1. 1 AI@Azusa Pacific University Sheldon Liang, Ph D Computer Science Department

2. 2 AI@Azusa Pacific University Sense, Communicate, Actuate

3. 3 Natural?  Natural Language?  Refers to the language spoken by people, e.g. English, Chinese, Swahili, as opposed to artificial languages, like C++, Java, etc.  Natural Language Processing  Applications that deal with natural language in a way or another and it is the subfield of Artificial Intelligence  Computational Linguistics  Doing linguistics on computers  More on the linguistic side than NLP, but closely related  Natural Language?  Refers to the language spoken by people, e.g. English, Chinese, Swahili, as opposed to artificial languages, like C++, Java, etc.  Natural Language Processing  Applications that deal with natural language in a way or another and it is the subfield of Artificial Intelligence  Computational Linguistics  Doing linguistics on computers  More on the linguistic side than NLP, but closely related AI@Azusa Pacific University

4. 4 What is Artificial Intelligence?  The use of computer programs and programming techniques to cast light on the principles of intelligence in general and human thought in particular (Boden)  AI is the study of how to do things which at the moment people do better (Rich & Knight)  AI is the science of making machines do things that would require intelligence if done by men. (Minsky)  The use of computer programs and programming techniques to cast light on the principles of intelligence in general and human thought in particular (Boden)  AI is the study of how to do things which at the moment people do better (Rich & Knight)  AI is the science of making machines do things that would require intelligence if done by men. (Minsky) AI@Azusa Pacific University

5. 5 Why Natural Language Processing? Why Natural Language Processing?  kJfmmfj mmmvvv nnnffn333  Uj iheale eleee mnster vensi credur  Baboi oi cestnitze  Coovoel2^ ekk; ldsllk lkdf vnnjfj?  Fgmflmllk mlfm kfre xnnn!  kJfmmfj mmmvvv nnnffn333  Uj iheale eleee mnster vensi credur  Baboi oi cestnitze  Coovoel2^ ekk; ldsllk lkdf vnnjfj?  Fgmflmllk mlfm kfre xnnn! AI@Azusa Pacific University

6. 6 Computers Lack Knowledge!  Computers “see” text in English the same you have seen the previous text!  People have no trouble understanding language  Common sense knowledge  Reasoning capacity  Experience  Computers have  No common sense knowledge  No reasoning capacity Unless we teach them!  Computers “see” text in English the same you have seen the previous text!  People have no trouble understanding language  Common sense knowledge  Reasoning capacity  Experience  Computers have  No common sense knowledge  No reasoning capacity Unless we teach them! AI@Azusa Pacific University

7. 7 Why Natural Language Processing?  Huge amounts of data  Internet = at least 8 billion pages  Intranet  Applications for processing large amounts of texts  Require NLP expertise  Huge amounts of data  Internet = at least 8 billion pages  Intranet  Applications for processing large amounts of texts  Require NLP expertise  Classify text into categories  Index and search large texts  Automatic translation  Speech understanding  Understand phone conversations  Information extraction  Extract useful information from resumes  Automatic summarization  Condense 1 book into 1 page  Question answering  Knowledge acquisition  Text generations / dialogs AI@Azusa Pacific University

8. 8 Where does it fit in the CS taxonomy? Computers & Applications Artificial Intelligence AlgorithmsDatabasesNetworking Robotics Search Natural Language Processing Information Retrieval Machine Translation Language Analysis SemanticsParsing AI@Azusa Pacific University

9. 9 Situating NLP computer science psychology/cognitive science linguistics math/statistics philosophy communication NLP AI@Azusa Pacific University

10. 10 Theoretical foundations  math: statistics, calculus, algebra, modeling  computational paradigms: connectionist, rule- based, cognitively plausible  linguistics: LFG, HPSG, GB, OT, CG, etc.  architectures: stacks, automata, networks, compilers  math: statistics, calculus, algebra, modeling  computational paradigms: connectionist, rule- based, cognitively plausible  linguistics: LFG, HPSG, GB, OT, CG, etc.  architectures: stacks, automata, networks, compilers AI@Azusa Pacific University

11. 11 Some areas of research  Corpora, tools, resources, standards  Language/grammar engineering  Machine (assisted) translation, tools  Language modeling  Lexicography  Speech  Corpora, tools, resources, standards  Language/grammar engineering  Machine (assisted) translation, tools  Language modeling  Lexicography  Speech AI@Azusa Pacific University

12. 12 Linguistics Essentials AI@Azusa Pacific University

13. 13 The Description of Language  Language = Words and Rules  Dictionary (vocabulary) + Grammar  Dictionary set of words defined in the language open (dynamic)  Traditional paper based  Electronic machine readable dictionaries; can be obtained from paper-based  Grammar set of rules which describe what is allowable in a language  Classic Grammars meant for humans who know the language  definitions and rules are mainly supported by examples  no (or almost no) formal description tools; cannot be programmed  Explicit Grammar (CFG, Dependency Grammars, Link Grammars,...) formal description can be programmed & tested on data (texts)  Language = Words and Rules  Dictionary (vocabulary) + Grammar  Dictionary set of words defined in the language open (dynamic)  Traditional paper based  Electronic machine readable dictionaries; can be obtained from paper-based  Grammar set of rules which describe what is allowable in a language  Classic Grammars meant for humans who know the language  definitions and rules are mainly supported by examples  no (or almost no) formal description tools; cannot be programmed  Explicit Grammar (CFG, Dependency Grammars, Link Grammars,...) formal description can be programmed & tested on data (texts) AI@Azusa Pacific University

14. 14 Linguistics Levels of Analysis  Speech  Written language  Phonology: sounds / letters / pronunciation  Morphology: the structure of words  Syntax: how these sequences are structured  Semantics: meaning of the strings  Interaction between levels where each level has an input and an output.  Speech  Written language  Phonology: sounds / letters / pronunciation  Morphology: the structure of words  Syntax: how these sequences are structured  Semantics: meaning of the strings  Interaction between levels where each level has an input and an output. AI@Azusa Pacific University

15. 15 Phonetics/Orthography  Input:  acoustic signal (phonetics) / text (orthography)  Output:  phonetic alphabet (phonetics) / text (orthography)  Deals with:  Phonetics:  consonant & vowel (& others) formation in the vocal tract  classification of consonants, vowels,... in relation to frequencies, shape & position of the tongue and various muscles  intonation  Orthography: normalization, punctuation, etc.  Input:  acoustic signal (phonetics) / text (orthography)  Output:  phonetic alphabet (phonetics) / text (orthography)  Deals with:  Phonetics:  consonant & vowel (& others) formation in the vocal tract  classification of consonants, vowels,... in relation to frequencies, shape & position of the tongue and various muscles  intonation  Orthography: normalization, punctuation, etc. AI@Azusa Pacific University

16. 16 Phonology -- pronunciation  Input:  sequence of phones/sounds (in a phonetic alphabet); or “normalized” text (sequence of (surface) letters in one language’s alphabet) [NB: phones vs. phonemes]  Output:  sequence of phonemes (~ (lexical) letters; in an abstract alphabet)  Deals with:  relation between sounds and phonemes (units which might have some function on the upper level)  e.g.: [u] ~ oo (as in book), [æ] ~ a (cat); i ~ y (flies)  Input:  sequence of phones/sounds (in a phonetic alphabet); or “normalized” text (sequence of (surface) letters in one language’s alphabet) [NB: phones vs. phonemes]  Output:  sequence of phonemes (~ (lexical) letters; in an abstract alphabet)  Deals with:  relation between sounds and phonemes (units which might have some function on the upper level)  e.g.: [u] ~ oo (as in book), [æ] ~ a (cat); i ~ y (flies) AI@Azusa Pacific University

17. 17 Morphology -- the structure of words  Input: sequence of phonemes (~ (lexical) letters)  Output:  sequence of pairs (lemma, (morphological) tag)  Deals with:  composition of phonemes into word forms and their underlying lemmas (lexical units) + morphological categories (inflection, derivation, compounding)  e.g. quotations ~ quote/V + -ation(der.V->N) + NNS.  Input: sequence of phonemes (~ (lexical) letters)  Output:  sequence of pairs (lemma, (morphological) tag)  Deals with:  composition of phonemes into word forms and their underlying lemmas (lexical units) + morphological categories (inflection, derivation, compounding)  e.g. quotations ~ quote/V + -ation(der.V->N) + NNS. AI@Azusa Pacific University

18. 18...and Beyond  Input:  sentence structure (tree): annotated nodes (autosemantic lemmas, (morphosyntactic) tags, deep functions)  Output:  logical form, which can be evaluated (true/false)  Deals with:  assignment of objects from the real world to the nodes of the sentence structure  e.g.: (I/Sg/Pat/t (see/Perf/Pred/t) Tom/Sg/Ag/f) ~ see(Mark-Twain[SSN:...],Tom-Sawyer[SSN:...])  Input:  sentence structure (tree): annotated nodes (autosemantic lemmas, (morphosyntactic) tags, deep functions)  Output:  logical form, which can be evaluated (true/false)  Deals with:  assignment of objects from the real world to the nodes of the sentence structure  e.g.: (I/Sg/Pat/t (see/Perf/Pred/t) Tom/Sg/Ag/f) ~ see(Mark-Twain[SSN:...],Tom-Sawyer[SSN:...]) AI@Azusa Pacific University

19. 19 Phonology  (Surface « Lexical) Correspondence  “symbol-based” (no complex structures)  Ex.: (stem-final change)  lexical: b a b y + s (+ denotes start of ending)  surface: b a b i e s (phonetic-related: b é b ì 0s )  Arabic: (interfixing, inside-stem doubling)  lexical: kTb+uu+CVCCVC (CVCC...vowel/consonant pattern)  surface: kuttub  (Surface « Lexical) Correspondence  “symbol-based” (no complex structures)  Ex.: (stem-final change)  lexical: b a b y + s (+ denotes start of ending)  surface: b a b i e s (phonetic-related: b é b ì 0s )  Arabic: (interfixing, inside-stem doubling)  lexical: kTb+uu+CVCCVC (CVCC...vowel/consonant pattern)  surface: kuttub AI@Azusa Pacific University

20. 20 Phonology Examples  German (umlaut) (satz ~ sentence)  lexical: s A t z + e (A denotes “umlautable” a)  surface: s ä t z e (phonetic: z æ c e, vs. zac )  Turkish ( vowel harmony )  lexical: e v + l A r (~house)  surface: e v l e r  German (umlaut) (satz ~ sentence)  lexical: s A t z + e (A denotes “umlautable” a)  surface: s ä t z e (phonetic: z æ c e, vs. zac )  Turkish ( vowel harmony )  lexical: e v + l A r (~house)  surface: e v l e r AI@Azusa Pacific University

21. 21 Morphology: Morphemes & Order  Scientific study of forms of words  Grouping of phonemes into morphemes  sequence deliverables  deliver, able and s (3 units)  could as well be some “ID” numbers:  e.g. deliver ~ 23987, s ~ 12, able ~ 3456  Morpheme Combination  certain combinations/sequencing possible, other not:  deliver+able+s, but not able+derive+s; noun+s, but not noun+ing  typically fixed (in any given language)  Scientific study of forms of words  Grouping of phonemes into morphemes  sequence deliverables  deliver, able and s (3 units)  could as well be some “ID” numbers:  e.g. deliver ~ 23987, s ~ 12, able ~ 3456  Morpheme Combination  certain combinations/sequencing possible, other not:  deliver+able+s, but not able+derive+s; noun+s, but not noun+ing  typically fixed (in any given language) AI@Azusa Pacific University

22. 22 The Dictionary (or Lexicon)  Repository of information about words:  Morphological:  description of morphological “behavior”: inflection patterns/classes  Syntactic:  Part of Speech  relations to other words:  subcategorization (or “surface valency frames”)  Semantic:  semantic features  frames ...and any other! (e.g., translation)  Repository of information about words:  Morphological:  description of morphological “behavior”: inflection patterns/classes  Syntactic:  Part of Speech  relations to other words:  subcategorization (or “surface valency frames”)  Semantic:  semantic features  frames ...and any other! (e.g., translation) AI@Azusa Pacific University

23. 23 AI@Azusa Pacific University Sense, Communicate, Actuate

24. 24 (Surface) Syntax  Input:  sequence of pairs (lemma, (morphological) tag)  Output:  sentence structure (tree) with annotated nodes (all lemmas, (morphosyntactic) tags, functions), of various forms  Deals with:  the relation between lemmas & morphological categories and the sentence structure  uses syntactic categories such as Subject, Verb, Object,...  e.g.: I/PP1 see/VB a/DT dog/NN ~  ((I/sg)SB ((see/pres)V (a/ind dog/sg)OBJ)VP)S  Input:  sequence of pairs (lemma, (morphological) tag)  Output:  sentence structure (tree) with annotated nodes (all lemmas, (morphosyntactic) tags, functions), of various forms  Deals with:  the relation between lemmas & morphological categories and the sentence structure  uses syntactic categories such as Subject, Verb, Object,...  e.g.: I/PP1 see/VB a/DT dog/NN ~  ((I/sg)SB ((see/pres)V (a/ind dog/sg)OBJ)VP)S AI@Azusa Pacific University

25. 25 Issues in Syntax Issues in Syntax “the dog ate my homework” - Who did what? 1.Identify the part of speech (POS) Dog = noun ; ate = verb ; homework = noun English POS tagging: 95% Can be improved! Part of speech tagging on other languages almost inexistent 2. Identify collocations mother in law, hot dog Compositional versus non-compositional collocates “the dog ate my homework” - Who did what? 1.Identify the part of speech (POS) Dog = noun ; ate = verb ; homework = noun English POS tagging: 95% Can be improved! Part of speech tagging on other languages almost inexistent 2. Identify collocations mother in law, hot dog Compositional versus non-compositional collocates AI@Azusa Pacific University

26. 26 Issues in Syntax Issues in Syntax  Shallow parsing: “the dog chased the bear” “the dog” “chased the bear” subject - predicate Identify basic structures NP-[the dog] VP-[chased the bear] Shallow parsing on new languages Shallow parsing with little training data  Shallow parsing: “the dog chased the bear” “the dog” “chased the bear” subject - predicate Identify basic structures NP-[the dog] VP-[chased the bear] Shallow parsing on new languages Shallow parsing with little training data AI@Azusa Pacific University

27. 27 Issues in Syntax Issues in Syntax  Full parsing: John loves Mary Current precisions: 85-88% Help figuring out (automatically) questions like: Who did what and when? AI@Azusa Pacific University

28. 28 Meaning (semantics) Meaning (semantics)  Input:  sentence structure (tree) with annotated nodes (lemmas, (morphosyntactic) tags, surface functions)  Output:  sentence structure (tree) with annotated nodes (semantic lemmas, (morpho-syntactic) tags, deep functions)  Deals with:  relation between categories such as “Subject”, “Object” and (deep) categories such as “Agent”, “Effect”; adds other cat’s  e.g. ((I)SB ((was seen)V (by Tom)OBJ)VP)S ~  (I/Sg/Pat/t (see/Perf/Pred/t) Tom/Sg/Ag/f)  Input:  sentence structure (tree) with annotated nodes (lemmas, (morphosyntactic) tags, surface functions)  Output:  sentence structure (tree) with annotated nodes (semantic lemmas, (morpho-syntactic) tags, deep functions)  Deals with:  relation between categories such as “Subject”, “Object” and (deep) categories such as “Agent”, “Effect”; adds other cat’s  e.g. ((I)SB ((was seen)V (by Tom)OBJ)VP)S ~  (I/Sg/Pat/t (see/Perf/Pred/t) Tom/Sg/Ag/f) AI@Azusa Pacific University

29. 29 Issues in Semantics  Understand language! How?  “plant” = industrial plant  “plant” = living organism  Words are ambiguous  Importance of semantics?  Machine Translation: wrong translations  Information Retrieval: wrong information  Anaphora Resolution: wrong referents  Understand language! How?  “plant” = industrial plant  “plant” = living organism  Words are ambiguous  Importance of semantics?  Machine Translation: wrong translations  Information Retrieval: wrong information  Anaphora Resolution: wrong referents AI@Azusa Pacific University

30. 30  The sea is at the home for billions of factories and animals  The sea is home to million of plants and animals  English  French [commercial MT system]  Le mer est a la maison de billion des usines et des animaux  French  English  The sea is at the home for billions of factories and animals  The sea is home to million of plants and animals  English  French [commercial MT system]  Le mer est a la maison de billion des usines et des animaux  French  English Why Semantics? AI@Azusa Pacific University

31. 31 Issues in Semantics  How to learn the meaning of words?  From dictionaries: plant, works, industrial plant -- (buildings for carrying on industrial labor; "they built a large plant to manufacture automobiles") plant, flora, plant life -- (a living organism lacking the power of locomotion) They are producing about 1,000 automobiles in the new plant The sea flora consists in 1,000 different plant species The plant was close to the farm of animals.  How to learn the meaning of words?  From dictionaries: plant, works, industrial plant -- (buildings for carrying on industrial labor; "they built a large plant to manufacture automobiles") plant, flora, plant life -- (a living organism lacking the power of locomotion) They are producing about 1,000 automobiles in the new plant The sea flora consists in 1,000 different plant species The plant was close to the farm of animals. AI@Azusa Pacific University

32. 32 Issues in Semantics  Learn from annotated examples:  Assume 100 examples containing “plant” previously tagged by a human  Train a learning algorithm  Precisions in the range 60%-70%-(80%) How to choose the learning algorithm? How to obtain the 100 tagged examples?  Learn from annotated examples:  Assume 100 examples containing “plant” previously tagged by a human  Train a learning algorithm  Precisions in the range 60%-70%-(80%) How to choose the learning algorithm? How to obtain the 100 tagged examples? AI@Azusa Pacific University

33. 33 Issues in Learning Semantics  Learning?  Assume a (large) amount of annotated data = training  Assume a new text not annotated = test  Learn from previous experience (training) to classify new data (test)  Decision trees, memory based learning, neural networks  Machine Learning Which one performs best?  Learning?  Assume a (large) amount of annotated data = training  Assume a new text not annotated = test  Learn from previous experience (training) to classify new data (test)  Decision trees, memory based learning, neural networks  Machine Learning Which one performs best? AI@Azusa Pacific University

34. 34 Issues in Semantics  Automatic annotation of data  Active learning  Identify only the hard examples  Co-training  Identify the examples where several techniques agree on the semantic tag  Collecting from Web users  Open Mind Word Expert  Automatic annotation of data  Active learning  Identify only the hard examples  Co-training  Identify the examples where several techniques agree on the semantic tag  Collecting from Web users  Open Mind Word Expert AI@Azusa Pacific University

35. 35 Problems faced by Natural Language-Understanding Systems AI@Azusa Pacific University

36. 36 Key NLP problem: ambiguity AI@Azusa Pacific University Human Language is highly ambiguous at all levels acoustic level recognize speech vs. wreck a nice beach morphological level saw: to see (past), saw (noun), to saw (present, inf) syntactic level I saw the man on the hill with a telescope semantic level One book has to be read by every student

37. 37 Key NLP problem: Ambiguity AI@Azusa Pacific University Human Language is highly ambiguous at all levels acoustic level recognize speech vs. wreck a nice beach morphological level saw: to see (past), saw (noun), to saw (present, inf) syntactic level I saw the man on the hill with a telescope semantic level One book has to be read by every student

38. 38 Language Model AI@Azusa Pacific University  A formal model about language  Two types  Non-probabilistic  Allows one to compute whether a certain sequence (sentence or part thereof) is possible  Often grammar based  Probabilistic  Allows one to compute the probability of a certain sequence  Often extends grammars with probabilities  A formal model about language  Two types  Non-probabilistic  Allows one to compute whether a certain sequence (sentence or part thereof) is possible  Often grammar based  Probabilistic  Allows one to compute the probability of a certain sequence  Often extends grammars with probabilities

39. 39 Example of Bad Language Model AI@Azusa Pacific University

40. 40 Example of Bad Language Model AI@Azusa Pacific University

41. 41 Example of Bad Language Model AI@Azusa Pacific University

42. 42 A Good Language Model AI@Azusa Pacific University  Non-Probabilistic  “I swear to tell the truth” is possible  “I swerve to smell de soup” is impossible  Probabilistic  P(I swear to tell the truth) ~.0001  P(I swerve to smell de soup) ~ 0  Non-Probabilistic  “I swear to tell the truth” is possible  “I swerve to smell de soup” is impossible  Probabilistic  P(I swear to tell the truth) ~.0001  P(I swerve to smell de soup) ~ 0

43. 43 Language Model Application AI@Azusa Pacific University  Spelling correction  Mobile phone texting  Speech recognition  Handwriting recognition  Disabled users  …  Spelling correction  Mobile phone texting  Speech recognition  Handwriting recognition  Disabled users  …

44. 44 Speech & Text segmentation  In spoken language, sounds representing succesive letters blend into each other  This makes the conversion of the analog signal to discrete characters very difficult  Regarding Text Segmentation, Some written languages like chinese, japanese and thai don’t have signal word boundaries.  So any significant text parsing requires identifying word boundaries, which is often a non-trivial tasks  In spoken language, sounds representing succesive letters blend into each other  This makes the conversion of the analog signal to discrete characters very difficult  Regarding Text Segmentation, Some written languages like chinese, japanese and thai don’t have signal word boundaries.  So any significant text parsing requires identifying word boundaries, which is often a non-trivial tasks AI@Azusa Pacific University

45. 45 Word sense disambiguation  Word sense disambiguation is the problem of selecting a sense for a word from a set of predefined possibilities.  Sense Inventory usually comes from a dictionary or thesaurus.  Knowledge intensive methods, supervised learning, and (sometimes) bootstrapping approaches  Word sense discrimination is the problem of dividing the usages of a word into different meanings, without regard to any particular existing sense inventory.  Unsupervised techniques  Word sense disambiguation is the problem of selecting a sense for a word from a set of predefined possibilities.  Sense Inventory usually comes from a dictionary or thesaurus.  Knowledge intensive methods, supervised learning, and (sometimes) bootstrapping approaches  Word sense discrimination is the problem of dividing the usages of a word into different meanings, without regard to any particular existing sense inventory.  Unsupervised techniques AI@Azusa Pacific University

46. 46 Word sense disambiguation Computers versus Humans  Polysemy – most words have many possible meanings.  A computer program has no basis for knowing which one is appropriate, even if it is obvious to a human…  Ambiguity is rarely a problem for humans in their day to day communication, except in extreme cases…  Polysemy – most words have many possible meanings.  A computer program has no basis for knowing which one is appropriate, even if it is obvious to a human…  Ambiguity is rarely a problem for humans in their day to day communication, except in extreme cases… AI@Azusa Pacific University

47. 47 Word sense disambiguation Ambiguity for a Computer  The fisherman jumped off the bank and into the water.  The bank down the street was robbed!  Back in the day, we had an entire bank of computers devoted to this problem.  The bank in that road is entirely too steep and is really dangerous.  The plane took a bank to the left, and then headed off towards the mountains.  The fisherman jumped off the bank and into the water.  The bank down the street was robbed!  Back in the day, we had an entire bank of computers devoted to this problem.  The bank in that road is entirely too steep and is really dangerous.  The plane took a bank to the left, and then headed off towards the mountains. AI@Azusa Pacific University

48. 48 Syntactic ambiguity Syntactic ambiguity  There are often multiple possible parse trees for a given sentence.  Choosing the most appropriate one usually requires semantic and contextual information.  Specific problem components here are: 1.Sentence boundary disambiguation 2.Imperfect input 3.Foreign or regional accents etc.  There are often multiple possible parse trees for a given sentence.  Choosing the most appropriate one usually requires semantic and contextual information.  Specific problem components here are: 1.Sentence boundary disambiguation 2.Imperfect input 3.Foreign or regional accents etc. AI@Azusa Pacific University

49. 49 Syntactic ambiguity Syntactic ambiguity AI@Azusa Pacific University

50. 50 Statistical NLP  Statistical NLP uses stochastic, probabilistic and statistical methods to resolve some difficulties of NLP  Methods for disambiguation of an involve the use of corpora & Markov models.  Technology for statistical NLP comes from machine learning and data mining both of which involve learning from data.  Statistical NLP uses stochastic, probabilistic and statistical methods to resolve some difficulties of NLP  Methods for disambiguation of an involve the use of corpora & Markov models.  Technology for statistical NLP comes from machine learning and data mining both of which involve learning from data. AI@Azusa Pacific University

51. 51 Statistical NLP -- Corpus AI@Azusa Pacific University  Corpus: text collection for linguistic purposes  Tokens How many words are contained in Tom Sawyer?  71.370  Types How many different words are contained in T.S.?  8.018  Hapax Legomena words appearing only once

52. 52 Statistical NLP – Word Counts AI@Azusa Pacific University  The most frequent words are function words wordfreqwordfreq the3332in906 and2972that877 a1775he877 to1725I783 of1440his772 was1161you686 it1027Tom679

53. 53 Major Tasks in NLP  Speech Recognition  Natural Language Generation  Machine Translation  Information Retrieval  Information Extraction  Text Simplification  Automatic summarization  Foreign Language Reading & writing aid  Speech Recognition  Natural Language Generation  Machine Translation  Information Retrieval  Information Extraction  Text Simplification  Automatic summarization  Foreign Language Reading & writing aid AI@Azusa Pacific University

54. 54 Speech Recognition  It is the process of converting a speech signal to a sequence of words, by means of an algorithm (as computer program).  Applications are : 1.Voice dialing 2.Call routing 3.Simple data entry 4.Preparation of structure documents  It is the process of converting a speech signal to a sequence of words, by means of an algorithm (as computer program).  Applications are : 1.Voice dialing 2.Call routing 3.Simple data entry 4.Preparation of structure documents AI@Azusa Pacific University

55. 55 Natural Language generation  It is a task of generating Natural Language from a machine representation system such as a knowledge base or a logical form. Ex: Choose randomly among outputs: – Visitant which came into the place where it will be Japanese has admired that there was Mount Fuji.  Top 10 outputs according to bigram probabilities: – Visitors who came in Japan admire Mount Fuji. – Visitors who came in Japan admires Mount Fuji. – Visitors who arrived in Japan admire Mount Fuji. – A visitor who came in Japan admire Mount Fuji. – The visitor who came in Japan admire Mount Fuji. – Visitors who came in Japan admire Mount Fuji. – The visitor who came in Japan admires Mount Fuji. – Mount Fuji is admired by a visitor who came in Japan.  It is a task of generating Natural Language from a machine representation system such as a knowledge base or a logical form. Ex: Choose randomly among outputs: – Visitant which came into the place where it will be Japanese has admired that there was Mount Fuji.  Top 10 outputs according to bigram probabilities: – Visitors who came in Japan admire Mount Fuji. – Visitors who came in Japan admires Mount Fuji. – Visitors who arrived in Japan admire Mount Fuji. – A visitor who came in Japan admire Mount Fuji. – The visitor who came in Japan admire Mount Fuji. – Visitors who came in Japan admire Mount Fuji. – The visitor who came in Japan admires Mount Fuji. – Mount Fuji is admired by a visitor who came in Japan. AI@Azusa Pacific University

56. 56 Conclusion AI@Azusa Pacific University  Overview of some probabilistic and machine learning methods for NLP  Also very relevant to bioinformatics !  Analogy between parsing  A sentence  A biological string (DNA, protein, mRNA, …)  Overview of some probabilistic and machine learning methods for NLP  Also very relevant to bioinformatics !  Analogy between parsing  A sentence  A biological string (DNA, protein, mRNA, …)

57. 57 AI@Azusa Pacific University Sense, Communicate, Actuate

58. 58 Machine Translations Machine Translation or MT is a sub-field of computational linguistics that investigates usage of computer software to translate text or speech from one natural language to another Machine Translation or MT is a sub-field of computational linguistics that investigates usage of computer software to translate text or speech from one natural language to another AI@Azusa Pacific University

59. 59 Issues in Machine Translations  Text to Text Machine Translations  Speech to Speech Machine Translations  Most of the work has addressed pairs of widely spread languages like English-French, English-Chinese  How to translate text?  Learn from previously translated data  Need parallel corpora  French-English, Chinese-English have the Hansards  Reasonable translations  Chinese-Hindi – no such tools available today!  Text to Text Machine Translations  Speech to Speech Machine Translations  Most of the work has addressed pairs of widely spread languages like English-French, English-Chinese  How to translate text?  Learn from previously translated data  Need parallel corpora  French-English, Chinese-English have the Hansards  Reasonable translations  Chinese-Hindi – no such tools available today! AI@Azusa Pacific University

60. 60 Issues in Machine Translations  How to obtain parallel texts?  From the Web! How?  From Web users! How?  Once we have the texts, how to get most out of them?  Word alignments  Obtain lexicons  Import knowledge from well studied languages  How to obtain parallel texts?  From the Web! How?  From Web users! How?  Once we have the texts, how to get most out of them?  Word alignments  Obtain lexicons  Import knowledge from well studied languages AI@Azusa Pacific University

61. 61 Information Extraction  It’s a type of information retrieval whose goal is to automatically extract structured or semi structured information from unstructured machine readable documents.  Its significance is determined by the growing amount of information available in unstructured form, for instance on the Internet.  It’s a type of information retrieval whose goal is to automatically extract structured or semi structured information from unstructured machine readable documents.  Its significance is determined by the growing amount of information available in unstructured form, for instance on the Internet. AI@Azusa Pacific University

62. 62 Issues in Information Extraction  “There was a group of about 8-9 people close to the entrance on Highway 75”  Who? “8-9 people”  Where? “highway 75”  Extract information  Detect new patterns:  Detect hacking / hidden information / etc.  Gov./mil. puts lots of money put into IE research  “There was a group of about 8-9 people close to the entrance on Highway 75”  Who? “8-9 people”  Where? “highway 75”  Extract information  Detect new patterns:  Detect hacking / hidden information / etc.  Gov./mil. puts lots of money put into IE research AI@Azusa Pacific University

63. 63 Information Retrieval Information Retrieval (IR) is a science of searching  for information in documents,  for documents themselves,  for metadata or  searching with in databases (any kind). Information Retrieval (IR) is a science of searching  for information in documents,  for documents themselves,  for metadata or  searching with in databases (any kind). AI@Azusa Pacific University

64. 64 Issues in Information Retrieval  Index meaning  Search for plant (=living organism) should not retrieve texts with plant (=industrial plant)  But should retrieve documents including “flora” or other related terms  Index parsed relations  Index meaning  Search for plant (=living organism) should not retrieve texts with plant (=industrial plant)  But should retrieve documents including “flora” or other related terms  Index parsed relations AI@Azusa Pacific University

65. 65 Issues in Information Retrieval  Retrieve specific information  Question Answering  “What is the height of mount Everest?”  11,000 feet  Current state-of-the-art 40-50% Improve precision with the use of more common sense knowledge Perform domain specific question answering  Retrieve specific information  Question Answering  “What is the height of mount Everest?”  11,000 feet  Current state-of-the-art 40-50% Improve precision with the use of more common sense knowledge Perform domain specific question answering AI@Azusa Pacific University

66. 66 Issues in Information Retrieval  Find information across languages!  Cross Language Information Retrieval  “What is the minimum age requirement for car rental in Italy?”  Search also Italian texts for “eta minima per noleggio macchine”  Integrate large number of languages  Integrate into performant IR engines  Find information across languages!  Cross Language Information Retrieval  “What is the minimum age requirement for car rental in Italy?”  Search also Italian texts for “eta minima per noleggio macchine”  Integrate large number of languages  Integrate into performant IR engines AI@Azusa Pacific University

67. 67 Automatic Summarization  It is the creation of a shortened version of a text by a computer program.  As access to data has increased so has interest in automatic summarization. An example of the use of summarization technology is search engines such as Google.  Technologies that can make a coherent summary, of any kind of text, need to take into account several variables such as length, writing –style and syntax to make a useful summary.  It is the creation of a shortened version of a text by a computer program.  As access to data has increased so has interest in automatic summarization. An example of the use of summarization technology is search engines such as Google.  Technologies that can make a coherent summary, of any kind of text, need to take into account several variables such as length, writing –style and syntax to make a useful summary. AI@Azusa Pacific University

68. 68 Foreign Language Writing Aid Foreign Language Writing Aid  It is a computer program that assists a non-native language user in their target language.  Assistive operations can be classified into two categories: on-the-fly prompts and post-writing checks.  Assisted aspects of writing include: Lexical syntax, Lexical semantics, idiomatic expression transfer, etc.  On-line dictionaries can also be considered as a type of foreign language writing aid.  It is a computer program that assists a non-native language user in their target language.  Assistive operations can be classified into two categories: on-the-fly prompts and post-writing checks.  Assisted aspects of writing include: Lexical syntax, Lexical semantics, idiomatic expression transfer, etc.  On-line dictionaries can also be considered as a type of foreign language writing aid. AI@Azusa Pacific University

69. 69 Language & speech technology have advanced rapidly in the last decades. AI@Azusa Pacific University

70. 70 It is EveR-2 Muse, a robot version of a Korean woman in her twenties (Eve+R for robot), can hold a conversation or sing a song, make eye contact, and express anger, sorrow and joy. But according to her creator, most Koreans found her homely in comparison to her predecessor AI@Azusa Pacific University

71. 71 Achievements of AI/ NLP  Sphinx can recognise continuous speech.  Deep Thought is an international grand master chess player. Without training for each speaker, it operates in near real time using a vocabulary of 1000 words and has 94% word accuracy.  Navlab is a truck that can drive along a road at 55mph in normal traffic.  Carlton and United Breweries use an AI planning system to plan production of their beer.  Natural language interfaces to databases can be obtained on a PC.  Machine Learning methods have been used to build expert systems.  Expert systems are used regularly in finance, medicine, manufacturing, and agriculture  Sphinx can recognise continuous speech.  Deep Thought is an international grand master chess player. Without training for each speaker, it operates in near real time using a vocabulary of 1000 words and has 94% word accuracy.  Navlab is a truck that can drive along a road at 55mph in normal traffic.  Carlton and United Breweries use an AI planning system to plan production of their beer.  Natural language interfaces to databases can be obtained on a PC.  Machine Learning methods have been used to build expert systems.  Expert systems are used regularly in finance, medicine, manufacturing, and agriculture AI@Azusa Pacific University

72. 72 If this dream comes alive…  Even a person who is ignorant of computer knowledge can interact with it through a colloquial interaction.  Almost all systems will be automated.  Many problems will have found a solution.  No one needs to learn computer languages any more, instead they can interact with the computer in their natural (regional) languages themselves.  It would be a matter of jubilance for the world as a whole…..  Even a person who is ignorant of computer knowledge can interact with it through a colloquial interaction.  Almost all systems will be automated.  Many problems will have found a solution.  No one needs to learn computer languages any more, instead they can interact with the computer in their natural (regional) languages themselves.  It would be a matter of jubilance for the world as a whole….. AI@Azusa Pacific University

73. 73 So lets await that wonderful day & work in this direction…. AI@Azusa Pacific University