MODULE 3: TRANSLATION AND

TECHNOLOGY
NDJIMI MALAKA, PHD.

The lecturer, Ndjimi Malaka, PhD.

 AIMS
• TO INTRODUCE TRANSLATION TECHNOLOGY
• TO DISCUSS THE DEFINITIONS OF TERMS REFERRING TO THE USE OF COMPUTERS IN
TRANSLATION ACTIVITIES.
• TO DISCUSS THE DEVELOPMENT OF DIFFERENT MACHINE TRANSLATION AND COMPUTER-AIDED
TRANSLATION TOOLS AND THEIR USES.
• TO DESCRIBE AND INVESTIGATE SPECIFIC MT APPROACHES .
• TO GIVE A DETAILED DESCRIPTION OF DIFFERENT MACHINE TRANSLATION SYSTEM DESIGNS
ALSO KNOWN AS ‘ARCHITECTURES’.
• TO EVALUATE TRANSLATION TOOLS.
• TO REFLECT ON CURRENT DEVELOPMENTS IN TRANSLATION TECHNOLOGY.
The lecturer, Ndjimi Malaka, PhD.
 CONTENTS

1. DEFINITION OF TERMS
2. MACHINE TRANSLATION
3. MACHINE TRANSLATION SYSTEMS
4. COMPUTER-AIDED TRANSLATION TOOLS AND RESOURCES
5. EVALUATING TRANSLATION TOOLS
6. RECENT DEVELOPMENTS AND FUTURE DIRECTIONS

The lecturer, Ndjimi Malaka, PhD.

 KEY QUESTIONS

a. WHAT IS INVOLVED IN TT?

b. WHAT MT SYSTEMS ARE YOU AWARE OF?
c. CAN TECHNOLOGY REPLACE HUMAN TRANSLATORS?
d. IS IT POSSIBLE TO HAVE A ‘PERFECT MT’?
e. WHY IS TRANSLATION TECHNOLOGY IMPORTANT IN TODAY’S GLOBALIZED WORLD?
f. HOW DOES TRANSLATION TECHNOLOGY HELP BUSINESSES?

The lecturer, Ndjimi Malaka, PhD.

• EXPAND ON THE FOLLOWING ASSERTION:
 TRANSLATION TECHNOLOGY ALLOWS COMPANIES TO GO GLOBAL QUICKLY, EASILY, AND
AFFORDABLY BY PROVIDING THE TOOLS THEY NEED TO CREATE, LOCALIZE, MANAGE, AND
PUBLISH CONTENT FOR AUDIENCES WORLDWIDE. SCALABILITY, AUTOMATION, INTELLIGENCE,
AND ACCURACY BECOME ACHIEVABLE AND UNLOCK BUSINESS OPPORTUNITIES THAT WERE
HARDLY IMAGINABLE WITH MANUAL TRANSLATION PROCESSES.

The lecturer, Ndjimi Malaka, PhD.

 1. DEFINITION OF TERMS

AIM: TO DISCUSS THE DEFINITIONS OF TERMS REFERRING TO THE USE OF COMPUTERS IN

TRANSLATION ACTIVITIES.
TOPICS:
A. MACHINE TRANSLATION (MT);
B. HUMAN-AIDED/ASSISTED MACHINE TRANSLATION (HAMT);
C. MACHINE-AIDED HUMAN TRANSLATION (MAHT)
D. COMPUTER-AIDED/ASSISTED TRANSLATION (CAT);

The lecturer, Ndjimi Malaka, PhD.

A. MACHINE TRANSLATION
PROCESS OF USING COMPUTER SOFTWARE TO TRANSLATE A TEXT OR SPOKEN WORDS FROM
ONE LANGUAGE TO ANOTHER.
B. HUMAN-AIDED/ASSISTED MACHINE TRANSLATION (HAMT)
HUMAN-AIDED MACHINE TRANSLATION (HAMT) IS THE FUSION OF HUMAN EXPERTISE AND AI-
POWERED MACHINE TRANSLATION. HAMT COMBINES THE STRENGTHS OF BOTH APPROACHES TO
OVERCOME THE LIMITATIONS OF PURE MT. THIS METHODOLOGY INVOLVES PROFESSIONAL
HUMAN TRANSLATORS WORKING ALONGSIDE AI SYSTEMS TO ENHANCE TRANSLATION QUALITY.
C. MACHINE-AIDED HUMAN TRANSLATION (MAHT)
IN MACHINE TRANSLATION (MT), TRANSLATION PROPER IS PERFORMED BY A COMPUTER, EVEN IF
THE HUMAN HELPS BY PREEDITING, POSTEDITING, OR ANSWERING QUESTIONS TO DISAMBIGUATE
THE SOURCE TEXT. IN COMPUTER-AIDED TRANSLATION, OR MORE PRECISELY MACHINE-AIDED
HUMAN TRANSLATION (MAHT), BY CONTRAST, TRANSLATION IS PERFORMED BY A HUMAN, AND
THE COMPUTER OFFERS SUPPORTING TOOLS.

The lecturer, Ndjimi Malaka, PhD.

D. COMPUTER-AIDED/ASSISTED TRANSLATION (CAT);
COMPUTER-ASSISTED TRANSLATION (CAT) IS A SOFTWARE APPLICATION USED IN WEBSITE
TRANSLATION WORKFLOWS, HELPING TO AUTOMATE THE PROCESS. THIS TRANSLATION TOOL
ALLOWS TRANSLATORS AND REVIEWERS TO HANDLE TRANSLATION AND REVISION TASKS
FASTER, EFFICIENTLY, AND CONSISTENTLY.

The lecturer, Ndjimi Malaka, PhD.

 2. MACHINE TRANSLATION (MT)

• AIMS:
• TO DISCUSS THE DEVELOPMENT OF DIFFERENT MACHINE TRANSLATION TOOLS.
TOPICS:
a. OVERVIEW
b. MACHINE TRANSLATION MODEL
c. LEVELS OF MT

The lecturer, Ndjimi Malaka, PhD.

ORIGINALLY, AUTOMATIC SYSTEMS WITH NO HUMAN INVOLVEMENT.
THE APPLICATION OF COMPUTERS TO THE TASK OF TRANSLATING TEXTS FROM
ONE NATURAL LANGUAGE TO ANOTHER.
INPUT IN THE FORM OF FULL SENTENCES AT A TIME AND GENERATING
CORRESPONDING FULL SENTENCES (NOT NECESSARILY OF GOOD QUALITY).
THE ATTEMPT TO AUTOMATE ALL OR PART OF THE PROCESS OF TRANSLATING
FROM ONE HUMAN LANGUAGE TO ANOTHER.

The lecturer, Ndjimi Malaka, PhD.

 Machine translation model

The lecturer, Ndjimi Malaka, PhD.

 2.1 LEVELS OF MT

The lecturer, Ndjimi Malaka, PhD.

 1.Basic level system
Less than 50,000 entries
in its largest dictionary.

Restricted dictionary
expansion.

Suitable for home use.

Restricted to single-
clause/basic sentence
translations.
The lecturer, Ndjimi Malaka, PhD.
 2. STANDARD LEVEL SYSTEM
1. HAS MORE THAN 50,000 ENTRIES IN ITS LARGEST
DICTIONARY,
2. ALLOWS FOR DICTIONARY EXPANSION,
3. ALLOWS MORE THAN SINGLE-CLAUSE/BASIC
SENTENCE TRANSLATIONS,
4. IS SUITABLE FOR HOME USE AND STAND-ALONE
OFFICE USE.
The lecturer, Ndjimi Malaka, PhD.
 Has more than 75,000 entries
in its smallest dictionary,

Allows for dictionary

3. Advanced expansion,
level system
Allows more than single-clause/basic
sentence translations, and

Is suitable for offices with

networked facilities.
The lecturer, Ndjimi Malaka, PhD.
 MACHINE TRANSLATION SYSTEM BASED ON USAGE

The lecturer, Ndjimi Malaka, PhD.

 Home - refers to machine translation systems
for home users who have few or no
translation skills.

Online - is designed specifically for the

translation of electronic documents obtained
from the Web.

The third type is designed for professional

translators,

The last for employees of large companies.

The lecturer, Ndjimi Malaka, PhD.

EXAMPLES OF SUCCESSFUL COMMERCIAL MACHINE
TRANSLATION SYSTEMS

• SPANAM (SPANISH AMERICAN) AND ENGSPAN (ENGLISH SPANISH), WHICH

WERE DEVELOPED BY THE PAN AMERICAN HEALTH ORGANIZATION (PAHO).

• EXAMPLES OF EXPERIMENTAL MACHINE TRANSLATION SYSTEMS INCLUDE

THE ALT-J/E (AUTOMATIC LANGUAGE TRANSLATOR JAPANESE TO ENGLISH),
AND ALT-J/C (AUTOMATIC LANGUAGE TRANSLATOR JAPANESE TO CHINESE)
SYSTEMS, BOTH DEVELOPED BY NIPPON TELEGRAPH AND TELEPHONE
CORPORATION (NTT).
The lecturer, Ndjimi Malaka, PhD.
 DISCUSSION QUESTIONS
1. IS MT EXCLUSIVE TO MACHINES? DO YOU THINK HUMANS SHOULD BE INVOLVED? WHAT
SHOULD THEIR ROLE BE?
2. CAN AN EXPERIENCED TRANSLATOR USE A BASIC LEVEL SYSTEM? IF YES, IN WHAT
SITUATIONS? IF NOT, WHY NOT?
3. WHICH TRANSLATION SYSTEM WOULD YOU USE IF YOU HAD TO TRANSLATE A NOVEL?

The lecturer, Ndjimi Malaka, PhD.

 3. MACHINE TRANSLATION SYSTEMS

• AIMS:
a. TO DESCRIBE AND INVESTIGATE SPECIFIC MT APPROACHES .
b. TO GIVE A DETAILED DESCRIPTION OF DIFFERENT MACHINE TRANSLATION SYSTEM DESIGNS
ALSO KNOWN AS ‘ARCHITECTURES’.

The lecturer, Ndjimi Malaka, PhD.

TOPICS:
a. IMPORTANCE
b. MAJOR HISTORICAL DEVELOPMENTS
c. FIRST-GENERATION SYSTEMS
d. SECOND-GENERATION SYSTEMS
e. ARCHITECTURES
f. MACHINE TRANSLATION SYSTEMS APPROACHES
g. HYBRID AND INTERACTIVE MACHINE TRANSLATION SYSTEMS
h. INTERACTIVE MACHINE TRANSLATION SYSTEMS
i. ONLINE MACHINE TRANSLATION SYSTEMS

The lecturer, Ndjimi Malaka, PhD.

• IMPORTANCE:
1. WHEN INSTANT TRANSLATIONS ARE NEEDED, HUMAN TRANSLATORS ARE NOT ABLE TO
SUPPLY THEM FAST ENOUGH.
2. COST-EFFECTIVE ALTERNATIVE TO HUMAN TRANSLATORS IN A VARIETY OF SITUATIONS.

Machine translation is an interdisciplinary enterprise that combines a

number of fields of study such as lexicography, linguistics,
computational linguistics, computer science and language
engineering. It is based on the hypothesis that natural languages can
be fully described, controlled and mathematically coded.
The lecturer, Ndjimi Malaka, PhD.
 3.1 MAJOR HISTORICAL DEVELOPMENTS

 IN 1629, DESCARTES MAY HAVE BEEN THE FIRST TO PROPOSE THE IDEA THAT A LANGUAGE
COULD BE REPRESENTED BY CODES AND THAT WORDS OF DIFFERENT LANGUAGES WITH
EQUIVALENT MEANING COULD SHARE THE SAME CODE.
 IN THE EARLY YEARS OF MACHINE TECHNOLOGY, THE COMMON TERM USED WAS ‘AUTOMATIC
TRANSLATION’ OR ‘MECHANICAL TRANSLATION’.
 ONLY AFTER THE SECOND WORLD WAR THAT THE POSSIBILITIES OF
LANGUAGE TRANSLATION USING STORED-PROGRAM COMPUTERS WERE EXPLORED.

The lecturer, Ndjimi Malaka, PhD.

The lecturer, Ndjimi Malaka, PhD.
 Major historical developments
Georgetown University and IBM introduced the world’s first machine translation
(MT) system. The approach was rule-based and lexicographical, which means
that it relied on pre-programmed rules and dictionaries. Although this early
1950s: form of MT proved unreliable and slow, it was still revolutionary—a stepping
stone on the path toward more advanced technology.

The United States Department of Defense and Defense Advanced Research

1970s: Projects Agency (DARPA) started developing speech recognition technologies
that paved the way for voice-to-text technologies
The arrival of electronic dictionaries and terminological databases during this
decade was another major turning point. These tools helped to make translation
1980s: more accessible by providing translators with instant access to information
(terminology with its translation) that could be used during the project.

The precursors of modern translation management systems (TMS) entered the

Mid-1980s: scene from the hand of Coventry Lanchester Polytechnic University and its ALP
System.

The lecturer, Ndjimi Malaka, PhD.

 Late 1980s to early 1990s: IBM researchers introduced statistical machine
translation (SMT). These systems were word-based and trained to translate
one language into another by comparing large amounts of parallel texts in
both languages (bilingual corpora). For example, they would analyze how
often the German phrase “das auto” was translated as “the car” vs “the
vehicle” vs “the automobile”, and choose the most frequent translation for the
text at hand.

Early 1990s: Most commercial computer-assisted (or aided) translation

(CAT) tools appeared during this decade—a milestone that transformed
translation technology forever. It enabled a whole new generation of
translators to work more efficiently and effectively.

The lecturer, Ndjimi Malaka, PhD.

 Late 1990s: A new version of IBM’s statistical translation
engine, this time phrase-based instead of word-based,
was released. It became the commercial standard for
years to come until Google entered the fray in 2006 with
their neural machine translation (NMT) technology.

Early 2000s: The first cloud-based TMS solutions

appeared in the market, enabling translation teams to
work more flexibly and collaborate with other company
members regardless of location

The lecturer, Ndjimi Malaka, PhD.

 2006: Google launched Google Translate—still statistical—which
took the world by storm. The system first translated the input text into
English before translating it into the target language. The system used
predictive algorithms, which would guess which words should come
next, based on the words and phrases it had “learned” before. These
guesses often resulted in poor grammatical accuracy.

2016: Google Translate introduced neural machine

translation (NMT), which outperformed phrase-based CAT tools and
became the new commercial standard.

The lecturer, Ndjimi Malaka, PhD.

 CHRONOLOGY OF MACHINE TRANSLATION
DEVELOPMENT

The lecturer, Ndjimi Malaka, PhD.

 3.1.1 PIONEER YEARS

• THE PIONEER YEARS BEGAN IN 1949 WITH THE WELL-KNOWN MEMORANDUM

FROM WARREN WEAVER THAT EFFECTIVELY MARKED THE BEGINNING OF MACHINE
TRANSLATION RESEARCH.
• THIS MEMORANDUM FOCUSED ON FOUR ISSUES:
MEANING AND CONTEXT,
LANGUAGE AND LOGIC,
TRANSLATION AND CRYPTOGRAPHY TECHNIQUES,
LANGUAGE AND INVARIANTS

The lecturer, Ndjimi Malaka, PhD.

CRYPTOGRAPHIC TECHNIQUES TO MECHANIZE TRANSLATION:
1. COLOSSUS - BUILT IN 1943 IN LONDON BY TOMMY FLOWERS, AN
ENGINEER AT THE BRITISH GENERAL POST OFFICE.
2. ENIAC (ELECTRONIC NUMERICAL INTEGRATOR AND COMPUTER).
DESIGNED BY J. PRESPER ECKERT AND JOHN WILLIAM MAUCHLY OF
THE UNIVERSITY OF PENNSYLVANIA, ENIAC WAS BUILT IN 1945 BY THE
US ARMY.

The lecturer, Ndjimi Malaka, PhD.

 3.2 FIRST-GENERATION SYSTEMS

The first public

The system
demonstration of a First generation,
was considered
machine translation
employed word- only a ‘toy
system was system’ with
the Russian–English for-word
Georgetown translation 250 words, six
University System, a
collaborative
methods with no grammar rules
clear built-in and 49
effort between IBM
and Georgetown linguistic sentences.
University, carried out component.
in 1954.

The lecturer, Ndjimi Malaka, PhD.

 3.3 SECOND-GENERATION SYSTEMS (LATE 1970S)
SPANAM (Spanish Spanish–English machine
American) translation system

ENGSPAN (English An English–Spanish system by

Spanish) PAHO

METAL (Mechanical A German-English machine translation system built by the US

Translation and Air Force at the University of Texas in Austin with support from
Analysis of Language) Siemens

The lecturer, Ndjimi Malaka, PhD.

 Covered all the languages
spoken in the European
Community at that time (70s-
1992).
EUROTRA (European
Translation)
PaTrans (Patent Translation)
translating patent texts from
SGS English into Danish

CAT2 (Constructors, Atoms and

EUROTRA (European Translators) machine translation
Translation system involving 13 languages in
Germany

The lecturer, Ndjimi Malaka, PhD.

• MACHINE TRANSLATION RESEARCH CONTINUED THROUGHOUT THE 1980S:
 MU MACHINE TRANSLATION SYSTEM - DEVELOPED AT KYOTO UNIVERSITY.
 FIFTH GENERATION MULTILINGUAL INTERLINGUA-BASED MACHINE TRANSLATION SYSTEM PROJECT
(1987–95) - A SYSTEM FOR TRANSLATING IN AND OUT OF CHINESE, INDONESIAN, JAPANESE,
MALAY AND THAI DEVELOPED BY JAPAN’S CENTER OF INTERNATIONAL COOPERATION FOR
COMPUTERIZATION (CICC) AND FOUR ASIAN COUNTRIES.
 PENSEE BY OKI.
 HICATS (HITACHI COMPUTER AIDED TRANSLATION SYSTEM) BY HITACHI.
 MELTRAN-J/E (JAPANESE/ENGLISH) BY MITSUBISHI ELECTRIC CORPORATION.

They all consist of only word and sentence structure analysis with much of the lexical
ambiguities unresolved. Their domains are restricted to certain subject fields such as
computer science and information technology. These machine translation systems
require extensive
The lecturer, Ndjimi Malaka, PhD. pre-editing and post-editing by human translators
 CHECK!

 IMPORTANCE MTS
 MAJOR HISTORICAL DEVELOPMENTS
 DISTINGUISH BETWEEN 1ST AND 2ND GENERATION SYSTEMS

The lecturer, Ndjimi Malaka, PhD.

 3.4 ARCHITECTURES

The components and their configurations involved in a

machine translation system are sometimes referred to
as the machine translation ARCHITECTURES.

The lecturer, Ndjimi Malaka, PhD.

 translation
The function of a source-language
monolingual dictionary is to present
grammatical information (morphology,
A set of monolingual syntax and semantics).

system components and bilingual

dictionaries. A bilingual dictionary is consulted by the
system when a source-language word is
subsequently matched to its target-
language equivalent.

A parser assigns a structure to each string

made up of a word or phrase in the
source-language text based on the stored
grammatical information already pre-
determined for that language.
Machine

Parser
The goal of the parser is to identify the
relationships between source-language
words and their structural representations

The lecturer, Ndjimi Malaka, PhD.

• A STRUCTURAL REPRESENTATION PROVIDES GRAMATICAL INFORMATION
RELATED TO THESE WORDS OR PHRASES.
• FOR EXAMPLE, THE WORD ‘SUPPLIES’ IN THE SENTENCE ‘THE INSTANT HOT AIR
SUPPLIES THE NECESSARY HEAT TO ALL LABORATORIES’ HAS THE STRUCTURAL
REPRESENTATIONS OF A VERB IN THE PRESENT TENSE AND IN THE DECLARATIVE
MOOD.
• THE GRAMATICAL INFORMATION IS ‘ATTACHED’ TO THE WORDS AND PHRASES
OF THE SOURCE-LANGUAGE TEXT BY MEANS OF THE PARSING PROCESS.
• THE CLOSER A SOURCE LANGUAGE IS TO ITS TARGET LANGUAGE
GENEALOGICALLY, FOR EXAMPLE ITALIAN TO SPANISH, THE LESS ELABORATE THE
PARSER TENDS TO BE.
The lecturer, Ndjimi Malaka, PhD.
• ‘THE INSTANT HOT AIR SUPPLIES THE NECESSARY HEAT TO ALL LABORATORIES.’

• (*A-SUPPLIES
(TENSE PRESENT)
(MOOD DECLARATIVE)
(PUNCTUATION PERIOD)
(SOURCE (*O-HOT AIR
(REFERENCE DEFINITE)
(NUMBER SINGULAR)
(ATTRIBUTE (*P-INSTANT))))
• (THEME (*U-HEAT
(REFERENCE DEFINITE)
(NUMBER SINGULAR)
(ATTRIBUTE (*P-NECESSARY))))
• (GOAL_TO (*O-ALL LABORATORIES
• (REFERENCE INDEFINITE)
• (NUMBER PLURAL)))
• EXAMPLE OF STRUCTURAL REPRESENTATIONS

The lecturer, Ndjimi Malaka, PhD.

3.5 MACHINE TRANSLATION SYSTEMS APPROACHES

The lecturer, Ndjimi Malaka, PhD.

 TRUE/FALSE
1. THE DTA WAS DESIGNED TO TRANSLATE BETWEEN VERY DIFFERENT LANGUAGES, SUCH
AS ENGLISH – CHINESE.
2. THE SOURCE-LANGUAGE TEXT IS TREATED AS A STRING OF DISCONNECTED WORDS.
3. THE DTA USES BILINGUAL DICTIONARIES TO TRANSLATE TEXTS.
4. THE DTA IS A PHRASE-FOR-PHRASE TRANSLATION SYSTEM.
5. THE DTA IS SIMPLE AND REQUIRES A SHORT PROCESSING TIME.

The lecturer, Ndjimi Malaka, PhD.

 3.5.1 DIRECT TRANSLATION APPROACH

• DIRECT TRANSLATION WAS THE FIRST APPROACH EMPLOYED IN MACHINE TRANSLATION

DEVELOPMENT AND IS CONSIDERED TO BE THE FIRST GENERATION OF MACHINE
TRANSLATION SYSTEMS.
• DESIGNED TO TRANSLATE BETWEEN TWO CLOSELY RELATED LANGUAGES.
• THE SOURCE-LANGUAGE TEXT IS TREATED AS
A STRING OF WORDS, AND A NUMBER OF OPERATIONS ARE PERFORMED BY REPLACING
SOURCE-LANGUAGE WORDS WITH TARGET-LANGUAGE WORDS, RE-ORDERING WORDS
UNTIL THEY END UP WITH A STRING OF SYMBOLS IN THE TARGET LANGUAGE

The lecturer, Ndjimi Malaka, PhD.

A DIRECT SYSTEM IS ESSENTIALLY A DICTIONARY-BASED
SYSTEM THAT MATCHES EACH SOURCE-LANGUAGE WORD
TO ITS TARGET-LANGUAGE EQUIVALENT.
THE TRANSLATION TASK IS A SINGLE PROCESSING
OPERATION THAT STORES ALL DATA IN ONE BILINGUAL
DICTIONARY WITH NO SEPARATE GRAMMAR MODULE.
WORD-FOR-WORD TRANSLATION.

The lecturer, Ndjimi Malaka, PhD.

 CONS OF THE DTA
PRIMITIVE AND REQUIRED A LONG PROCESSING TIME.
NO LINGUISTIC ANALYSIS WAS CARRIED OUT ON THE SOURCE LANGUAGE TEXT BEFORE ITS
TRANSLATION WAS GENERATED.
DOES NOT HAVE THE CAPABILITY TO RESOLVE AMBIGUITIES, TO DEAL WITH
METAPHORICAL EXPRESSIONS OR TO TRANSLATE SENTENCES BETWEEN UNRELATED LANGUAGE
PAIRS.
THIS APPROACH WAS SIMPLE AND CHEAP BUT THE OUTPUT RESULTS WERE POOR AND MIMIC –
FOR OBVIOUS REASONS – THE SYNTACTIC STRUCTURES OF THE SOURCE LANGUAGE.
THE SYSTEM MODEL WAS DESIGNED AND BUILT BY MATHEMATICIANS AND
ENGINEERS.
UNRELIABLE AND INSUFFICIENTLY POWERFUL.
The lecturer, Ndjimi Malaka, PhD.


 3.5.1.1 MODELS OF THE DTA

Early direct machine translation systems include Météo, Weidner, CULT

and the old Systran. The current Systran, originally designed to
translate from Russian into English, can now accommodate a larger
number of language pairs.

The lecturer, Ndjimi Malaka, PhD.

 DIRECT TRANSLATION MODEL

The lecturer, Ndjimi Malaka, PhD.

 REFLECTION
• WHAT IS THE MAIN IDEA BEHIND THE DTA?

The lecturer, Ndjimi Malaka, PhD.

 3.5.2. RULE-BASED APPROACHES

The lecturer, Ndjimi Malaka, PhD.

• INVOLVE THE APPLICATION OF MORPHOLOGICAL,
SYNTACTIC AND/OR SEMANTIC RULES TO THE ANALYSIS
OF A SOURCE-LANGUAGE TEXT AND SYNTHESIS OF A
TARGET-LANGUAGE TEXT.

The lecturer, Ndjimi Malaka, PhD.

 3.5.2.1 THE INTERLINGUA APPROACH

IN THE INTERLINGUA APPROACH, A SOURCE-LANGUAGE TEXT IS CONVERTED INTO A

HIGHLY ABSTRACT REPRESENTATION THAT CAPTURES ALL THE ESSENTIAL SYNTACTIC AND
SEMANTIC INFORMATION THAT CAN THEN BE CONVERTED INTO SEVERAL TARGET
LANGUAGES.

AN ‘INTERLINGUA’ REPRESENTS ‘ALL SENTENCES THAT MEAN THE “SAME” THING IN THE
SAME WAY, REGARDLESS OF THE LANGUAGE THEY HAPPEN TO BE IN’.

IT IS DESIGNED TO BE LANGUAGE-INDEPENDENT.

The lecturer, Ndjimi Malaka, PhD.
• AN INTERLINGUA IS INTENDED TO FUNCTION IN STAGES AS THE INTERMEDIARY
BETWEEN NATURAL LANGUAGES.
THE PROCESS:
1. DURING THE ANALYSIS STAGE, A SOURCE-LANGUAGE TEXT IS ANALYSED AND
TRANSFORMED INTO ITS INTERLINGUA REPRESENTATION.

2. TARGET LANGUAGE SENTENCES ARE PRODUCED FROM THIS INTERLINGUA

REPRESENTATION WITH THE HELP OF TARGET-LANGUAGE DICTIONARIES AND
GRAMMAR RULES DURING THE SYNTHESIS STAGE.

3. INTERLINGUA SYSTEMS ARE HIGHLY MODULAR IN THE SENSE THAT ONE PART OF
THE SYSTEM DOES NOT AFFECT OTHER PARTS.

FOR EXAMPLE, IN A DUTCH TO RUSSIAN MACHINE TRANSLATION SYSTEM, IF THE

DUTCH PARSER IS BEING UPGRADED IT DOES NOT AFFECT THE RUSSIAN SENTENCE
GENERATOR.

The lecturer, Ndjimi Malaka, PhD.

 INTERLINGUA MODELS

The lecturer, Ndjimi Malaka, PhD.

The lecturer, Ndjimi Malaka, PhD.
 This illustrates a multilingual system using the interlingua approach,
which started with Dutch as the source language and Russian as the
target language. With modularity, it is possible to add three other
source languages (French, Italian and Russian) and generate three
other target languages (Dutch, Italian and French). In all, a total of 12
combinations of language pairs (from Dutch into Russian, French and
Italian; from French into Dutch, Russian and Italian; from Italian into
Russian, Dutch and French; and from Russian into Dutch, French and
Italian) from four languages can be produced by such a system.

The lecturer, Ndjimi Malaka, PhD.

COMPREHENSION
A. WHAT IS AN INTERLINGUA?
B. HOW FAR DO YOU AGREE WITH THE FOLLOWING STATEMENTS?

1. IN THE INTERLINGUA APPROACH, A SOURCE-LANGUAGE TEXT IS

TRANSFORMED INTO A NEUTRAL LANGUAGE BEFORE BEING
TRANSLATED INTO ANY LANGUAGE.
2. AN ‘INTERLINGUA’ IS COMPOSED OF SIMILAR MEANINGS IN ANY
LANGUAGE.
3. IT IS LANGUAGE-BOUND.

The lecturer, Ndjimi Malaka, PhD.

COMPLETE THE PROCESS:
1. ANALYSIS STAGE, A SOURCE-LANGUAGE TEXT IS ANALYSED AND…

2. TARGET LANGUAGE SENTENCES ARE PRODUCED FROM THIS INTERLINGUA

REPRESENTATION WITH THE HELP OF…

3. INTERLINGUA SYSTEMS ARE HIGHLY MODULAR IN THE SENSE THAT ONE

PART OF THE SYSTEM DOES NOT AFFECT…

The lecturer, Ndjimi Malaka, PhD.

 3.5.2.2 THE TRANSFER APPROACH

• THE TRANSFER APPROACH IS LESS AMBITIOUS

THAN THE INTERLINGUA APPROACH, AND
CONSISTS OF THREE STAGES.
•

The lecturer, Ndjimi Malaka, PhD.

• 1. THE ANALYSIS STAGE AIMS TO CONVERT A SOURCE LANGUAGE TEXT INTO
AN ABSTRACT SOURCE-LANGUAGE REPRESENTATION.

• 2. FOLLOWING THIS, THE TRANSFER OF THE SOURCE-LANGUAGE

REPRESENTATION INTO ITS EQUIVALENT TARGET-LANGUAGE
REPRESENTATION TAKES PLACE.

• 3. THE LAST STAGE IS WHERE A TARGET-LANGUAGE TEXT IS GENERATED.

The lecturer, Ndjimi Malaka, PhD.

• SPECIFIC DICTIONARIES ARE USED AT EACH STAGE: A
SOURCE-LANGUAGE DICTIONARY AT THE ANALYSIS STAGE,
A BILINGUAL DICTIONARY AT THE TRANSFER STAGE, AND A
TARGET LANGUAGE DICTIONARY AT THE GENERATION
STAGE AS ILLUSTRATED IN THE FIGURE.

The lecturer, Ndjimi Malaka, PhD.

The lecturer, Ndjimi Malaka, PhD.
 This approach uses contrastive knowledge of
the two languages. As an example, Figure x
shows the transfer stage where the source-
language representation of the English
phrase ‘the beautiful little girl’ undergoes a
parsing process to restructure the English
phrase into its Spanish translation, ‘la
pequeña muchacha hermosa’.

The lecturer, Ndjimi Malaka, PhD.

 TRANSFER USING TREE-TO-TREE PARSING

The lecturer, Ndjimi Malaka, PhD.

Like the other rule-based approach, the transfer approach is suitable
for building a multilingual machine translation system. However, unlike
the interlingua approach where only one interlingua is responsible for all
the language pairs, the transfer approach uses different transfer models
for each language pair.

Figure y shows an example of a transfer-based multilingual machine

translation system of three languages able to generate six language
pairs (from Polish into Hungarian and Romanian; from Hungarian into
Polish and Romanian; and from Romanian into Polish and Hungarian).

The lecturer, Ndjimi Malaka, PhD.

The lecturer, Ndjimi Malaka, PhD.
• CONS:

1. IT RELIES ON DICTIONARIES, WHICH MAY NOT

NECESSARILY CONTAIN SUFFICIENT KNOWLEDGE TO
RESOLVE AMBIGUITIES.
2. MOREOVER, FAILURE AT THE ANALYSIS STAGE MAY
RESULT IN ZERO OUTPUT BECAUSE THE TRANSFER
PROCESS CANNOT TAKE PLACE.

The lecturer, Ndjimi Malaka, PhD.

 CHECK IT!

1. IS THE TRANSFER APPROACH LESS AMBITIOUS THAN THE INTERLINGUA

APPROACH? WHY? WHY NOT?
2. WHAT HAPPENS AT THE THREE STAGES OF THE TA?
3. WHAT SPECIFIC DICTIONARIES ARE USED AT EACH STAGE?

The lecturer, Ndjimi Malaka, PhD.

 3.5.3. CORPUS-BASED APPROACHES

Make use of linguistic information in a

corpus to create new translations. All corpus-based
machine translation systems use a set of so-called
‘reference translations’ containing source language
texts and their translations. Source and target-
language texts are aligned and the equivalent
translation is extracted using a specific statistical
method or by matching a number of examples
extracted from the corpus.

The lecturer, Ndjimi Malaka, PhD.

 Corpus-based approaches
The statistical-based
approach

The example-based
approach

The lecturer, Ndjimi Malaka, PhD.

 3.5.3.1 THE STATISTICAL-BASED APPROACH

• A SOURCE-LANGUAGE TEXT IS FIRST SEGMENTED INTO STRINGS OF WORDS AND PHRASES;

• THE SOURCE-LANGUAGE SEGMENTS ARE THEN COMPARED TO AN EXISTING LARGE
ALIGNED BILINGUAL CORPUS CONSISTING OF ORIGINAL TEXTS AND THEIR TRANSLATIONS,
AND A STATISTICAL METHOD IS THEN EMPLOYED ON THE ALIGNED BILINGUAL CORPUS TO
OBTAIN NEW TARGET-LANGUAGE SEGMENTS.
• FROM THE NEW SEGMENTS, USING THE THEOREM, A NEW TARGET LANGUAGE TEXT IS
PRODUCED.

The lecturer, Ndjimi Malaka, PhD.

• THIS APPROACH IS CLEARLY QUITE DIFFERENT FROM THE RULE-BASED APPROACHES
PREVIOUSLY DISCUSSED THAT EMPLOY PRE-DETERMINED LINGUISTIC RULES TO ANALYSE
SOURCE-LANGUAGE TEXTS IN ORDER TO GENERATE TRANSLATIONS

The lecturer, Ndjimi Malaka, PhD.

• THE PRINCIPAL HYPOTHESIS OF THIS APPROACH IS THAT ONE SOURCE-LANGUAGE
SENTENCE (S) CAN HAVE A LARGE NUMBER OF TRANSLATIONS (T), AND EACH OF THESE HAS A
VARYING PROBABILITY (P) OF BEING CORRECT. THE PROBABILITY IS CALCULATED USING BAYES’
RULE, WHICH STATES THAT:
•

Where P(T⏐S) is the probability of T given the translation S; P(T) is the

probability of randomly selecting the text T, which is calculated from
the frequency in the corpus; P(S⏐T) is the probability assumed by the
translation model used by the algorithm assigned to S being translated
into T; and P(S) is the frequency of observing the text S in the corpus.

The lecturer, Ndjimi Malaka, PhD.

 STATISTICAL-BASED MODEL

The lecturer, Ndjimi Malaka, PhD.

 3.5.3.2 THE EXAMPLE-BASED APPROACH

The lecturer, Ndjimi Malaka, PhD.

• EXAMPLE-BASED MACHINE TRANSLATION IS ALSO REFERRED TO AS
ANALOGY-, MEMORY-, PATTERN-, CASE- OR SIMILARITY-BASED
TRANSLATION.
• AN EXAMPLE-BASED MACHINE TRANSLATION REQUIRES A
BILINGUAL CORPUS OF TRANSLATION PAIRS AND EMPLOYS AN
ALGORITHM TO MATCH THE CLOSEST EXAMPLE OF A SOURCE-
LANGUAGE SEGMENT TO ITS TARGET-LANGUAGE
SEGMENT AS THE BASIS FOR TRANSLATING THE NEW SOURCE TEXT.
• A MATCHED PAIR OF SEGMENTS IS CALLED AN ‘EXAMPLE’.

The lecturer, Ndjimi Malaka, PhD.

• A SEGMENT CAN BE OF ANY LENGTH OR OPERATE AT ANY
LINGUISTIC LEVEL, BUT, IDEALLY, IT SHOULD BE AT THE SENTENCE
LEVEL.
• THREE MAIN TASKS ARE INVOLVED IN THE TRANSLATION PROCESS
OF AN EXAMPLE BASED SYSTEM: MATCHING SEGMENTS FROM THE
NEW SOURCE
TEXT AGAINST EXISTING PAIRS OF EXAMPLES EXTRACTED FROM AN
ALIGNED BILINGUAL CORPUS, THEN ALIGNING CORRESPONDING
TRANSLATION SEGMENTS AND RECOMBINING THEM TO GENERATE
A TARGET TEXT.
The lecturer, Ndjimi Malaka, PhD.
 EXAMPLE-BASED MODEL

The lecturer, Ndjimi Malaka, PhD.

• THE BASIC IDEA OF AN EXAMPLE BASED TRANSLATION IS TO ‘TRANSLATE A SOURCE
SENTENCE BY IMITATING THE TRANSLATION OF A SIMILAR SENTENCE ALREADY IN THE
DATABASE’. HOWEVER, IN MOST CASES, MORE THAN ONE ‘IMITATION’ MAY BE NEEDED TO
TRANSLATE A COMPLETELY NEW SOURCE-LANGUAGE SENTENCE. THE FOLLOWING TABLE
ILLUSTRATES THIS.

The lecturer, Ndjimi Malaka, PhD.

The lecturer, Ndjimi Malaka, PhD.
 When an English sentence such as ‘She sells flowers in the farmers’
market every day’ (ES3.3 in Table 3.2) needs to be translated into Malay,
the database of examples is searched for similar but not necessarily
identical source-language strings such as ‘The lady in the farmers’
market is my cousin’ (ES3.1) and ‘She sells flowers every day’ (ES3.2).
Along with source sentences such as these, their Malay translations are
also extracted (see MS3.1 and MS3.2). By referring to these examples, the
new source-language sentence may then be translated by ‘imitating’ the
matching parts of the Malay examples and recombining them to turn
them into new target-language sentence

The lecturer, Ndjimi Malaka, PhD.

 DISCUSSION
1. WHY ARE THEY CALLED CORPUS-BASED APPROACHES?
2. IS STATISTICS A VIABLE LANGUAGE MEASURE?
3. WHAT SHORCOMINGS CAN YOU THINK OF AN EXAMPLE
BASED-APPROACH?

The lecturer, Ndjimi Malaka, PhD.

 3.6 HYBRID AND INTERACTIVE MACHINE TRANSLATION
SYSTEMS
A rule-based system is deductive in nature as it is based on a set of linguistic
rules set up by its designers.

Moreover, it does not in principle store any translation results or reuse

previously translated segments.
A corpus-based system on the other hand, is inductive in nature because the rules are
derived from a given set of translation examples and modification is achieved
through the addition of new translation examples.

The rule-based approach is often expensive, and may produce inconsistent results
when new linguistic rules are added.
The lecturer, Ndjimi Malaka, PhD.
In contrast, the corpus-based approach is flexible enough to
process sentences even if they are ill formed. However, when
long sentences are involved, the processing time tends to be
lengthy.

Machine translation research is unlikely to progress significantly by

the refinement of one approach in preference to another. Instead,
‘hybrid’ and other innovative approaches may be the best way
forward.
The lecturer, Ndjimi Malaka, PhD.
 3.6.1 HYBRIDIZATIONS COMBINING TWO DIFFERENT
MACHINE TRANSLATION SYSTEMS

The lecturer, Ndjimi Malaka, PhD.

1. CAT2 (A RULE-BASED) MACHINE TRANSLATION SYSTEM.
2. EXAMPLE-BASED EDGAR MACHINE TRANSLATION SYSTEM.
3. PANGLOSS MARK II AN EXAMPLE-BASED ENGINE AND AN
INTERLINGUA APPROACH IN ONE SINGLE SYSTEM.
4. OTELO, WHICH IS MADE UP OF TWO MACHINE TRANSLATION
SYSTEMS – IBM LMT AND LOGOS – WITH A TRANSLATION
MEMORY SYSTEM CALLED IBM TRANSLATION MANAGER.

The lecturer, Ndjimi Malaka, PhD.

 3.7 INTERACTIVE MACHINE TRANSLATION SYSTEMS

• INVOLVE HUMANS IN THE PROCESS.

• EXAMPLES OF INTERACTIVE MACHINE TRANSLATION SYSTEMS INCLUDE LINGSTAT,
TRANSTYPE2 AND WEBDIPLOMAT (WEB DISTRIBUTED INTELLIGENT PROCESSING OF
LANGUAGE FOR OPERATIONAL MACHINE AIDED TRANSLATION) FROM CARNEGIE MELLON
UNIVERSITY, AND DBMT (DIALOGUE-BASED MACHINE TRANSLATION SYSTEM) FROM THE
UNIVERSITY OF GRENOBLE.

The lecturer, Ndjimi Malaka, PhD.

 3.8 ONLINE MACHINE TRANSLATION SYSTEMS

Machine translation was

originally designed for The Developers of machine
formal written language, translation systems have
but the demand for the created different systems to
translation of e-mail meet the needs of different
messages that contain less end-users, for instance, client–
formal syntactic structures server machine translation
has increased in recent systems.
years.

The lecturer, Ndjimi Malaka, PhD.

CLIENT–SERVER SITUATION:
1. AN ORGANIZATION OR A TRANSLATION COMPANY HOSTS THE MACHINE
TRANSLATION SYSTEM ON A SERVER.
2. THE ‘CLIENT’, EITHER A COMPANY OR AN INDIVIDUAL, CAN ACCESS THE MACHINE
TRANSLATION SYSTEM USING USER IDENTIFICATION AND A PASSWORD.

The lecturer, Ndjimi Malaka, PhD.

 EXAMPLES OF OMTS
1. EUROPEAN COMMISSION SYSTRAN MACHINE TRANSLATION SYSTEM - OFFERS TRANSLATION
SERVICES TO TRANSLATORS, ADMINISTRATORS AND THE OFFICIALS OF EUROPEAN UNION
INSTITUTIONS AND BODIES. THE SYSTEM HAS 18 LANGUAGE PAIRS WITH ENGLISH, FRENCH,
GERMAN AND SPANISH ACTING AS THE MAIN SOURCE LANGUAGES.

The lecturer, Ndjimi Malaka, PhD.

2. BABELFISH

The lecturer, Ndjimi Malaka, PhD.

• PROMT

The lecturer, Ndjimi Malaka, PhD.

WORLDLINGO

The lecturer, Ndjimi Malaka, PhD.

• THE GROWING NUMBER OF ONLINE MACHINE TRANSLATION SYSTEMS AND IMPROVEMENTS IN
THEIR PERFORMANCE MEAN THAT MILLIONS OF INTERNET USERS CAN NOW READ PAGES
ORIGINALLY POSTED IN LANGUAGES THAT THEY DO NOT UNDERSTAND.
• FOR END-USERS WHO OCCASIONALLY REQUIRE THE TRANSLATION OF WEB PAGES, SMALL
CHUNKS OF TEXT AND E-MAILS, FREE ONLINE MACHINE TRANSLATION SYSTEMS SUCH AS
BABELFISH, PROMT-ONLINE AND WORLDLINGO ARE EXTREMELY HELPFUL.
• SINCE MOST ONLINE MACHINE TRANSLATION SYSTEMS ARE FOR GENERAL PURPOSES, THE
TRANSLATIONS GENERATED MAY NOT BE OF THE HIGHEST QUALITY BUT THERE IS ENOUGH
INFORMATION FOR THE END-USERS TO AT LEAST UNDERSTAND THE OVERALL CONTENT.

The lecturer, Ndjimi Malaka, PhD.

3.9 COMMERCIAL MACHINE TRANSLATION SYSTEMS
Much of the previous discussion has been about the development of
machine translation systems from a research perspective. In this section,
the focus is on commercial machine translation systems.

• THE FOLLOWING TABLE PRESENTS AN INDICATIVE SUMMARY OF THE DIFFERENT TYPES OF

COMMERCIAL MACHINE TRANSLATION SYSTEMS CURRENTLY AVAILABLE. THE CLASSIFICATION IS
BASED ON THE INTENDED TARGET MARKET, FOR EXAMPLE HOME VERSUS PROFESSIONAL, AND
THE NUMBER OF LANGUAGES INVOLVED.

The lecturer, Ndjimi Malaka, PhD.

 Table v. Classification of commercial machine translation systems

The lecturer, Ndjimi Malaka, PhD.

 THE LANGUAGES AND DIRECTION OF TRANSLATION IN MACHINE TRANSLATION SYSTEMS ARE
USUALLY PRE-DETERMINED.
 THE DIRECTION OF TRANSLATION FOR THE TRILINGUAL CLASS VARIES FROM SYSTEM TO SYSTEM
AND IT IS NOT POSSIBLE TO LIST EACH ONE.
 SIMILARLY, THE DIRECTIONS ARE NOT LISTED FOR THE MULTILINGUAL SYSTEMS, WHICH INVOLVE
MORE THAN THREE LANGUAGES.
 ALMOST ALL COMERCIAL SYSTEMS ARE SAID TO HAVE RULE-BASED RATHER THAN CORPUS-BASED
ARCHITECTURES.
 CORPUS-BASED SYSTEMS ARE NOW BEING COMMERCIALLY DEVELOPED AND COULD MAKE THEIR
APPEARANCE IN THE MARKET IN THE FUTURE.

The lecturer, Ndjimi Malaka, PhD.

 THE TYPE LABELLED AS ‘PROFESSIONAL’ IN THE TABLE CATERS FOR INDIVIDUAL PROFESSIONAL
TRANSLATORS, WHILE ‘CLIENT–SERVER’ IS PRIMARILY INTENDED FOR EXTERNAL CONTRACTOR
COMPANIES OR TRANSLATION COMPANIES IN ORDER TO SUPPORT THEIR TEAMS OF
PROFESSIONAL TRANSLATORS EITHER IN-HOUSE OR WORKING AS ‘SUPPLIERS’ OF TRANSLATION
SERVICES ON A FREELANCE BASIS.
 THE TYPE LABELLED ‘PORTAL’ REFERS TO MACHINE TRANSLATION COMPANIES THAT SUPPLY
MACHINE TRANSLATION SYSTEMS TO OTHER COMPANIES OR WEBSITES.
 THE TYPE LABELLED ‘MOBILE’ MAY REFER TO PLUG-IN SYSTEMS; FOR EXAMPLE, A PALM PILOT
PLUGGED INTO A NETWORKED COMPUTER USING EITHER A CABLE OR WIRELESS MODEM TO
REMOTELY ACCESS A MACHINE TRANSLATION SYSTEM.

The lecturer, Ndjimi Malaka, PhD.

 CMTS

The lecturer, Ndjimi Malaka, PhD.

The lecturer, Ndjimi Malaka, PhD.
 ONE OF THE MAIN COMMERCIAL FUNCTIONS OF MACHINE TRANSLATION IS TO IMPROVE THE
PRODUCTIVITY OF HUMAN TRANSLATORS WHO ARE ABLE TO CHECK AND POLISH TRANSLATIONS.
 ALTHOUGH A MACHINE TRANSLATION SYSTEM IS UNLIKELY TO BE AS ACCURATE AS A
PROFESSIONAL TRANSLATOR, IT IS FASTER AND CHEAPER.
 THE SUCCESSFUL USE OF ANY COMMERCIAL MACHINE TRANSLATION SYSTEM REQUIRES A LARGE
INVESTMENT ON THE PART OF THE END-USER SUCH AS BEING COMMITTED TO CREATE, MAINTAIN
AND UPDATE TERMINOLOGY AND TRANSLATION DATABASES IN ORDER TO MAXIMIZE THE
USABILITY OF A SYSTEM.

The lecturer, Ndjimi Malaka, PhD.

3.10 REASONS FOR USING MACHINE TRANSLATION
SYSTEMS
1. DISSEMINATION,
2. ASSIMILATION,
3. INFORMATION EXCHANGE,
4. ACCESS.

The lecturer, Ndjimi Malaka, PhD.

 4. HUMAN-AIDED MACHINE TRANSLATION

• WHAT IS HAMT?

The lecturer, Ndjimi Malaka, PhD.

 A system wherein the computer is
responsible for producing the
translation per se, but may interact
with a human monitor at many
stages along the way.

The lecturer, Ndjimi Malaka, PhD.

THE MACHINE CARRIES OUT MOST OF THE WORK BUT IT MIGHT NEED HUMAN
ASSISTANCE EITHER AT THE TEXT-PREPARATION STAGE OR THE OUTPUT STAGE.
THE FORMER PROCESS IS KNOWN AS ‘PRE-EDITING’ (TO DISCOVER ANY ELEMENTS
SUCH AS ODD PHRASES OR IDIOMS AND TYPOGRAPHICAL ERRORS THAT MAY
CREATE PROBLEMS FOR THE MACHINE TRANSLATION SYSTEM DURING THE
TRANSLATION PROCESS.
AND THE LATTER ‘POST-EDITING’ - INVOLVES CORRECTING THE TRANSLATION
OUTPUT GENERATED BY THE MACHINE TRANSLATION SYSTEM, A TASK
PERFORMED BY THE HUMAN EDITOR OR TRANSLATOR IN ORDER TO BRING THE
TEXT TO A CERTAIN PRE-DETERMINED STANDARD IN TERMS OF LANGUAGE STYLE
AND APPROPRIATE USE OF TERMS.

The lecturer, Ndjimi Malaka, PhD.

The lecturer, Ndjimi Malaka, PhD.
Human-aided machine translation model
A pre-edited text is one that has been edited by a human, in most cases by
someone other than the author, prior to the translation process, whereas a
controlled-language text is usually written following certain strict linguistic rules.

Sometimes, a source-language text can also be edited using the controlled-

language vocabulary and linguistic rules.

Ideally, pre-edited and controlled language texts are free from ambiguity and
complex sentences.

Unedited text, on the other hand, has had no editing prior to translation.

For systems that have an interactive mode, a human is allowed to correct or select
appropriate equivalents during the automatic translation process. Otherwise,
corrections can only be performed at the post-editing stage, which is after the
machine translation
The lecturer, Ndjimi Malaka, PhD. system has produced the translation.
 EXAMPLES OF HUMAN-AIDED MACHINE
TRANSLATION SYSTEMS
• MATRA PRO AND LITE
 DEVELOPED AT THE NATIONAL CENTRE FOR SOFTWARE TECHNOLOGY BASED IN MUMBAI,
INDIA, THAT TRANSLATE FROM ENGLISH INTO HINDI.
 HUMAN-AIDED MACHINE TRANSLATION SYSTEMS HAVE BEEN IMPLEMENTED AT SCHREIBER
TRANSLATIONS, INC., FOREIGN LANGUAGE SERVICES, INC. AND RALPH MCELROY
TRANSLATION COMPANY, ALL COMPANIES THAT ARE EMPLOYED TO TRANSLATE PATENTS FOR
THE UNITED STATES (US) PATENT AND TRADEMARK OFFICE.

The lecturer, Ndjimi Malaka, PhD.

 CHECK IT!
COMPLETE THE STATEMENTS
1. PRE-EDITING IS TO DISCOVER ANY ELEMENTS SUCH AS…
2. POST-EDITING INVOLVES CORRECTING THE TRANSLATION…
3. THE FUNCTION OF THE HUMAN COMPONENT IS…

The lecturer, Ndjimi Malaka, PhD.

 5. MACHINE-AIDED HUMAN TRANSLATION (MAHT)

Machine-aided human translation has

been described as the use of computer
software by translators to perform part
of the process of translation.

The lecturer, Ndjimi Malaka, PhD.

 Integrated machine-aided human translation systems are
sometimes known as ‘workbenches’ or ‘workstations’, as they
combine a number of tools. The focus in this type of
translation is on the human translator, who uses an assortment
of tools such as spell-checkers, electronic glossaries, electronic
dictionaries, terminology databases and collections of
previously translated texts and their originals, that is
translation ‘memory’, to support the translation process.

The lecturer, Ndjimi Malaka, PhD.

• EXAMPLES OF COMMERCIAL MACHINE-AIDED HUMAN TRANSLATION SYSTEMS

1. TRANSLATOR’S WORKBENCH BY TRADOS GMBH,

2. TRANSIT BY STAR AG,
3. SDLX TRANSLATION SUITE BY SDL INTERNATIONAL
AND DÉJÀ VU BY ATRIL.

The lecturer, Ndjimi Malaka, PhD.

 MACHINE-AIDED HUMAN TRANSLATION MODEL

The lecturer, Ndjimi Malaka, PhD.

 6. COMPUTER-AIDED TRANSLATION TOOLS
AND RESOURCES (CATTR)
• TRUE/FALSE
1. A ‘WORKBENCH’ OR A ‘WORKSTATION’ IS A SINGLE INTEGRATED SYSTEM THAT IS MADE
UP OF A NUMBER OF TRANSLATION TOOLS AND RESOURCES.
2. A MULTILINGUAL TEXT ARCHIVE CONTAINS SEGMENTED, ALIGNED, PARSED AND
DECLASSIFIED MULTILINGUAL TEXTS.
3. THE DIFFERENCE BETWEEN A MACHINE TRANSLATION SYSTEMS AND TRANSLATION
MEMORY SYSTEMS IS THAT MTS GENERATES TRANSLATIONS AUTOMATICALLY; WHEREAS,
TMS ALLOW PROFESSIONAL TRANSLATORS TO BE IN CHARGE OF THE DECISION-MAKING
PROCESS.

The lecturer, Ndjimi Malaka, PhD.

 Perfect
matching

Fuzzy matching

The translation
workflow
Workbenches
CATTR Alignment
Terminology
management
systems
Segmentation

Filter

The lecturer, Ndjimi Malaka, PhD.

 Terminology management
Cloud-based
systems TMS solutions
Types of TMS Integrated TMS
solutions
Standalone TMS
software

The lecturer, Ndjimi Malaka, PhD.

 6.1 WORKBENCHES

A ‘workbench’ or a ‘workstation’ is a single integrated system that is

made up of a number of translation tools and resources such as a
translation memory, an alignment tool, a tag filter, electronic
dictionaries, terminology databases, a terminology management
system and spell and grammar-checkers.

The lecturer, Ndjimi Malaka, PhD.

The lecturer, Ndjimi Malaka, PhD.
 6.1.1 TRANSLATION MEMORY SYSTEMS

A multilingual text archive contains segmented, aligned, parsed and classified

multilingual texts, allowing storage and retrieval of aligned multilingual text
segments against various search conditions

The lecturer, Ndjimi Malaka, PhD.

 Unlike machine translation systems, which generate translations
automatically, translation memory systems allow professional translators
to be in charge of the decision-making whether to accept or reject a term
or an equivalent phrase or ‘segment’ suggested by the system during the
translation process. Translators can also build their own ‘memory’.

The lecturer, Ndjimi Malaka, PhD.

 Virtually all translation memory systems are language-
independent and support international character sets that
represent many, if not all, alphabets and scripts digitally.

The lecturer, Ndjimi Malaka, PhD.

• TRANSLATION MEMORY TECHNOLOGY WORKS BY REUSING PREVIOUSLY TRANSLATED
TEXTS AND THEIR ORIGINALS IN ORDER TO FACILITATE THE PRODUCTION OF NEW
TRANSLATIONS.
• IT CAN ALSO INTERFACE WITH DATABASES OF STORED SPECIALIZED TERMINOLOGIES THAT
CAN BE ACCESSED AND RETRIEVED FOR REUSE IN NEW TRANSLATIONS.

The lecturer, Ndjimi Malaka, PhD.

 CHARACTERISTICS
A perfect or exact match
A translation memory
occurs when a new source
system has no linguistic
Perfect matching language segment is
component, and two
completely identical
different approaches
including spelling,
are employed to
punctuation
extract translation
and inflections, to the old
segments from
segment found in the
the previously stored
database, that is in the
texts. These are known
translation ‘memory’
as perfect matching
and fuzzy Occurs when an old and a new
matching. Other source-language segment are similar
characteristics such as but not exactly identical. Even a very
filter, segmentation Fuzzy matching small difference such as punctuation
and alignment leads to a fuzzy match.
will also be discussed

The lecturer, Ndjimi Malaka, PhD.

 A filter is a feature that converts a source
language text from one format into
Filter another giving a translator the
flexibility to work with texts of different
formats

is the process of breaking a text up into

units consisting of a word or a string of
words that is linguistically acceptable.
Segmentation
Segmentation is needed in order for a
translation memory to perform the matching
(perfect and fuzzy) process.

The lecturer, Ndjimi Malaka, PhD.

 is the process of binding a source-language segment
to its corresponding target-language segment. The
purpose of alignment is to create a new translation
Alignment memory database or to add to an existing one. The
corresponding pairs of source and target-language
segments are called ‘translation units’ (see Table).

The lecturer, Ndjimi Malaka, PhD.

The lecturer, Ndjimi Malaka, PhD.
 Suppose that an English source text
needs to be translated into French. The English text is
then compared to a database of previously translated
English–French texts to find out if any of the segments
The translation in the new English source text matches the segments of
old or stored English source texts in the database.
workflow

At this stage, identical or similar segments in

English and French are identified and extracted
by the translation memory system.

The lecturer, Ndjimi Malaka, PhD.

The lecturer, Ndjimi Malaka, PhD.
 6.2 TERMINOLOGY MANAGEMENT SYSTEMS

For professional translators who specialize in highly technical subject

fields, terminology is a crucial component of their translation work.
A terminology – that is a codified collection of terms – can be defined as ‘a
systematic arrangement of concepts within a special language. Concepts, not terms.
Systematic, not alphabetic’.

Terminology is arranged by concept. Each concept has a label – or set of labels if synonymous
– called a ‘term’, which is a single word or a string of words used to represent it in the
language of the specialized field.
The lecturer, Ndjimi Malaka, PhD.
• CONCEPTS ARE ARRANGED ‘SYSTEMATICALLY’ TO REFLECT THE ORGANIZATION OF
KNOWLEDGE IN A PARTICULAR SUBJECT FIELD, FOR EXAMPLE TO EXHIBIT A HIERARCHICAL
RELATIONSHIP OF SCIENTIFIC CLASSIFICATION OR TAXONOMY.

The lecturer, Ndjimi Malaka, PhD.

 6.2.1 TYPES OF TERMINOLOGY MANAGEMENT
SYSTEMS

• STANDALONE TMS SOFTWARE REFERS TO SYSTEMS THAT ARE INSTALLED LOCALLY ON

INDIVIDUAL COMPUTERS OR SERVERS, REQUIRING USERS TO DOWNLOAD AND MAINTAIN
SOFTWARE LICENSES AND DATABASES. WHILE STANDALONE TMS OFFER GREATER CONTROL
AND CUSTOMIZATION OPTIONS, THEY MAY ALSO INVOLVE HIGHER UPFRONT COSTS AND
TECHNICAL COMPLEXITIES.
• INTEGRATED TMS SOLUTIONS ARE SEAMLESSLY INTEGRATED WITHIN TRANSLATION
SOFTWARE OR CONTENT MANAGEMENT SYSTEMS, PROVIDING A UNIFIED PLATFORM FOR
MANAGING BOTH TRANSLATION WORKFLOWS AND TERMINOLOGY.

The lecturer, Ndjimi Malaka, PhD.

• CLOUD-BASED TMS SOLUTIONS ARE HOSTED ON REMOTE SERVERS AND ACCESSED VIA THE
INTERNET, OFFERING GREATER SCALABILITY, ACCESSIBILITY, AND COLLABORATION CAPABILITIES
COMPARED TO TRADITIONAL STANDALONE TMS. CLOUD-BASED TMS ELIMINATE THE NEED FOR
LOCAL INSTALLATION AND MAINTENANCE, MAKING THEM IDEAL FOR DISTRIBUTED TEAMS OR
ORGANIZATIONS WITH DIVERSE TRANSLATION NEEDS. THEY ALSO OFFER REAL-TIME UPDATES
AND SYNCHRONIZATION, ENSURING THAT USERS HAVE ACCESS TO THE LATEST
TERMINOLOGICAL DATA AT ALL TIMES.

The lecturer, Ndjimi Malaka, PhD.

 6.2.2 KEY FEATURES AND FUNCTIONS OF
TERMINOLOGY MANAGEMENT SYSTEMS

• ONE OF THE PRIMARY FUNCTIONS OF A TERMINOLOGY MANAGEMENT SYSTEM IS

TO FACILITATE THE CREATION AND ORGANIZATION OF A CENTRALIZED TERMINOLOGY
DATABASE. THIS DATABASE SERVES AS A REPOSITORY FOR STORING TERMINOLOGICAL DATA,
INCLUDING TERMS, DEFINITIONS, TRANSLATIONS, USAGE GUIDELINES, AND CONTEXTUAL
INFORMATION. BY CENTRALIZING TERMINOLOGY IN A STRUCTURED DATABASE, TMS PROVIDE
USERS WITH EASY ACCESS TO RELEVANT TERMINOLOGICAL DATA AND ENSURE CONSISTENCY
IN LANGUAGE USAGE ACROSS ALL CONTENT.

The lecturer, Ndjimi Malaka, PhD.

• TERMINOLOGY MANAGEMENT SYSTEMS ALSO OFFER TOOLS FOR TERMINOLOGY
EXTRACTION AND ALIGNMENT, ALLOWING USERS TO IDENTIFY AND EXTRACT KEY TERMS
FROM SOURCE DOCUMENTS AND ALIGN THEM WITH CORRESPONDING TRANSLATIONS. THIS
PROCESS HELPS ENSURE THAT TRANSLATIONS ARE CONSISTENT WITH SOURCE TERMINOLOGY
AND FACILITATES THE CREATION OF BILINGUAL OR MULTILINGUAL TERMINOLOGICAL
DATABASES.
• ANOTHER KEY FEATURE OF TMS IS TERMINOLOGY STANDARDIZATION AND CONTROL. TMS
PROVIDE TOOLS FOR ESTABLISHING STANDARDIZED DEFINITIONS, TRANSLATIONS, AND USAGE
GUIDELINES FOR KEY TERMS, ENSURING COHERENCE AND CONSISTENCY IN LANGUAGE. BY
ENFORCING STANDARDIZED TERMINOLOGY, TMS HELP MAINTAIN BRAND CONSISTENCY,
IMPROVE TRANSLATION QUALITY, AND ENHANCE COMMUNICATION AMONG TEAMS AND
STAKEHOLDERS.

The lecturer, Ndjimi Malaka, PhD.

• INTEGRATION WITH TRANSLATION WORKFLOWS IS ANOTHER IMPORTANT FUNCTION OF
TERMINOLOGY MANAGEMENT SYSTEMS. MANY TMS OFFER SEAMLESS INTEGRATION WITH
TRANSLATION SOFTWARE OR CONTENT MANAGEMENT SYSTEMS, PROVIDING A UNIFIED PLATFORM
FOR MANAGING BOTH TRANSLATION WORKFLOWS AND TERMINOLOGY. THIS INTEGRATION
STREAMLINES THE TRANSLATION PROCESS, IMPROVES EFFICIENCY, AND FACILITATES
COLLABORATION AMONG TRANSLATORS AND TEAMS.
• FURTHERMORE, SOME TMS OFFER ADVANCED FEATURES SUCH AS REPORTING AND ANALYTICS
CAPABILITIES, ALLOWING USERS TO TRACK AND ANALYZE TERMINOLOGICAL DATA, USAGE
PATTERNS, AND TRANSLATION QUALITY METRICS. THESE FEATURES PROVIDE VALUABLE INSIGHTS
INTO TERMINOLOGY USAGE AND HELP IDENTIFY AREAS FOR IMPROVEMENT IN LANGUAGE-RELATED
PROJECTS.
• TERMINOLOGY MANAGEMENT SYSTEMS OFFER A RANGE OF FEATURES AND FUNCTIONS DESIGNED
TO STREAMLINE THE MANAGEMENT OF TERMINOLOGY AND ENSURE CONSISTENCY AND
ACCURACY IN LANGUAGE-RELATED PROJECTS. BY LEVERAGING THESE KEY FEATURES EFFECTIVELY,
ORGANIZATIONS CAN OPTIMIZE THEIR TERMINOLOGY MANAGEMENT WORKFLOWS AND ACHIEVE
BETTER RESULTS IN THEIR LINGUISTIC ENDEAVORS.
The lecturer, Ndjimi Malaka, PhD.
 6.2.3 HOW TO IMPLEMENT AND UTILIZE
TERMINOLOGY MANAGEMENT SYSTEMS

The first step in implementing a Terminology Management System is

to define clear objectives and goals for terminology management.
This involves identifying key stakeholders, establishing terminology
standards and guidelines, and defining the scope and requirements
of the TMS implementation. By aligning terminology management
efforts with organizational goals and priorities, organizations can
ensure that the TMS implementation is tailored to meet their specific
needs and objectives.

The lecturer, Ndjimi Malaka, PhD.

 Once objectives and goals are defined, the next step is to select the
appropriate TMS solution that aligns with organizational requirements
and preferences. This may involve evaluating different TMS vendors,
assessing features and functionalities, and considering factors such as
scalability, compatibility, and cost-effectiveness. It’s important to choose a
TMS solution that not only meets current needs but also has the flexibility to
adapt to future requirements and growth.

After selecting a TMS solution, the next step is to plan and execute the
implementation process. This may involve tasks such as data migration,
configuration, customization, and integration with existing systems and
workflows. It’s important to involve key stakeholders throughout the
implementation process and provide adequate training and support to
ensure a smooth transition and adoption of the TMS.

The lecturer, Ndjimi Malaka, PhD.

• ONCE THE TMS IS IMPLEMENTED, IT’S ESSENTIAL TO ESTABLISH CLEAR WORKFLOWS AND
PROCESSES FOR UTILIZING THE SYSTEM EFFECTIVELY. THIS INCLUDES DEFINING ROLES AND
RESPONSIBILITIES, ESTABLISHING PROTOCOLS FOR DATA ENTRY AND MANAGEMENT, AND
IMPLEMENTING QUALITY CONTROL MEASURES TO ENSURE THE ACCURACY AND RELIABILITY OF
TERMINOLOGICAL DATA. REGULAR MONITORING AND EVALUATION OF TMS USAGE AND
PERFORMANCE ARE ALSO IMPORTANT TO IDENTIFY AREAS FOR IMPROVEMENT AND
OPTIMIZATION.
• IMPLEMENTING AND UTILIZING A TERMINOLOGY MANAGEMENT SYSTEM REQUIRES CAREFUL
PLANNING, COORDINATION, AND INTEGRATION INTO EXISTING WORKFLOWS. BY
FOLLOWING BEST PRACTICES AND STRATEGIES, ORGANIZATIONS CAN OPTIMIZE THE
IMPLEMENTATION AND UTILIZATION OF TMS TO MAXIMIZE THEIR BENEFITS AND ACHIEVE
GREATER CONSISTENCY, ACCURACY, AND EFFICIENCY IN THEIR LANGUAGE-RELATED PROJECTS.

The lecturer, Ndjimi Malaka, PhD.

 CHECK IT!
• SUMMARISE
1. TYPES OF TERMINOLOGY MANAGEMENT SYSTEMS
2. KEY FEATURES AND FUNCTIONS OF TERMINOLOGY MANAGEMENT SYSTEMS
3. HOW TO IMPLEMENT AND UTILIZE TERMINOLOGY MANAGEMENT SYSTEMS

The lecturer, Ndjimi Malaka, PhD.

1. TYPES
a. STANDALONE TMS SOFTWARE
b. INTEGRATED TMS SOLUTIONS
c. CLOUD-BASED TMS SOLUTIONS
2. KEY FEATURES
a. FACILITATE THE CREATION AND ORGANIZATION OF A CENTRALIZED TERMINOLOGY
DATABASE
b. TERMINOLOGY EXTRACTION AND ALIGNMENT
c. STANDARDIZATION AND CONTROL
d. INTEGRATION WITH TRANSLATION WORKFLOWS
e. REPORTING AND ANALYTICS CAPABILITIES

The lecturer, Ndjimi Malaka, PhD.

F. STREAMLINE THE MANAGEMENT OF TERMINOLOGY
3. HOW TO IMPLEMENT
a. DEFINE CLEAR OBJECTIVES AND GOALS FOR TERMINOLOGY MANAGEMENT
b. SELECT THE APPROPRIATE TMS SOLUTION THAT ALIGNS WITH ORGANIZATIONAL
REQUIREMENTS AND PREFERENCES.
c. PLAN AND EXECUTE THE IMPLEMENTATION PROCESS.
d. ESTABLISH CLEAR WORKFLOWS AND PROCESSES FOR UTILIZING THE SYSTEM
EFFECTIVELY.

The lecturer, Ndjimi Malaka, PhD.

 6.2.3 BENEFITS AND IMPACT OF TERMINOLOGY
MANAGEMENT SYSTEMS

 TRANSLATION CONSISTENCY AND QUALITY.

 ENHANCE COMMUNICATION AND COLLABORATION.
 INCREASE PRODUCTIVITY AND EFFICIENCY.
 HAVE A POSITIVE IMPACT ON OVERALL ORGANIZATIONAL EFFICIENCY AND EFFECTIVENESS.

The lecturer, Ndjimi Malaka, PhD.

The lecturer, Ndjimi Malaka, PhD.
 6.2.4 THE 10 BEST AI TRANSLATION TOOLS TO USE
IN 2024

The lecturer, Ndjimi Malaka, PhD.

• 1. QUILLBOT AI

The lecturer, Ndjimi Malaka, PhD.

 A RECENT ADDITION TO THE QUILLBOT SUITE OF WRITING SERVICES—QUILLBOT TRANSLATOR
IS AN AI-POWERED TRANSLATION TOOL INTRODUCED IN 2023. IT OFFERS TRANSLATIONS IN
OVER 40 LANGUAGES AND CAN DETECT THE INPUT LANGUAGE AUTOMATICALLY.
 IT IS ESPECIALLY HANDY FOR PEOPLE USING IT FOR PARAPHRASING OR EDITING PURPOSES, AS
THEY CAN TRANSLATE TEXT ON THE SAME PLATFORM.
 YOU CAN TRANSLATE UP TO 5,000 CHARACTERS AT ONCE. THE TOOL IS COMPLETELY FREE,
WITH NEW LANGUAGES BEING ADDED CONSISTENTLY, MAKING IT THE PERFECT AI TOOL FOR
NON-PROFITS, INDIVIDUALS, AND BUSINESSES.

The lecturer, Ndjimi Malaka, PhD.

QUILLBOT AI BEST FEATURES
 TRANSLATE SHORTER AND LONGER TEXTS INTO YOUR TARGET LANGUAGE
 ACCESS FREE TRANSLATION INTO OVER 40 LANGUAGES
 EDIT, PARAPHRASE, CITE, AND TRANSLATE ON A SINGLE PLATFORM
 EXPERIENCE A MOBILE-FRIENDLY AND EASY-TO-USE INTERFACE
 TRANSLATE INTO NEW LANGUAGES AS THE APP IS CONTINUOUSLY ADDING MORE
LANGUAGES AND FEATURES

The lecturer, Ndjimi Malaka, PhD.

QUILLBOT AI LIMITATIONS
• AS THE PRODUCT IS STILL IN ITS EARLY STAGES, IT SUPPORTS FEWER LANGUAGES THAN ITS
COMPETITORS
• DOES NOT HAVE THE OPTION TO TRANSLATE DOCUMENTS
• ALLOWS ONLY THE TRANSLATION OF 5,000 CHARACTERS AT ONCE

The lecturer, Ndjimi Malaka, PhD.

2. COPY.AI
 COPY.AI LEVERAGES ARTIFICIAL INTELLIGENCE TO OFFER STELLAR TRANSLATION CAPABILITIES.
IT DOES WHAT MACHINE TRANSLATION OFTEN MISSES—TRANSLATING WORDS WHILE
RETAINING THEIR NUANCE AND INTENTION. YOU CAN USE IT TO CREATE MARKETING COPY,
BLOG POSTS, COLD EMAIL SEQUENCES, AND MORE.
 IT IS ESPECIALLY USEFUL FOR THE MARKETING AND SALES TEAMS, AS IT WORKS AS AN AI
COPYWRITING TOOL TO TRANSCREATE AND LOCALIZE YOUR MESSAGES. TEAMS CAN ADD
THEIR BRAND VOICE GUIDE TO ENSURE THE RESULTS ARE AS ACCURATE AND RELEVANT AS
POSSIBLE.
 IT OFFERS TRANSLATIONS IN OVER 25 LANGUAGES. THE BEST PART IS ITS ZERO RETENTION
DATA PRIVACY POLICY. AND THE ONLY DRAWBACK IS THAT IT IS NOT PRIMARILY A
TRANSLATION TOOL.

The lecturer, Ndjimi Malaka, PhD.

COPY.AI BEST FEATURES
 EXPLORE OVER 25 LANGUAGES
 GET RELEVANT AND HIGH-QUALITY TRANSLATIONS
 ENHANCE YOUR TRANSLATIONS BY ADDING BRAND VOICE GUIDES FOR BETTER RESULTS
 NAVIGATE EFFORTLESSLY THROUGH A USER-FRIENDLY INTERFACE
 UTILIZE PRE-BUILT WORKFLOW TEMPLATES
 LEVERAGE ITS VERSATILITY FOR WRITING AND TRANSLATING BLOGS, PRODUCT DESCRIPTIONS,
CONTENT TRANSLATION, AND MORE
 ENSURE YOUR DATA’S SECURITY AND PRIVACY WITH ROBUST MEASURES
 INTEGRATE SEAMLESSLY WITH OVER 2,000 PLATFORMS

The lecturer, Ndjimi Malaka, PhD.

COPY.AI LIMITATIONS
 PRIMARILY NOT A TRANSLATION TOOL, SO FEATURES ARE LIMITED
 SOME USERS HAVE COMPLAINED ABOUT NON-RESPONSIVE CUSTOMER SUPPORT

The lecturer, Ndjimi Malaka, PhD.

• 3. GOOGLE TRANSLATE
 STARTED AS A FREE MACHINE TRANSLATION SERVICE IN 2006, GOOGLE TRANSLATE CAN
EASILY AND ACCURATELY TRANSLATE TEXT, IMAGES, DOCUMENTS, AND WEBSITES. IN 2016, IT
SHIFTED TO NEURAL MACHINE TRANSLATION TO ALLOW FOR MORE NATURAL AND RELEVANT
TRANSLATIONS.
 THE TOOL IS EASILY ACCESSIBLE, WITH AN INTUITIVE INTERFACE THAT ALLOWS USERS TO
TRANSLATE INTO OVER 130 LANGUAGES.
 WHAT’S SPECIAL ABOUT THIS TOOL IS THAT IT CAN TRANSLATE WRITTEN WORDS, WEBSITES,
VARIOUS DOCUMENTS, SPEECH, IMAGES, AND HANDWRITTEN LANGUAGES. IT ALSO WORKS
FOR BILINGUAL CONVERSATION TRANSLATION AND AS AN AI TRANSCRIPTION TOOL.
 GOOGLE TRANSLATE IS PERFECT FOR INDIVIDUALS, LANGUAGE LEARNERS, AND TEAMS WHO
DO NOT REQUIRE EXTENSIVE TRANSLATION FEATURES.

The lecturer, Ndjimi Malaka, PhD.

GOOGLE TRANSLATE BEST FEATURES
 ENJOY A SIMPLE AND INTERACTIVE USER EXPERIENCE
 ACCESS TRANSLATIONS FOR OVER 130 LANGUAGES FOR FREE
 LEVERAGE DETECT LANGUAGE FEATURE
 ACCESS OFFLINE TRANSLATIONS IN THE MOBILE APP
 TRANSLATE DIRECTLY BY SCANNING THE TEXT WITH YOUR CAMERA
 DOWNLOAD LANGUAGES FOR OFFLINE TRANSLATIONS
 SAVE WORDS AND PHRASES FOR FASTER TRANSLATION
 FACILITATE SMOOTH CONVERSATION BETWEEN PEOPLE SPEAKING DIFFERENT LANGUAGES

The lecturer, Ndjimi Malaka, PhD.

GOOGLE TRANSLATE LIMITATIONS
 CAN OFFER INACCURATE TRANSLATIONS
 DOES NOT GUARANTEE THE PRIVACY OF THE SHARED INPUT OR OUTPUT
 SOME USERS COMPLAINED ABOUT THE TOOL NOT CAPTURING THE NUANCE OF THE
LANGUAGES

The lecturer, Ndjimi Malaka, PhD.

The lecturer, Ndjimi Malaka, PhD.
• 4. SMARTLING
 SMARTLING IS AN AI-ENABLED TRANSLATION TOOL THAT OFFERS QUICK AND ACCURATE
TRANSLATIONS AND CONTENT LOCALIZATION. IT OFFERS SEVERAL FEATURES, LIKE
PERFORMANCE AND PROGRESS DASHBOARDS, CONTEXTUAL CAT TOOLS, AND PRE-BUILT
INTEGRATIONS.
 WHETHER YOU WANT HUMAN TRANSLATION OR NEURAL MACHINE TRANSLATION,
SMARTLING OFFERS IT ALL. IT SUPPORTS TRANSLATIONS IN OVER 150 LANGUAGES AND
ALLOWS YOU TO INTEGRATE YOUR CONTENT MANAGEMENT SYSTEM FOR SMOOTH CONTENT
MANAGEMENT.
 YOU CAN TRANSLATE TUTORIALS, TESTIMONIALS, ADS, VIDEOS, INTERVIEWS, AND MORE INTO
YOUR TARGET LANGUAGE. SMARTLING ALSO OFFERS SPEECH TRANSLATION, TRANSCRIPTION,
AND CAPTIONING SERVICES.

The lecturer, Ndjimi Malaka, PhD.

SMARTLING BEST FEATURES
 TRANSLATE FOR OVER 150 LANGUAGES
 UTILIZE HUMAN TRANSLATION
 INTEGRATE SMARTLING WITH CMS EASILY
 EXPERIENCE A FULLY TRANSPARENT WORKFLOW
 BENEFIT FROM STRINGENT TRANSLATION QUALITY ASSESSMENT
 DEVELOP STYLE GUIDES AND GLOSSARIES
 USE TRANSLATION MEMORY AND TERMINOLOGY MANAGEMENT

The lecturer, Ndjimi Malaka, PhD.

SMARTLING LIMITATIONS
• DOES NOT OFFER TRANSPARENT PRICING
• HAS A STEEP LEARNING CURVE

The lecturer, Ndjimi Malaka, PhD.

The lecturer, Ndjimi Malaka, PhD.
• 5. DEEPL

The lecturer, Ndjimi Malaka, PhD.

 AMONG THE MANY MACHINE TRANSLATION TOOLS AVAILABLE ON THE MARKET, DEEPL IS
WELL-KNOWN FOR ITS ACCURATE AND NUANCED TRANSLATIONS. IT FOCUSES ON
CAPTURING THE NUANCE AND MEANING BEHIND WORDS TO OFFER THE MOST RELEVANT
OUTPUT.
 THE INTERFACE IS INTUITIVE AND SUPPORTS 30 LANGUAGES. WITH THE HELP OF NEURAL
NETWORKS, THE TOOL CREATES ACCURATE AND CONTEXTUALLY APPROPRIATE
TRANSLATIONS.
 YOU CAN USE IT TO TRANSLATE PDFS, WORD DOCUMENTS, AND POWERPOINT FILES. IT ALSO
ALLOWS YOU TO SELECT ALTERNATIVE WORDS OR PHRASES AND AUTOMATICALLY ADJUSTS
THE OTHER PHRASES AS PER THE SELECTION.
 DEEPL’S FREE VERSION ENABLES YOU TO CREATE YOUR OWN GLOSSARY, OFFERS AN IN-BUILT
DICTIONARY, AND SETS RULES REGARDING SPECIFIC TRANSLATIONS.
 WITH DEEPL PRO, YOU CAN ALSO ADJUST THE TONE OF YOUR TRANSLATIONS, INTEGRATE IT
WITH YOUR CAT TOOL, AND MORE.
The lecturer, Ndjimi Malaka, PhD.
DEEPL BEST FEATURES
 USE AN EASY AND FRIENDLY INTERFACE
 INTEGRATE SEAMLESSLY WITH WINDOWS, IOS, AND CHROME
 ACCESS EXPERT AND EXCEPTIONAL ACCURACY IN THE LANGUAGES OFFERED
 SET RULES FOR WORDS AND PHRASES
 TRANSLATE LARGER FILES EASILY WITH SIMPLE LANGUAGE TRANSLATION
 MAINTAIN THE TEXT’S ORIGINAL FORMATTING WITH THE PRO VERSION
 ENSURE HIGH DATA SECURITY FOR YOUR INPUTS AND OUTPUTS
DEEPL LIMITATIONS
 MANY FEATURES ARE ONLY AVAILABLE IN THE PAID VERSION
 CAN TRANSLATE ONLY 1,500 CHARACTERS IN THE FREE VERSION
 LIMITED CUSTOMIZATION OPTIONS
The lecturer, Ndjimi Malaka, PhD.
6. MICROSOFT BING TRANSLATOR
• MICROSOFT BING TRANSLATOR IS A PART OF MICROSOFT COGNITIVE SERVICES THAT WORKS
WITH MICROSOFT OFFICE, EDGE, BING, SKYPE, AND MORE. IT OFFERS TEXT, SPEECH, MULTI-
DEVICE CONVERSATIONS, IMAGES, AND OFFLINE TRANSLATIONS (ONLY FOR ANDROID).
• THE TRANSLATION TOOL SUPPORTS OVER 100 LANGUAGES. THE INTERFACE IS USER-FRIENDLY.
IT OFFERS SEAMLESS INTEGRATION WITH MICROSOFT PRODUCTS. YOU CAN ADD THE
TRANSLATOR TO YOUR OUTLOOK IF YOU REGULARLY TRANSLATE EMAILS.

The lecturer, Ndjimi Malaka, PhD.

MICROSOFT BING TRANSLATOR BEST FEATURES
 TRANSLATE EFFORTLESSLY WITH AN EASY-TO-USE INTERFACE
 ENGAGE IN REAL-TIME CONVERSATIONS AND SPEECH TRANSLATION
 INTEGRATE SEAMLESSLY WITH MICROSOFT APPS
 DETECT LANGUAGE FOR QUICK TRANSLATION AUTOMATICALLY
 TRANSLATE QUICKLY BY USING WIDELY USED PHRASES AVAILABLE ON THE PLATFORM
 TRANSLATE INTO OVER 100 LANGUAGES
MICROSOFT BING TRANSLATOR LIMITATIONS
 USERS HAVE COMPLAINED ABOUT INACCURATE LANGUAGE TRANSLATION IN LONG TEXTS
 THE SPEAKING FEATURE IS NOT AVAILABLE FOR ALL LANGUAGES
 DOES NOT ALLOW YOU TO CHANGE TRANSLATED WORDS

The lecturer, Ndjimi Malaka, PhD.

The lecturer, Ndjimi Malaka, PhD.
7. UNBABEL
• UNBABEL IS A LANGOPS PLATFORM THAT OFFERS A COMBINATION OF HUMAN AND AI
TRANSLATIONS TO OFFER HIGHLY ACCURATE RESULTS. IT IS BEST FOR BUSINESSES AND TEAMS AS IT
OFFERS A COMPREHENSIVE SET OF ADVANCED FEATURES, INCLUDING REAL-TIME REPORTING,
TRANSPARENCY, AND AI-DRIVEN WORKFLOWS THAT SAVE TIME AND COSTS.
• THE AI QUALITY ESTIMATION OFFERS A TRANSPARENT LOOK INTO YOUR TRANSLATION
PERFORMANCE. WHAT’S MORE, IT IS ISO 27001 CERTIFIED, MAKING IT A SECURE PLATFORM.
UNBABEL BEST FEATURES
• ACCESS TRANSLATIONS INTO 90 LANGUAGES
• LEVERAGE A COMBINATION OF AI AND HUMAN TRANSLATORS FOR HIGH-QUALITY TRANSLATIONS
• INTEGRATE WITH A WIDE RANGE OF TOOLS LIKE ZENDESK, SALESFORCE, AND HUBSPOT
• EVALUATE THE QUALITY OF TRANSLATIONS TO ENSURE ACCURACY
• TAILOR TRANSLATION WORKFLOWS TO SUIT YOUR NEEDS
• USE READY-MADE TEMPLATES

The lecturer, Ndjimi Malaka, PhD.

UNBABEL LIMITATIONS
 CAN MAKE ERRORS IN TRANSLATIONS
 SOME USERS COMPLAINED ABOUT THE LONG TRANSLATION TIME

The lecturer, Ndjimi Malaka, PhD.

8. SMARTCAT
 SMARTCAT IS AN AI TRANSLATION TOOL THAT OFFERS CONTEXTUAL TRANSLATIONS AND AN INTUITIVE
INTERFACE. IT OFFERS COLLABORATIVE WORKFLOWS AND QUICK TRANSLATIONS, MAKING IT PERFECT FOR
GLOBAL ENTERPRISES AND TEAMS. IT ALLOWS YOU TO ADD AS MANY USERS AS YOU WANT.
 THE TRANSLATION QUALITY AND ACCURACY IMPROVE AS YOU CONTINUE TO USE IT. THE TOOL SUPPORTS
OVER 280 LANGUAGES AND 50 FILE FORMATS. YOU CAN CONNECT WITH 30+ INTEGRATIONS TO
STREAMLINE YOUR WORK.
SMARTCAT BEST FEATURES
 ACCESS GLOSSARIES AND TRANSLATION MEMORIES FOR IMPROVED CONSISTENCY
 VIEW ACCURACY RATINGS FOR SPECIFIC LANGUAGE PAIRS
 TRANSLATE INTO 280+ LANGUAGES
 LEVERAGE A COMBINATION OF HUMAN EXPERTISE WITH AI TRANSLATIONS
 HARNESS THE POWER OF GPT-4 FOR ENHANCED TRANSLATION CAPABILITIES
 USE AI MATCHING TO FIND THE BEST LINGUISTS
 EDIT AND VERIFY TRANSLATED CONTENT IN THE SMARTCAT EDITOR
SMARTCAT LIMITATIONS
 NO FREE VERSION
 SOME USERS COMPLAINED ABOUT UNRESPONSIVE CUSTOMER SUPPORT

The lecturer, Ndjimi Malaka, PhD.

The lecturer, Ndjimi Malaka, PhD.
9. REVERSO TRANSLATION
REVERSO TRANSLATION BEST FEATURES
 GET EASY AND FAST TRANSLATION
 CREATE VOCABULARY AND LEARN WITH FLASHCARDS
 ACCESS IN-CONTENT EXAMPLES
 TRANSLATE DOCUMENTS AND TEXTS WITH EASE
 BENEFIT FROM ROBUST DATA PROTECTION
 UNLOCK LONGER TEXT TRANSLATION, INSTANT TRANSLATION, AND NO ADS WITH THE PAID
VERSION
REVERSO TRANSLATION LIMITATIONS
 MANY FEATURES ARE AVAILABLE IN THE PAID VERSION
 SUPPORTS LESS LANGUAGE THAN ALTERNATIVES

The lecturer, Ndjimi Malaka, PhD.

The lecturer, Ndjimi Malaka, PhD.
10. SYSTRAN
SYSTRAN BEST FEATURES
 ENSURES A ROBUST PRIVACY POLICY FOR DATA PROTECTION
 TRAIN YOUR OWN AI MODEL FOR BETTER RESULTS
 GET INDUSTRY- AND DOMAIN-SPECIFIC TRANSLATION
 CUSTOMIZE TRANSLATIONS TO MEET YOUR UNIQUE NEEDS
 CREATE YOUR GLOSSARY AND TRANSLATION MEMORY
 USE ADVANCED USER DICTIONARIES
 SPECIFY CUSTOM RULES TO REFINE TRANSLATION OUTPUT
SYSTRAN LIMITATIONS
 DOCUMENT TRANSLATION IS ONLY AVAILABLE IN THE PAID VERSION
 STEEP LEARNING CURVE

The lecturer, Ndjimi Malaka, PhD.

• WHICH OF THESE WOULD YOU SELECT TO WORK WITH? WHY?

The lecturer, Ndjimi Malaka, PhD.

 EXAMPLES OF COMMERCIAL MACHINE TRANSLATION SYSTEMS

The lecturer, Ndjimi Malaka, PhD.

 EXAMPLES OF ONLINE MACHINE TRANSLATION SYSTEMS

The lecturer, Ndjimi Malaka, PhD.

The lecturer, Ndjimi Malaka, PhD.
The lecturer, Ndjimi Malaka, PhD.
 EXAMPLES OF ONLINE DICTIONARIES

The lecturer, Ndjimi Malaka, PhD.

The lecturer, Ndjimi Malaka, PhD.
The lecturer, Ndjimi Malaka, PhD.
The lecturer, Ndjimi Malaka, PhD.
 EXAMPLES OF ONLINE THESAURI

The lecturer, Ndjimi Malaka, PhD.

The lecturer, Ndjimi Malaka, PhD.
 EXAMPLES OF ONLINE GLOSSARIES

The lecturer, Ndjimi Malaka, PhD.

The lecturer, Ndjimi Malaka, PhD.
The lecturer, Ndjimi Malaka, PhD.
 EXAMPLES OF ONLINE ENCYCLOPAEDIA

The lecturer, Ndjimi Malaka, PhD.

 LANGUAGE KEY

The lecturer, Ndjimi Malaka, PhD.

 GENERAL FRAMEWORKS FOR EVALUATING
TRANSLATION TOOLS

Meeting stated or implied needs of an end-user when

Functionality functions of the system operate under specific conditions.

• MAINTAINING THE LEVEL OF PERFORMANCE BY THE SYSTEM WHEN

Reliability OPERATING UNDER SPECIFIC CONDITIONS.

The lecturer, Ndjimi Malaka, PhD.

 The ease of operating, understanding and
Usability: learning each task of the system as a whole.

The performance of the system in relation to

Efficiency: the amount of resources available.

The capability of the system to undergo modifications such as

corrections, improvement and adaptations for different
Maintainability: requirements and working environments.

The ability to transfer the system from one environment

Portability: to another such as to different operating systems.
The lecturer, Ndjimi Malaka, PhD.
 COURSE REVIEW

1. WHAT IS INVOLVED IN TT?

2. WHAT MT SYSTEMS ARE YOU AWARE OF?
3. CAN TECHNOLOGY REPLACE HUMAN TRANSLATORS?
4. IS IT POSSIBLE TO HAVE A ‘PERFECT MT’?
5. WHY IS TRANSLATION TECHNOLOGY IMPORTANT IN TODAY’S GLOBALIZED WORLD?
6. HOW DOES TRANSLATION TECHNOLOGY HELP BUSINESSES?

The lecturer, Ndjimi Malaka, PhD.

7. WHAT IS YOUR UNDERSTANDING OF THE FOLLOWING CONCEPTS:
a. MACHINE TRANSLATION (MT);
b. HUMAN-AIDED/ASSISTED MACHINE TRANSLATION (HAMT);
c. MACHINE-AIDED HUMAN TRANSLATION (MAHT)
d. COMPUTER-AIDED/ASSISTED TRANSLATION (CAT);
8. WHAT ARE THE LEVELS OF MACHINE TRANSLATION?
9. HOW CAN WE CLASSIFY MACHINE TRANSLATION SYSTEM BASED ON USAGE?
10. IS MT EXCLUSIVE TO MACHINES? DO YOU THINK HUMANS SHOULD BE INVOLVED? WHAT
SHOULD THEIR ROLE BE?
11. CAN AN EXPERIENCED TRANSLATOR USE A BASIC LEVEL SYSTEM? IF YES, IN WHAT
SITUATIONS? IF NOT, WHY NOT?
12. WHICH TRANSLATION SYSTEM WOULD YOU USE IF YOU HAD TO TRANSLATE A NOVEL?

The lecturer, Ndjimi Malaka, PhD.

13. DESCRIBE A FIRST-GENERATION SYSTEM.
14. GIVE EXEMPLES OF SECOND-GENERATION SYSTEMS.
15. WHAT IS AN ARCHITECTURE?
16. WHAT ARE THE MACHINE TRANSLATION SYSTEM COMPONENTS?
17. WHAT IS A STRUCTURAL REPRESENTATION?
18. DISCUSS MACHINE TRANSLATION SYSTEMS APPROACHES.
19. WHAT ARE HYBRID AND INTERACTIVE MACHINE TRANSLATION SYSTEMS?
20. GIVE EXAMPLES OF INTERACTIVE MACHINE TRANSLATION SYSTEMS?

The lecturer, Ndjimi Malaka, PhD.

21. TRUE/FALSE
a. A ‘WORKBENCH’ OR A ‘WORKSTATION’ IS A SINGLE INTEGRATED SYSTEM THAT IS MADE UP
OF A NUMBER OF TRANSLATION TOOLS AND RESOURCES.
b. A MULTILINGUAL TEXT ARCHIVE CONTAINS SEGMENTED, ALIGNED, PARSED AND
DECLASSIFIED MULTILINGUAL TEXTS.
c. THE DIFFERENCE BETWEEN A MACHINE TRANSLATION SYSTEMS AND TRANSLATION MEMORY
SYSTEMS IS THAT MTS GENERATES TRANSLATIONS AUTOMATICALLY; WHEREAS, TMS ALLOW
PROFESSIONAL TRANSLATORS TO BE IN CHARGE OF THE DECISION-MAKING PROCESS.

The lecturer, Ndjimi Malaka, PhD.

22. WHAT IS YOUR UNDERSTANDING OF THE FOLLOWING CONCEPS:
a. WORKBENCHES
b. TRANSLATION MEMORY,
c. AN ALIGNMENT TOOL,
d. FILTER,
e. TERMINOLOGY DATABASES,
f. TERMINOLOGY MANAGEMENT SYSTEM
g. PERFECT MATCHING
h. FUZZY MATCHING
i. SEGMENTATION
j. TRANSLATION UNIT

The lecturer, Ndjimi Malaka, PhD.

K. THE TRANSLATION WORKFLOW
23. ELABORATE ON:
a. TYPES OF TERMINOLOGY MANAGEMENT SYSTEMS.
b. KEY FEATURES AND FUNCTIONS OF TERMINOLOGY MANAGEMENT SYSTEMS
c. HOW TO IMPLEMENT AND UTILIZE TERMINOLOGY MANAGEMENT SYSTEMS

The lecturer, Ndjimi Malaka, PhD.

 NDJIMI DUMBA WATEMBO MALAKA, DOCTOR IN LINGUISTICS.

The lecturer, Ndjimi Malaka, PhD.