Scribd
Upload
20 views21 pages

Lecture02 Ambiguity

This lecture discusses the various levels of linguistic representation in Natural Language Processing (NLP), including phonetics, morphology, syntax, semantics, and pragmatics. It highlights the challenges of ambiguity present in natural language and the importance of understanding context for accurate interpretation. Additionally, it covers the complexities of language variety, non-literal speech, and morphological structures that affect NLP techniques.

Uploaded by

yl5404
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
20 views21 pages

Lecture02 Ambiguity

This lecture discusses the various levels of linguistic representation in Natural Language Processing (NLP), including phonetics, morphology, syntax, semantics, and pragmatics. It highlights the challenges of ambiguity present in natural language and the importance of understanding context for accurate interpretation. Additionally, it covers the complexities of language variety, non-literal speech, and morphological structures that affect NLP techniques.

Uploaded by

yl5404
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 21

Natural Language

Processing
Lecture 2: Levels of Representation and Ambiguity

09/06/2024

Columbia COMS W4705


Daniel Bauer
Levels of Linguistic
Representation
phonetics sounds and sound /bɔɪ/
phonology patterns of language

morphology formation of words in- + validate + -ed


DT | NN | VBZ | DT | NN | TO | VB | PRP |.
the | boy | want+s | the | girl | to | like | him |.
S
VP
syntax word order S
VP
NP NP VP
DT NN VBZ DT NN TO VB PRP
the boy wants the girl to like him

word and sentence


semantics
meaning want
arg0 arg1
boy like arg0
in uence of context and arg1
girl
pragmatics
situation
fl
Language Understanding /
Generation as Translation
• Most NLP techniques can be understood as translation tasks
from one structure (level of representation) into another.

• For each translation step:

• Construct search space of possible translations.

• Find best paths through this space (decoding) according


to some performance measure.

• Modern NLP relies on Machine Learning to gure out these


translation steps.
fi
NLP is hard: Ambiguity
• Unlike arti cial languages, natural language is full of ambiguity.

• This can happen on all levels of representation.

• “Wreck a Nice Beach” , “Recognize Speech”

• “in ammable” = in + - ammable

• “Enraged Cow Injures Farmer with Axe”

• “Stolen Painting Found by Tree”

• “Red Tape Holds Up New Bridges”

• “Mouse”
fl
fi
fl
Lexical Ambiguity and
Translation
Lost in Translation
en: the attorney general
fr: L’avocat général
en: the general avocado

en: the spirit is strong but the esh is weak

ru: ...

en: the vodka is good but the meat is rotten


fl
https://www.star.radio/news/world-news/coronavirus-monkeys-escape-with-covid-19-samples-after-attacking-
lab-assistant/
More Real Headlines
• Ban on nude dancing on Governor’s desk

• Eye drops o shelf

• Kids Make Nutritious Snacks

• Drunk gets nine months in violin case

• Government head seeks arms

• Patient at death’s door – doctors pull him through

• In America a woman has a baby every 15 minutes


ff
Syntactic Structure
• What is the part-of-speech of each word? (noun, verb, adjective,
adverb, determiner, …)

• What are the constituents:

• Noun phrase: “Enraged cow”, “The cat with the hat”,


“Columbia University”

• What are the subjects and objects:

• “Dog bites man” vs. “Man bites dog”

• Modi cation:

• “John saw the man in the park with a telescope”


fi
Structural Ambiguity
• Interplay between constituent structure and modi cation.

• Prepositional Phrase (PP) attachment:

Enraged cow injures farmer with axe.

fi
Representing Modi cation
with Brackets

[Enraged cow] [injures [farmer [with axe]]]


NP NP PP

[Enraged cow] injures [farmer] [with axe]


NP NP PP

fi
More PP attachment
[Ban]
Ban on
on nude
[nudedancing
dancing]onongovernor’s
[governor’s
desk
desk]
NP NP NP

• What are the possible modi cations? Which one is correct?

[Ban]NP [on nude dancing]PP [on governor's desk]PP

[Ban]NP [on [nude dancing on governor's desk]NP]PP

[[Ban]NP [on nude dancing]PP]NP [on governor's desk]PP


fi
Noun-Noun Modi cation
• Compound nouns also have internal structure:

country song platinum album

1. [country [song [platinum album]]]

2. [country [[song platinum] album]]

3. [[country song] [platinum album]]

4. [[country [song platinum]] album]

5. [[[country song] platinum] album]


fi
Noun-Noun Modi cation
• What is the semantic relationship between nouns in a noun
compound?

• Water fountain: A fountain that supplies water.

• Water ballet: A ballet that takes place in water.

• Water meter: A device that measures water.

• Water barometer: A barometer that uses water (instead


of mercury) to measure air pressure.

• Water glass: A glass that is meant to hold water.


fi
Other tricky phenomena
• Need for deep semantic knowledge.
There was once a Wolf who saw a Lamb drinking at a river and wanted an excuse to eat it.

For that purpose, even though he himself was upstream, he accused the Lamb of stirring
up the water and keeping him from drinking. . .

Wolf

Lamb

Minsky 1975
Other tricky phenomena
• Winograd Schemes

• Coreference resolution requires requires semantic


understanding.

The trophy wouldn't t into the suitcases because it was too big.
fi
Other tricky issues:
Language Variety
• Problem: Most NLP techniques were developed on
English (speci cally nancial news written in American
English in the 1990s), or other languages with many
resources.

• Languages use di erent mechanisms to express meaning


(morphology vs. word-order).
fi
ff
fi
Other tricky issues: Domains
and Language Change

• Non-standard English

• Idioms: throw in the towel, get cold feet, kick the bucket

• Neologisms ( xed lexicon doesn’t work)

• noob, crowdsource, unfriend, retweet, bromance, …


fi
Non-literal speech

• Metaphor: This news is music to my ears.

• Sarcasm: Visiting the dentist is so much fun!

• Idioms: It's not rocket science.


Morphology
• Structure and formation of words.

• Derivational morphology: Create new words from old words


(can also change the part-of-speech).
anti- + dis- + establish + -ment + -arian + -ism

• In ectional morphology:

• Convey information about number, person, tense, aspect,


mood, voice, and the role a word plays in the sentence (case).

• English has few morphological categories, but many


languages are morphologically rich.
fl
Morphology
• Morphological categories in English

• Number (“dog”, “dog +s”)

• Person (“I run”, “She runs”)

• Tense (“He wait+ed”)

• Voice (“The issue was decid+ed”)

• Other examples from other languages?

You might also like