Some extra stuffSome extra stuff

• Semantic change that results in an antonym Semantic change that results in an antonym of the original word:of the original word:

• awfulawful: original meaning: 'awe-inspiring, filling : original meaning: 'awe-inspiring, filling (someone) with deep awe', as in (someone) with deep awe', as in the awful the awful majesty of the Creator; majesty of the Creator; new meaning:new meaning: 'breath-'breath-takingly bad; so bad that it fills (a person) with takingly bad; so bad that it fills (a person) with awe and amazement'awe and amazement'

• Compare it to Compare it to awesomeawesome• Similar pair: Similar pair: terrible vs. terrificterrible vs. terrific

Some extra stuffSome extra stuff

• cool vs. hotcool vs. hot: they could be both positive (e.g. : they could be both positive (e.g. That’s really cool; hot buys) but not alwaysThat’s really cool; hot buys) but not always (e.g. hot check)(e.g. hot check)

• Intermediate ones likeIntermediate ones like lukewarm lukewarm do not have do not have positive or negative meanings.positive or negative meanings.

Computational LinguisticsComputational Linguistics

Ling 400Ling 400

What is Linguistics good for?What is Linguistics good for?

• Language teachingLanguage teaching

• Medical: speech therapy for aphasic Medical: speech therapy for aphasic patientspatients

• Computer-related applicationsComputer-related applications– Computational linguistics (also called Computational linguistics (also called

natural language processing, language natural language processing, language engineering): computer simulation of engineering): computer simulation of human language processeshuman language processes

Some online materialsSome online materials

• Some relevant information including a Some relevant information including a lecture by Prof. Emily Bender is lecture by Prof. Emily Bender is available from the following web site.available from the following web site.

• http://depts.washington.edu/llc/olr/http://depts.washington.edu/llc/olr/linguistics/index.phplinguistics/index.php

Computational LinguisticsComputational Linguistics

• encoding/production:encoding/production: speech synthesis, speech synthesis, word processing help, production side word processing help, production side of an expert system, generation of of an expert system, generation of sentences in the target language in sentences in the target language in machine translation.machine translation.

• decoding/understanding:decoding/understanding: speech speech recognition, parsing, disambiguation via recognition, parsing, disambiguation via a network of semantic relations.a network of semantic relations.

Language ProductionLanguage Production

• thinkingthinking: cannot be simulated: cannot be simulated• speech/writing:speech/writing: computer simulation of computer simulation of

speech sounds is possible to some speech sounds is possible to some extent. Computer can help this process extent. Computer can help this process with a grammar checker, an input with a grammar checker, an input system and a word breaker (in a system and a word breaker (in a language like Japanese). But these language like Japanese). But these tasks do not simulate what people tasks do not simulate what people actually do when they talk.actually do when they talk.

Language Production (2)Language Production (2)

• Though not part of the natural production Though not part of the natural production process, turning speech into written text has process, turning speech into written text has some practical applications.some practical applications.

• This is very useful because speaking is This is very useful because speaking is usually quicker than writing. It would be like usually quicker than writing. It would be like having a personal secretary.having a personal secretary.

• This is also useful for someone who cannot This is also useful for someone who cannot write because of disability or injury.write because of disability or injury.

Language UnderstandingLanguage Understanding

• speech recognitionspeech recognition: difficult but possible if the : difficult but possible if the domain is restricted (e.g. speaker and/or domain is restricted (e.g. speaker and/or expected input types) expected input types) Why is this difficult?Why is this difficult?

• syntactic analysis:syntactic analysis: “parsing” (syntactic analysis by “parsing” (syntactic analysis by computer) is possible but needs computer) is possible but needs semantic/pragmatic information for semantic/pragmatic information for disambiguating instances of structural ambiguity.disambiguating instances of structural ambiguity.

• Interpretation (truth conditions)Interpretation (truth conditions): unclear as to how : unclear as to how to simulate this; usually done via semantic to simulate this; usually done via semantic representations (in some machine translation representations (in some machine translation systems).systems).

Ambiguity ResolutionAmbiguity Resolution

• The astronomer saw a comet with a The astronomer saw a comet with a telescope.telescope.

• The astronomer saw a comet with a The astronomer saw a comet with a brilliant tail.brilliant tail.

How do we solve its ambiguity?How do we solve its ambiguity?

One needs to know some information One needs to know some information about seeing and comets.about seeing and comets.

Semantic NetworkSemantic Network

Concrete ApplicationsConcrete Applications• corpus linguisticscorpus linguistics• machine translationmachine translation• text retrievaltext retrieval• text summarizationtext summarization• word processing help (discussed above)word processing help (discussed above)• expert systemsexpert systems• speech recognition/synthesis (touched upon speech recognition/synthesis (touched upon

above)above)• toys, gamestoys, games• automatic telephone interpretation systemautomatic telephone interpretation system• ultimately … artificial intelligence, roboticsultimately … artificial intelligence, robotics

Corpus LinguisticsCorpus Linguistics• This is a generic name for various This is a generic name for various

computer applications that make use of computer applications that make use of large language databases (called corpora)large language databases (called corpora)

• Having access to a large database enabled Having access to a large database enabled us to process linguistic data in a statistical us to process linguistic data in a statistical way, rather than in an analytical way.way, rather than in an analytical way.

• This conflict of two opposing views This conflict of two opposing views (statistical vs. analytical) is very apparent in (statistical vs. analytical) is very apparent in machine translation.machine translation.

Machine Translation (1)Machine Translation (1)

• text-to-text translation (great need for text-to-text translation (great need for translation at UN, EC (European translation at UN, EC (European Community)Community)

• Works best when two languages in Works best when two languages in question are similar in structurequestion are similar in structure

• Usually, pre-editing and/or post-editing Usually, pre-editing and/or post-editing by a human translator is required — by a human translator is required — machine-assisted translation. machine-assisted translation.

Machine Translation (2)Machine Translation (2)

• Traditionally, MT required parsing, possibly Traditionally, MT required parsing, possibly some semantic analysis, then mapping to a some semantic analysis, then mapping to a syntactic tree of the sentence in the target syntactic tree of the sentence in the target language.language.

• An alternative is appeal to statistical means An alternative is appeal to statistical means of mapping a surface string in the source of mapping a surface string in the source language to a surface string in the target language to a surface string in the target language.language.

http://www.excite.co.jp/world/english/web/http://www.excite.co.jp/world/english/web/

Machine Translation (3)Machine Translation (3)

• Difficulty with word-for-word translationDifficulty with word-for-word translation

Watasi-ni-wa futari kodomo-ga imasu.Watasi-ni-wa futari kodomo-ga imasu.

I -at-top two child-nom existI -at-top two child-nom exist

Literally, ‘As for me as a location, two Literally, ‘As for me as a location, two children exist’children exist’

Actually, its meaning is ‘I have two Actually, its meaning is ‘I have two children.’children.’

Computational SemanticsComputational Semantics

• The study of how to automate the process of The study of how to automate the process of constructing and reasoning with meaning constructing and reasoning with meaning representations of natural language representations of natural language expressions.expressions.

• This could play an important role in such This could play an important role in such application areas as machine translation application areas as machine translation when two typologically distinct languages are when two typologically distinct languages are involved (e.g. English and Japanese).involved (e.g. English and Japanese).

Text RetrievalText Retrieval

key word key word text/book text/bookkey word: morphologykey word: morphology1.1. Principles of Polymer MorphologyPrinciples of Polymer Morphology2.2. Image Analysis and Mathematical MorphologyImage Analysis and Mathematical Morphology3.3. Drainage Basin MorphologyDrainage Basin Morphology4.4. French MorphologyFrench MorphologyWe need morphological, syntactic, and semantic We need morphological, syntactic, and semantic

information to find the right text/book.information to find the right text/book.Further applications: search engines, etc.Further applications: search engines, etc.

Text SummarizationText Summarization

• We need to be able to select the right information from the electronic documents available (esp. on the web).

• Automatic text summarization is a technique that can help people to quickly grasp the concepts presented in a document by creating an abstract or summary of the original text.

Semantic WebSemantic Web

• Some people (e.g. Evergreen U) are trying to classify contents of web pages so that they are meaningful to computers. But this is not an easy task since the categories must presumably be pre-selected by people.

• The semanticThe semantic WebWeb provides a common provides a common framework that allows data to be shared and framework that allows data to be shared and reused across application, enterprise, and reused across application, enterprise, and community boundaries. community boundaries. http://www.w3.org/2001/sw/

Speech Speech Recognition/SynthesisRecognition/Synthesis

• actually being used on personal actually being used on personal computers (on a limited basis), computers (on a limited basis), automated telephone answering automated telephone answering system, etc.system, etc.

• Application of acoustic phonetics, Application of acoustic phonetics, phonologyphonology

AI and roboticsAI and roboticsFurby

Aibo