatGPT a Highly Fluent Grammatical Error Correction
System? A Comprehensive Evaluation. arXiv, https://arxiv.org/pdf/2304.01746.pdf.
Fauconnier, Gilles and Turner, Mark (2002). The Way We Think: Conceptual Blending and
the Mind’s Hidden Complexities. New York: Basic.
Frąckiewicz, Marcin (2023). ChatGPT-4 and the Future of Literary Translation. TS2 Space,
ts2.space/en/chatgpt-4-and-the-future-of-literary-translation-preserving-nuance-and-
creativity.
Frances, Claire and Zimotti, Giovanni (2023). Robots vs. Humans: Does ChatGPT Pose a
Challenge to Second Language Writing? Free Language Teaching technology
Magazine, https://fltmag.com/chatgpt-second-language-writing/.
 References 135

Fries, Charles F. (1927). The Teaching of the English Language. New York: Appleton
Century.
Fries, Charles F. (1945). Teaching and Learning English as a Foreign Language. Ann
Arbor: University of Michigan Press.
Frye, Northrop (1981). The Great Code: The Bible and Literature. Toronto: Academic
Press.
Fryer, Luke K., Coniam, David, Carpenter, Rollo, and Lăpușneanu, Diana (2020). Bots for
Language Learning Now: Current and Future Directions. Language Learning &
Technology 24(2): 8-22.
Gardner, Robert C. (1985). Social Psychology and Language Learning: The Role of
Attitudes and Motivation. London: Edward Arnold.
Gibbs, Raymond W. (2000). Making Good Psychology out of Blending Theory. Cognitive
Linguistics 11: 347-358.
Gkountara, Despoina and Prasad, Ramjee (2022). A Review of Artificial Intelligence in
Foreign Language Learning. 25th International Symposium on Wireless Personal
Multimedia Communications (WPMC), 134-139. Herning, Denmark, 2022, doi:
10.1109/wpmc55625.2022.10014767.
Gobbi, Andrea (2012). Ipotesi glottodidattica 2.0. Journal of e-Learning and Knowledge
Society 8: 47-56.
Godwin-Jones, Robert (2022). Chatbots in Language Learning: AI Systems on the Rise. In:
Birna Arnbjörnsdóttir, Branislav Bédi, Linda Bradley, Kolbrún Friðriksdóttir,
Holmfridur Garðarsdóttir, Silvie Thouësny, and Matthew J. Whelpton (eds.),
Intelligent CALL, Granular Systems, and Learner Data, 124-128. Research-
publishing.net.
Good, Irving J. (1965). Speculations Concerning the First Ultraintelligent Machine.
Advances in Computers 6: 31-83.
Gouin, François (1880). L’art d’enseigner et d’étudier les langues. Paris: Librairie
Fischbacher [The art of teaching and studying languages. Paris: Fischbacher
Bookstore].
Hackl, Bernd (2016). Body Language in the Classroom. Scientist Live,
https://www.scientistlive.com/content/body-language-classroom.
Halder, Nilimesh (2023). Unleashing Creativity with the Metaphor ChatGPT Plugin: A
Comprehensive Guide for Writers. Medium,
medium.com/@HalderNilimesh/unleashing-creativity-with-the-metaphor-chatgpt-
plugin-a-comprehensive-guide-for-writers-29b2d3643637.
Hartsell, Tarallynn and Yuen, Steve (2006). Video Streaming in Online Learning. AACE
Journal 14: 31-43.
Harwati, Luisa Neti (2018). Javanese Language Preservation in the Global Era:
Determining Effective Teaching Methods for Elementary School Students. Advances
in Language and Literary Studies 9: 37-42.
Holmes, Wayne, Persson, Jen, Chounta, Irene-Angelica, Wasson, Barbara, and Dimitrova,
Vania (2022). Artificial Intelligence and Education. Strasburg: Council of Europe.
Holterman, Bart and Deemter, Kees van (2023). Does ChatGPT Have a Theory of Mind?
arKiv 2305: https://arxiv.org/abs/2305.14020.
136 Marcel Danesi

Hong, Yee Chee and Ganapathy, Malini (2017). To Investigate ESL Students’ Instrumental
and Integrative Motivation towards English Language Learning in a Chinese School
in Penang: Case Study. English Language Teaching 10: 17-35.
Hymes, Dell (1966). Two Types of Linguistic Relativity. In: William Bright (ed.),
Sociolinguistics, 114-158. The Hague: Mouton.
Hymes, Dell (1971). On Communicative Competence. Philadelphia: University of
Pennsylvania Press.
Jakobovits, Leon A. (1970). Foreign Language Learning: A Psycholinguistic Analysis of
the Issues. Rowley, Mass.: Newbury House.
Johnson, Mark (1987). The Body in the Mind: The Bodily Basis of Meaning, Imagination,
and Reason. Chicago: University of Chicago Press.
Khan, Shahla Sattar and Takkac, Mehmet (2021). Motivational Factors for Learning
English as a Second Language Acquisition in Canada. Higher Education Studies 11:
160-170.
Kitao, Kenji (1995). The History of Language Laboratories—Origin and Establishment.
ERIC, https://eric.ed.gov/?id=ED381020.
Klimová, Blanka and Seraj, Prodhan Mahbub Ibna (2023). The Use of Chatbots in
University EFL Settings: Research Trends and Pedagogical Implications. Frontiers in
Psychology 14: https://doi.org/10.3389/fpsyg.2023.1131506.
Krashen, Stephen D. (1982). Principles and Practice in Second Language Acquisition.
Oxford: Pergamon.
Krashen, Stephen D. (1985). The Input Hypothesis. London: Longman.
Kurzweil, R. (2012). How to Create a Mind: The Secret of Human Thought Revealed. New
York: Viking.
Kurzweil, Ray (2005). The Singularity Is Near. Harmondsworth: Penguin.
Lado, Robert (1957). Linguistics across Cultures: Applied Linguistics for Language
Teachers. Ann Arbor: University of Michigan Press.
Lado, Robert (1964). Language Teaching: A Scientific Approach. New York: McGraw-
Hill.
Lakoff, George (1987). Women, Fire, and Dangerous Things: What Categories Reveal
about the Mind. Chicago: University of Chicago Press.
Lakoff, George and Johnson, Mark (1980). Metaphors We Live By. Chicago: Chicago
University Press.
Lakoff, George and Johnson, Mark (1999). Philosophy in Flesh: The Embodied Mind and
Its Challenge to Western Thought. New York: Basic.
Lakoff, Robin (1975). Language and Woman’s Place. New York: Harper and Row.
Lamendella, John T. (1979). The Neurofunctional Basis of Pattern Practice. TESOL
Quarterly 13: 5-19.
Langacker, R. W. (1987). Foundations of Cognitive Grammar. Stanford: Stanford
University Press.
Langacker, Ronald W. (1990). Concept, Image, and Symbol: The Cognitive Basis of
Grammar. Berlin: Mouton de Gruyter.
Langacker, Ronald W. (1999). Grammar and Conceptualization. Berlin: Mouton de
Gruyter.
 References 137

Laufer, Batia and Waldman, Tina (2011). Verb–Noun Collocations in Second Language
Writing: Corpus Analysis of Learners’ English. Language Learning 61: 647-672.
Lenci, Alessandro (2022). Verso Babele e oltre! Promesse e scommesse della nuova
Intelligenza Artificiale. Linguisticamente [Towards Babel and beyond! Promises and
bets of the new Artificial Intelligence. Linguistically], https://www.linguisticamente.
org/verso-babele-e-oltre-le-promesse-scommesse-della-nuova-intelligenza-
artificiale/.
Léon, Pierre R. (1962). Laboratoire de langues et correction phonétique [Language
laboratory and phonetic correction]. Paris: Didier.
Levy, Michael (1997). Computer-Assisted Language Learning. Oxford: Clarendon Press.
Lipari, Lisbeth (2014). Listening, Thinking, Being. University Park: Penn State University
Press.
Luckin, Rose (2020). AI in Education Will Help Us Understand How We Think. Financial
Times, https://www.ft.com/content/4f24adca-5186-11ea-8841-482eed0038b1.
Ludanskaa, Alissa (2023). How to Learn Spanish with ChatGPT’s AI Tool. Lingopole,
https://lingopie.com/blog/how-to-learn-spanish-with-chatgpts-ai-tool/.
Malinowski, Bronislaw (1922). The Problem of Meaning in Primitive Languages. In: C. K.
Ogden and I. A. Richards (eds.), The Meaning of Meaning, pp. 296-336. London:
Kegan Paul, Trench and Trubner.
Marr, Bernard (2023). The Intersection Of AI And Human Creativity: Can Machines Really
Be Creative? Forbes, www.forbes.com/sites/bernardmarr/2023/03/27/the-
intersection-of-ai-and-human-creativity-can-machines-really-be-creative/?.
Maturana, Humberto and Varela, Francisco (1973). Autopoiesis and Cognition. Dordrecht:
Reidel.
Maurizio, Carmelina (2021). Apprendere le lingue con l’intelligenza artificiale: app e
modelli utili. Agenda Digitale [Learning languages with artificial intelligence: useful
apps and models. Digital Agenda], https://www.agendadigitale.eu/cultura-digitale/ai-
e-apprendimento-delle-lingue-modelli-e-app-che-possono-aiutare/.
McLuhan, Marshall (1960). Classrooms without Walls. In: Edmund Carpenter (ed.),
Explorations in Communication. Boston: Beacon Press.
McLuhan, Marshall (1962). The Gutenberg Galaxy: The Making of Typographic Man.
Toronto: University of Toronto Press.
McLuhan, Marshall (1964). Understanding Media: The Extensions of Man. Cambridge,
MA: MIT Press.
McLuhan, Marshall (1968). Through the Vanishing Point. New York: Harper & Row.
McLuhan, Marshall (1970). Culture is Our Business. New York: Wipf & Stock.
McLuhan, Marshall and Leonard, George B. (1967). The Future of Education: The Class
of 1989. Look, February 21, pp. 23-24.
Memisevic, Roland (2022). Can Neural Networks “Think” in Analogies? Qualcomm,
https://www.qualcomm.com/news/onq/2022/09/Can-neural-networks-think-in-
analogies.
Morris, Charles W. (1938). Foundations of the Theory of Signs. Chicago: University of
Chicago Press.
Morris, Charles W. (1946). Signs, Language and Behavior. Englewood Cliffs, N. J.:
Prentice-Hall.
138 Marcel Danesi

Mumford, Michael D. (2003). Where Have We Been, Where Are We Going? Taking Stock
in Creativity Research. Creativity Research Journal 15: 107-120.
Neuman, Yair (2014). Introduction to Computational Cultural Psychology. Cambridge:
Cambridge University Press.
Neuman, Yair, Danesi, Marcel, and Vilenchik, Dan (2023). Using AI for Dialoguing with
Texts: From Psychology to Cinema and Literature. London: Routledge.
Nielsen, Michael (2012). Reinventing Discovery: The New Era of Networked Science.
Princeton: Princeton University Press.
Nghi, Tran Tin, Phuc, Tran Huu, and Tat, Thang Nguyen (2019). Applying AI Chatbot For
Teaching A Foreign Language: An Empirical Research. International Journal of
Scientific & Technology Research 8 (11): 897-902.
Norafkan, Mehran (2013). Learnability of Cultural Models through Authentic Materials:
Focus on Metaphorical Competence and Conceptual Fluency. Doctoral Dissertation,
Simon Fraser University.
Omaggio, Alice (1986). Teaching Language in Context. Boston: Heinle and Heinle.
Pardede, Parlindungan (2010). The Role of Pronunciation in a Foreign Language Program.
Conference paper presented at English Bimonthly Collegiate Forum, Jakarta,
Indonesia.
Palmer, Harold (1917). The Scientific Teaching and Study of Languages. Yonkers-on-
Hudson: World Book.
Palmer, Harold (1921). The Oral Method of Teaching Languages. Cambridge: Heffer.
Palmer, Harold (1922). The Principles of Language-Study. Oxford: Oxford University
Press.
Pegrum Mark (2009) From Blogs to Bombs: The Future of Digital Technologies in
Education. Perth: University of Western Australia Press.
Peterson, Mark (2013). Computer Games and Language Learning. New York: Palgrave
Macmillan.
Peters, Elke (2016). The Learning Burden of Collocations: The Role of Interlexical and
Intralexical Factors. Language Teaching Research 20 113-138.
Plato (1999). Cratylus, Project Gutenberg, ia804508.us.archive.org/26/items/cratylus.
Richards, Jack C. and Rodgers, Theodore S. (1986). Approaches and Methods in Language
Teaching: A Description and Analysis. Cambridge: Cambridge University Press.
Ringelberg, Kalli (2023). 21 AI and Automation Quotes to Inspire Leaders. Automation
Blog, https://www.wrk.com/blog/ai-automation-inspirational-quotes/.
Rivers, Wilga (1964). The Psychologist and the Foreign Language Teacher. Chicago:
University of Chicago Press.
Roy, Kavika (2022). Why Artificial Intelligence Will Never Replace Teachers. Medium,
medium.com/datatobiz/why-artificial-intelligence-will-never-replace-teachers-
c9d96dc7a12.
Rudra, Suchi (2021). Chatbots Allow Educators to Delegate Repetitive Tasks and Focus on
Teaching: AI-Enabled Chatbots Are Taking on the Roles of Tutor, College Adviser
and School Administration Assistant. Ed Tech Magazine, https://edtechmagazine.
com/k12/article/2021/09/chatbots-allow-educators-delegate-repetitive-tasks-and-
focus-teaching.
 References 139

Sabzalieva, Emma and Valentini, Arianna (2023). ChatGPT and Artificial Intelligence in
Higher Education. Paris: UNESCO.
Saussure, Ferdinand de (1916). Cours de linguistique générale [General linguistics
course]. Paris: Plon.
Schank, Roger C. (1984). The Cognitive Computer. Reading, Mass.: Addison-Wesley.
Scinicariello, Sharon (1997). Uniting Teachers, Learners, and Machines: Language
Laboratories and Other Choices. In: Michael D. Bush and Robert M. Terry (eds.),
Technology-Enhanced Language Learning, 185-213. Lincolnwood: National
Textbook.
Sebeok, Thomas A. and Danesi, Marcel (2000). The Forms of Meaning: Modeling Systems
Theory and Semiotics. Berlin: Mouton de Gruyter.
Selinker, Larry (1972). Interlanguage. International Review of Applied Linguistics 10: 209-
231.
Shannon, Claude E. (1948). A Mathematical Theory of Communication. Bell Systems
Technical Journal 27: 379-423.
Shirazi, Maryam G. and Talebinezhad, Mohammad R. (2013). Developing Intermediate
EFL Learners’ Metaphorical Competence through Exposure. Theory and Practice in
Language Studies 3: 135-141.
Silver, David, Hubert, Thomas, Schrittwieser, Julian, Antonoglou, Ioannis, Lai, Matthew,
Guez, Arthur, Lanctot, Marc, Sifre, Laurent, Kumaran, Dharshan, Graepel, Thore,
Lillicrap, Timothy, Simonyan, Karen, and Hassabis, Demis (2018). A General
Reinforcement Learning Algorithm That Masters Chess, Shogi, and Go Through Self-
Play. Science 362 (6419): 1140-1144.
Simon, Edwige (2023). 6 Ways to Use ChatGPT to Learn a Foreign Language.
International Center for Language Studies, https://www.icls.edu/6-ways-to-use-
chatgpt-to-learn-a-foreign-language.
Skjuve, Marita and Brandtzaeg, Petter Bae (2019). Measuring User Experience in Chatbots:
An Approach to Interpersonal Communication Competence. In: S. Bodrunova (ed.),
Internet Science. Cham: Springer.
Stern, H, H. (1983). Fundamental Concepts of Language Teaching. Oxford: Oxford
University Press.
Sweet, Henry (1899). The Practical Study of Languages: A Guide for Teachers and
Learners. London: J. M. Dent & Sons.
Sydorenko, Tetyana, Hsieh, Ching-Ni, Ahn, Seongmee, and Arnold, Nike (2017). Foreign
Language Learners’ Beliefs about CALL: The Case of a U.S. Midwestern University.
Computer Assisted Language Instruction Consortium Journal 34 (2): 196-218.
Taulli, Tom (2023). Generative AI: How ChatGPT and Other AI Tools Will Revolutionize
Business. New York: Springer.
Thornbury, Scott and Slade, Diana (2006). Conversation: From Description to Pedagogy.
Cambridge: Cambridge University Press.
Timothy, Maxwell (2023). How to Use ChatGPT as a Language Translation Tool. Make
Use Of, https://www.makeuseof.com/how-to-translate-with-chatgpt/.
Titone, Renzo (1968). Teaching Foreign Languages: An Historical Sketch. Washington, D.
C.: Georgetown University Press.
140 Marcel Danesi

Tonini, Elisabetta, Bischetti, Luca, Del Sette, Paola, Tosi, Eleonora, Lecce, Serena, and
Bambini, Valentina (2023). The Relationship between Metaphor Skills and Theory of
Mind in Middle Childhood: Task and Developmental Effects. Cognition 238:
https://www.sciencedirect.com/science/article/abs/pii/S0010027723001385.
Turner, Mark (2015). The Origin of Ideas: Blending, Creativity, and the Human Spark.
Oxford: Oxford University Press.
Van Ek, J. A. (1975). The Threshold Level in a European Unit/Credit System for Modern
Language Teaching by Adults. Strasbourg: Council of Europe.
Viëtor, Wilhelm (1882). Der Sprachunterricht muss umkerhren! Ein Beitrag zur
Uberündungsfrages [Language teaching must turn around! A contribution to the
question of transition]. Heilbronn: Henninger.
Vygotsky, Lev S. (1962). Thought and Language. Cambridge, Mass.: MIT Press.
Wachowiak, Lennart and Gromann, Dagmar (2023). Does GPT-3 Grasp Metaphors?
Identifying Metaphor Mappings with Generative Language Models. Proceedings of
the 61st Annual Meeting of the Association for Computational Linguistics Volume 1:
Long Papers, pages 1018-1032.
Warschauer, Mark and Healey, Deborah (1998). Computers and Language Learning: An
Overview. Language Teaching 31: 57-71.
Weizenbaum, Joseph (1966). ELIZA: A Computer Program for the Study of Natural
Language Communication between Man and Machine. Communications of the
Association for Computing Machinery 9: 36-45.
Widdowson, Henry G. (1978). Teaching Language as Communication. Oxford: Oxford
University Press.
Wierzbicka, Anna (2003). Cross-cultural pragmatics: The semantics of human interaction.
Berlin: Mouton de Gruyter.
Wilkins, David A. (1976). Notional Syllabuses. Oxford: Oxford University Press.
Williams, Keir (2023). Using ChatGPT to Enhance Written Communication: How
ChatGPT Spelling and Grammar Assistance Can Be Used to Benefit Dyslexic
Students. Dyslexia UK: https://www.dyslexiauk.co.uk/using-chatgpt-to-enhance-
written-communication-how-chatgpt-spelling-and-grammar-assistance-can-be-used-
to-benefit-dyslexic.
Winograd, Terry (1971). Procedures as a Representation for Data in a Computer Program
for Understanding Natural Language. MIT Library: hdl:1721.1/7095.
Witkin, Herman A. and Goodenough, Donald R. (1981). Cognitive Styles: Essence and
Origins. New York: International Universities Press.
Yin, Jiaqi, Goh, Tiong-Thye, Yang, Bing, and Xiaobin, Yang (2020). Conversation
Technology With Micro-Learning: The Impact of Chatbot-Based Learning on
Students’ Learning Motivation and Performance. Journal of Educational Computing
Research 59 (1): 073563312095206.
Yin, Qinghua e Satar, Müge (2020). English as a Foreign Language Learner Interactions
with Chatbots: Negotiation for Meaning. International Online Journal of Education
and Teaching 7: 390-410.
Zengengo (2023). https://www.zengengo.com/blog/how-to-use-chatgpt-in-your-foreign-
language-classroom.
 References 141

Zhai, Chunpeng and Wibowo, Santoso (2022). A Systematic Review on Cross-Culture,

Humor and Empathy Dimensions in Conversational Chatbots: The Case of Second
Language Acquisition. Heliyon 8: e12056.
Zheng, Danning, Song, Ruihua, Hu, Tianran, Fu, Hao, and Zhou, Jin (2020). Love is as
Complex as Math: Metaphor Generation System for Social Chatbot. arXiv,
https://doi.org/10.48550/arXiv.2001.00733.
Index

A blends, 14
Busuu, 130
acquisition, 10, 26, 29, 31, 32, 33, 48, 63,
86, 99, 117, 118, 136, 141
C
adjacency pairs, 76
affective filter, 32, 48, 62, 118 calquing, 89, 96, 97
algorithm, 26, 30, 34, 39, 40, 70, 82, 108, cataphoric, 54, 78
139 categorization, 54
AlphaGo, vii, viii Chagga, 93
analogy, 28, 29, 33 chatbot, vii, viii, ix, 16, 19, 25, 26, 34, 36,
anaphoric, 54, 78 38, 41, 43, 46, 47, 48, 49, 50, 51, 52, 53,
Arabic, 51, 107 54, 55, 57, 58, 59, 60, 63, 65, 68, 69, 71,
Aristotle, 53 77, 78, 80, 82, 83, 84, 85, 86, 87, 90, 91,
Army Specialized Training Program, 47 95, 101, 103, 104, 105, 106, 107, 108,
artificial intelligence (AI), vii, viii, ix, 1, 2, 111, 112, 113, 116, 117, 118, 121, 122,
12, 15, 16, 17, 18, 19, 20, 23, 25, 26, 27, 123, 125, 129, 130, 138, 140, 141
30, 32, 36, 37, 38, 40, 41, 42, 43, 47, 48, ChatGPT, viii, ix, 16, 17, 18, 19, 25, 30,
51, 57, 58, 61, 62, 63, 66, 69, 70, 75, 80, 31, 32, 36, 45, 46, 47, 48, 49, 50, 51, 52,
82, 85, 87, 90, 96, 99, 102, 112, 113, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 65,
116, 117, 118, 119, 120, 121, 122, 123, 66, 68, 69, 71, 72, 73, 77, 78, 80, 81, 82,
124, 125, 126,127, 128, 129, 133, 134, 84, 85, 86, 90, 91, 92, 95, 97, 99, 101,
135, 137, 138, 139 102, 104, 105, 106, 107, 111, 112, 113,
audio-lingual method (ALM), 1, 4, 5, 6, 7, 114, 115, 116, 117, 121, 122, 125, 127,
8, 10, 11, 12, 20, 30, 37, 66 129, 130, 133, 134, 135, 137, 139, 140
audio-visual method, 5 Chinese, 17, 19, 25, 50, 51, 53, 83, 136
Austin, John, 68, 94, 133 Chomsky, Noam, 22, 33, 134
autopoiesis, 113, 137 classroom without walls, 21, 22, 83
cognitive linguistics, 89, 91
B collocation, 34
Comenius, 10, 67
Babbel, 129 communicative competence, 7, 35, 45, 66,
Bantu, 93 67, 68, 72, 73, 74, 78, 82, 85, 86, 89, 92,
behaviorism, 134 94
Black, Max, 43, 133
blended pedagogy, 15, 123
144 Index

computer-assisted-language-learning drills, 4, 6, 9, 26, 29, 46, 47, 48, 55, 58,

(CALL), vii, 1, 2, 12, 13, 14, 15, 23, 42, 122
78, 93, 95, 112, 135, 139 drops, 131
conceptual blending, 109 Duolingo, 2, 130
conceptual competence, 45, 89, 90, 91, 92,
93, 94, 95, 96, 98, 99, 101, 103, 107 E
conceptual fluency, 89, 91, 98, 99, 107,
109, 111 ELSA, 2
conceptual interference, 91 emotional needs, 26
conceptual metaphor, 92, 93, 95, 98, 99, English, 2, 6, 25, 27, 28, 42, 45, 46, 49, 50,
100, 101, 106 54, 57, 59, 60, 72, 80, 81, 83, 85, 89, 91,
conceptual metaphor theory, 89, 90, 92, 98, 93, 94, 95, 96, 99, 100, 101, 103, 104,
100, 102 105, 133, 135, 136, 137, 138, 140
conceptual transfer, 91, 104
conceptualization, 42, 91, 92, 103, 104, F
108, 136
connotative, 59, 99 Farsi, 107
context, 8, 17, 25, 29, 32, 34, 40, 41, 42, feedback, 12, 17, 18, 22, 39, 41, 47, 50, 54,
43, 54, 56, 61, 62, 69, 70, 71, 72, 82, 85, 57, 61, 62, 84, 106, 117, 118, 121, 122
90, 94, 95, 97, 113, 116, 118, 121, 133, figurative, 59, 92, 97, 107, 108, 124
138 folk wisdom, 93
contrastive analysis (CA), 8, 10, 46, 47, 48, foreign language learning (FL-learning),
72, 78, 79, 104, 106 viii, ix, 4, 7, 8, 10, 11, 12, 13, 23, 25,
contrastive conceptual analysis, 104 26, 27, 29, 30, 31, 34, 39, 43, 52, 83, 87,
conversation analysis, 70 97, 101, 103, 117, 120, 123, 124, 125,
conversations, vii, 17, 35, 36, 41, 42, 47, 134, 135, 136
54, 57, 71, 72, 73, 74, 75, 76, 78, 79, 80, foreign language teaching (FLT), viii, ix, 1,
82, 86, 104, 107, 121, 122, 129, 131 2, 4, 6, 7, 8, 10, 12, 13, 14, 15, 16, 17,
coreference, 54, 57 19, 22, 23, 26, 33, 35, 37, 38, 41, 47, 50,
Council of Europe, 68, 127, 135, 140 53, 57, 62, 63, 65, 66, 67, 68, 82, 83, 84,
culture, ix, 5, 12, 13, 19, 21, 37, 42, 55, 63, 85, 87, 90, 91, 92, 99, 111, 112, 113,
65, 66, 67, 72, 74, 84, 85, 86, 93, 95, 98, 119, 120, 123, 124, 126, 127, 129
99, 101, 113, 114, 122, 125, 137, 141 French, 36, 51, 52, 58, 69, 71, 107
Fries, Charles, 6, 30, 135
Frye, Northrop, 93, 135
D
DALL-E, 129 G
deep learning, ix, x, 40, 41, 48, 59, 70, 92,
112, 127 games, vii, 11, 13, 41, 61, 62, 63, 134, 138
denotative, 59, 98 gamification, 62
dialect, 28 Generative AI, 1, 2, 16, 32, 45, 139
dialogue, 2, 7, 9, 18, 21, 22, 29, 65, 66, 71, German, 49
75, 76, 78, 85, 86, 97, 99, 133 Gouin, François, 8, 135
direct method (DM), 8, 9, 10, 26, 29, 30, grammar, 5, 8, 9, 16, 17, 21, 27, 29, 30, 32,
31, 37, 62, 66, 67, 91 33, 34, 35, 36, 44, 45, 51, 52, 53, 54, 55,
 Index 145

56, 57, 58, 59, 61, 62, 63, 65, 79, 81, 83, irony, 2, 74, 98, 100, 102, 117
92, 93, 108, 111, 114, 115, 116, 121, Italian, 5, 18, 25, 27, 42, 46, 59, 68, 80, 84,
122, 130, 131, 134, 136, 140 85, 89, 91, 96, 100, 101, 102, 103, 104,
grammar-translation, 27, 29, 44 106, 134
graphemics, 49
Greek, 53, 107 J
H Japanese, 53
Johnson, Mark, 89, 98, 100, 136
habit-formation, 4, 5
hallucination, 30 K
Hungarian, 107
Hymes, Dell, 35, 66, 68, 136 Korean, 51, 53, 81
Krashen, Stephen, 29, 31, 32, 62, 99, 117,
I 134, 136

idiom, 25 L
image schema, 91, 96, 99, 100, 101, 103,
104, 106, 108 Lado, Robert, 42, 136
imitation, 5, 6, 7, 9, 10, 29, 33 Lakoff, George, 79, 89, 98, 100, 136
individualized learning, 12 Langacker, Ronald, 108, 136
induction, 6, 30 language acquisition device, 33
inductive process, 30 language laboratory, 1, 2, 4, 10, 12, 137
industrial revolution, 9 language learning apps, 129
information theory, 39 learner-centric pedagogy, 120
input, 7, 9, 11, 16, 27, 30, 31, 32, 33, 39, learning style, 16, 18, 26, 37, 38, 39, 49,
40, 48, 49, 51, 62, 63, 81, 92, 99, 101, 61, 84, 114, 120
107, 112, 113, 117, 136 lesson plan, 16, 18, 119
input hypothesis, 31, 32, 63, 136 lexeme, 52
instruction, ix, 5, 6, 9, 12, 13, 18, 27, 29, linguistic competence, 32, 35, 45, 61, 66,
31, 33, 37, 51, 59, 68, 82, 100, 101, 117, 68
118, 139 linguistic metaphor, 92, 98
interactional pedagogy, 85 linguistics, 4, 8, 27, 33, 48, 70, 78, 134,
interactivity, 11, 62 135, 136, 139, 140
interference, 27, 28, 89 listening, 4, 9, 17, 20, 29, 41, 48, 49, 67,
interlanguage, 27, 28, 29, 30, 31, 39, 51, 111, 133, 137
139 literary translation, 90, 92
interlanguage theory, 27, 28, 29, 30
interlinguistic error, 27 M
International Phonetic Association, 9
internet, 11, 12, 15, 18, 20, 21, 22, 25, 33, machine learning, 16, 39, 40, 41, 43, 62,
39, 63, 71, 101, 106, 107, 108, 112, 121, 69, 70, 71, 112, 127
122, 125, 139 Malinowski, Bronislaw, 35, 137
intralinguistic error, 29 Mandarin, 17, 107
inuit, 42 mapping, 98, 99, 100, 102
146 Index

McLuhan, Marshall, 15, 20, 21, 22, 23, 83, pedagogy, ix, 8, 9, 13, 14, 15, 23, 27, 37,
113, 124, 125, 126, 127, 137 43, 62, 84, 85, 86, 91, 102, 106, 112,
Memrise, 130 119, 120, 121, 123, 125, 134, 139
metaphor, 33, 59, 60, 90, 92, 93, 95, 98, personalization, 38, 41, 62, 117
99, 100, 102, 106, 107, 109, 134, 135, phatic communion, 35
140, 141 phonetic, 17, 49, 137
metaphorical competence, 89, 98, 99, 101 phonics, 50
metonymy, 98, 100, 102 phonology, 45, 46, 48, 49
Modern Language Association of America, pinyin, 17
8 Plato, 53, 138
Modern Language Association of Great plugin, 47, 60, 106, 135
Britain, 8 Polish, 107
Mondly, 131 polysemy, 39, 58, 59, 70
monitor, 4, 19, 118 positive transfer, 8, 30, 72, 96, 103
monitoring, 12, 31, 32 pragmatic model, 35, 36
morpheme, 52 pragmatics, 34, 36, 70, 78, 140
morphology, 53 proficiency movement, 38, 68
Morris, Charles, 34, 137 pronunciation, 7, 8, 9, 30, 41, 45, 46, 47,
motivation, 16, 19, 37, 38, 39, 114, 116, 48, 49, 57, 58, 59, 61, 62, 83, 114, 130,
118, 133, 135, 136, 140 138
multimedia, 11, 135 proverb, 93, 123
psychology, 4, 8, 10, 27, 78, 135, 136, 138
N puzzles, 63, 134

native language (NL), 2, 7, 8, 9, 10, 18, 21, Q

25, 27, 28, 29, 30, 31, 49, 51, 52, 57, 72,
73, 86, 89, 90, 91, 96, 99, 103 quizzes, 18, 19, 20
negative transfer, 8, 27, 30, 72, 97, 103,
104 R
neural network, x, 16, 59, 71, 96, 127
N-gram, 39 reading, 5, 9, 17, 20, 29, 37, 41, 50, 51, 58,
67, 139
reform movement, 27, 43, 126, 127
O
register, 16, 69, 71, 81, 86, 108, 129
OpenAI, viii, ix, 129 repetition, 5, 6, 10, 78, 82
oral method, 35 role of the teacher, 14, 15
oral speech, 17, 41 role-playing, 7, 18
orthography, 49 Rosetta Stone, 130

P S
Palmer, Harold, 10, 35, 138 scripts, 51, 72, 73, 80, 92, 94
pattern practice, 5, 6, 7, 9, 29, 47 second language teaching, 2
Selinker, Larry, 27, 139
semantics, 34, 54, 140
 Index 147

sequence organization, 76, 77 translation, 18, 25, 29, 61, 84, 89, 90, 104,
Shannon, Claude, 39, 139 105, 106, 111, 113, 114, 133, 134, 139
SHRDLU, 75, 76 turn-taking, 76, 77
singularity, viii, 136
small talk, 72, 73, 75 U
social framing, 79
source domain, 93, 95, 97, 98, 99, 100, Universal Grammar (UG), 30, 33
101, 102, 103 Universal Grammar Theory, 33
Spanish, 25, 45, 46, 56, 60, 61, 65, 66, 67,
68, 72, 77, 96, 97, 99, 104, 105, 106, V
111, 114, 133, 137
speaking, 3, 4, 9, 17, 27, 29, 41, 45, 46, 47, Vietnamese, 53
48, 64, 67, 73, 80, 83, 91, 95, 111, 121, Viëtor, Wilhelm, 8, 140
131 virtual assistant, 38, 57, 67, 112, 114
speech act, 68, 72, 73, 74, 79, 94, 101 virtual reality (VR), 12, 41, 83, 131
speech recognition, 48, 61, 62, 82 vocabulary, 7, 9, 17, 29, 30, 36, 45, 48, 51,
Sweet, Henry, 10, 93, 103, 139 58, 59, 60, 61, 62, 63, 65, 83, 92, 111,
syntax, 34, 53, 71, 134 114, 115, 116, 121, 130, 131
voice recognition, 41
Vygotsky, Lev, 32, 140
T
target domain, 98, 100, 101, 102 W
technology, viii, 1, 2, 5, 6, 11, 13, 14, 15,
20, 22, 23, 41, 42, 48, 83, 85, 113, 119, Weizenbaum, Joseph, viii, 140
120, 122, 123, 125, 126, 127, 128, 133, Winograd, Terry, 75, 140
134, 135, 138, 139, 140 word class, 53
tests, 18, 19, 20, 38, 41, 45, 86 writing, ix, 5, 9, 16, 17, 19, 23, 29, 41, 42,
textbook, 37, 45, 55, 65, 69, 78, 82, 84, 49, 51, 55, 67, 70, 80, 106, 113, 116,
101, 123, 124, 125, 126, 139 117, 121, 122, 129, 130, 133, 134, 137
theory of mind, 95, 135, 140
tracking, 71, 130 Z
transfer theory, 7, 10, 27, 29, 30
transformer, ix, 16, 27, 111 zones of proximal development, 32