Chapter 15 Natural Language Processing (cont) 323-670 Artificial Intelligence ดร.วิภาดา เวทย์ประสิทธิ์ ภาควิชาวิทยาการคอมพิวเตอร์ คณะวิทยาศาสตร์ มหาวิทยาลัยสงขลานครินทร์

The same expression means different things in different context. NLP Problems Figure 15.1 P. 378 English sentences are incomplete descriptions of the information that are intended to convey. The same expression means different things in different context. No natural language program can be complete because of new words, expression, and meaning can be generated quite freely. There are lots of ways to say the same thing. 323-670 Artificial Intelligence Lecture36-40 Page 2

1) Processing written text NLP Problems 1) Processing written text using lexical, syntactic, and semantic knowledge of the language the require real world information 2) Processing spoken language using all information needed above plus additional knowledge about phonology handle ambiguities in speech 323-670 Artificial Intelligence Lecture36-40 Page 3

Natural Language processing NLP Natural Language processing Language translation / multilingual translation Language understanding Figure 14.5 p. 365 Interaction among component Figure 14.6 p. 366 A speech Waveform 323-670 Artificial Intelligence Lecture36-40 Page 4

1) Morphological Analysis 2) Syntactic Analysis 3) Semantic Analysis Step in NLP 1) Morphological Analysis 2) Syntactic Analysis 3) Semantic Analysis 4) Discourse Integration 5) Pragmatic Analysis boundaries between these five phrases are often fuzzy. 323-670 Artificial Intelligence Lecture36-40 Page 5

1. Morphological Analysis Individual words are analyzed into components Nonword tokens such as punctuation are separated from the words I want to print Bill’s .int file. file extension proper noun possessive suffix 323-670 Artificial Intelligence Lecture36-40 Page 6

linear sequence of words are transformed into structures 2. Syntactic Analysis linear sequence of words are transformed into structures show how words relate to each other English syntactic analyzer If do not pass the syntactic analyzer  reject (Boy the go to store the) 323-670 Artificial Intelligence Lecture36-40 Page 7

Example of syntactic analysis Figure 15.2 p. 382  RM2, RM5, RM5 A knowledge base Fragment Figure 15.3 p. 383 User073, F1, Printing, File_Structure, Waiting Mental Event/ Physical Event Animate/Event Partial meaning for a sentence Figure 15.4 p. 384 323-670 Artificial Intelligence Lecture36-40 Page 8

the structures created by the syntactic analyser are assign meanings 3. Semantic Analysis the structures created by the syntactic analyser are assign meanings mapping between the syntactic structure and objects in the task domain If no mapping  reject (colorless green ideas sleep furiously) 1) It must map individual words into appropriate objects in the knowledge base or database. 2) It must create the correct structures to correspond to the meanings of the individual words combine with each other. 323-670 Artificial Intelligence Lecture36-40 Page 9

4. Discourse Integration the meaning of the individual sentence may depend on the sentences that precede it and may influence the meanings of the sentences that follow it. (Ex. John want it.)  “It” depends on the previous sentence. Current user who type word “I” is User068 = Susan_Black We get F1 with filename in /wsmith/ directory 323-670 Artificial Intelligence Lecture36-40 Page 10

(Ex. Do you know what time it is?) 5. Pragmatic Analysis The structure representing what was said is reinterpreted to determine what was actually meant. (Ex. Do you know what time it is?)  we should understand what to do.... Understand to decide what to do as a result Representing the intended meaning Figure 15.5 P. 385 323-670 Artificial Intelligence Lecture36-40 Page 11

Grammar declarative representation Syntactic Processing Grammar declarative representation syntactic facts about the language Figure 15.6 p.387 Parser procedure compares the grammar against input sentences to produce parsed structure. Figure 15.7 p.388 A parse tree for a sentence 323-670 Artificial Intelligence Lecture36-40 Page 12

Syntactic Processing Top-down Parsing Begin with start symbol and apply the grammar rules forward until the symbols at the terminals of the tree correspond to the components of the sentence being parsed. Bottom-up Parsing Begin with the sentence to be parsed and apply the grammar rules backward until a single tree whose terminals are the words of the sentence and whose top node is the start symbol has been produced. 323-670 Artificial Intelligence Lecture36-40 Page 13

ATN : Augmented Transition Network similar to finite state machine Figure 15.8 p.392 An ATN network Figure 15.9 p.3923An ATN Grammar in List Form sentence  “The long file has printed.” S  NP  Q1  AUX  Q3  V  Q4 (F) halt NP  Det Q6  Adj Q6  N  Q7 (F) (S DCL (NP (FILE (LONG) DEFINITE)) HAS (VP PRINTED)) 323-670 Artificial Intelligence Lecture36-40 Page 14

DAGs : Direct Acyclic Graph Unification Grammar DAGs : Direct Acyclic Graph the graph corresponding to “the” and “file” are [CAT: DET [CAT: N [NP [DET: N LEX:the] LEX : file HEAD: file NUMBER: SING] NUMBER: SING]] 323-670 Artificial Intelligence Lecture36-40 Page 15

Lexicon disambiguaty or Word sense disambiguaty Semantic Processing Step 1. Lexicon  look up the individual words in a dictionary and extract their meaning. Step 2 Lexicon disambiguaty or Word sense disambiguaty words may have more than one meaning e.g. bank (ธนาคาร หรือ ตลิ่ง), diamond p.398 (เพชร หรือ รูปเหลี่ยม) use semantic marker  PHYSICAL OBJECT, ANIMATE OBJECT, ABSTRACT OBJECT e.g. I drop my diamond... 323-670 Artificial Intelligence Lecture36-40 Page 16

Semantic Processing use semantic marker  PHYSICAL OBJECT ANIMATE OBJECT ABSTRACT OBJECT e.g. I drop my diamond... note that “ My lawn hates the cold” is a correct semantic as well, although lawn can not act the verb hate..... but this sentence can be use in the good English sense. 323-670 Artificial Intelligence Lecture36-40 Page 17

Sentence level Processing 1) Semantic grammars 2) Case grammars 3) Conceptual parsing 4) Approximately compositional semantic interpretation 323-670 Artificial Intelligence Lecture36-40 Page 18

Sentence level Processing 1) Semantic grammars - semantic action associate with the grammar rule - Figure 15.10 p. 401 e.g. “I want to”  ACTION - Figure 15.11 p. 402 Parsing Result with semantic grammars 323-670 Artificial Intelligence Lecture36-40 Page 19

Sentence level Processing 2) Case grammars : passing process driven from the sentence’s main verb. - Figure 15.12 p. 404 Active and passive sentence : Susan printed a file. = The file was printed by Susan. - Figure 15.13 p. 404 Similar sentence : Mother baked for three hours. = The pie baked for three hours. - Word case p. 405 : (A) Agent, (I) Instrument, (F) Factitive, (L) Locatives, (S) Source, (G) Goal, (B) Beneficiary, (T) Time, (O) Object 323-670 Artificial Intelligence Lecture36-40 Page 20

Sentence level Processing 2) Case grammars - Figure 15.14 p. 406 Some verb case frames : open [_ _ O (I) (A)] A : Instigator of the action : die [_ _ D] D : Entity effect by the action John die. : kill [_ _ D (I) (A)] Bill killed John. Bill killed John with a knife. : want [_ _A O] John wanted some ice cream. John wanted Marry to go to store. 323-670 Artificial Intelligence Lecture36-40 Page 21

Sentence level Processing 3) Conceptual parsing : strategy for finding both the structure and the meaning of the sentence in one step. use verb-ACT dictionary Figure 15.15 p. 407 e.g. “want” 1) stative (wanting something to happen) 2) transitive ATRANS (wanting an object) 3) intransitive PTRANS (wanting a person) 323-670 Artificial Intelligence Lecture36-40 Page 22

Sentence level Processing 3) Conceptual parsing CD : Conceptual dependency structure : passing process driven from the sentence’s main verb with more details in the lower level. Figure 15.16 p. 407 “John wanted Mary to go to the store.” (PTRANS) Figure 15.17 p. 409 “John went to the park with the peacocks.” (PTRANS) 323-670 Artificial Intelligence Lecture36-40 Page 23

Sentence level Processing 4) Approximately compositional semantic interpretation - Semantic interpretation rules Figure 15.18 p. 411 - Combining Mapping Knowledge Figure 15.19 p. 412 wanting : agent : (animate), object : (state or event) 323-670 Artificial Intelligence Lecture36-40 Page 24

Discourse and Pragmatic Processing use to understand a single sentence. p. 415 – 416 Relationship between discourse contexts Names of individuals Dave went to the movie. Parts of actions. 1) John went on a business trip to New York. 2) He left on an early morning flight. Causal chains.... Planning sequences.... 323-670 Artificial Intelligence Lecture36-40 Page 25

Discourse and Pragmatic Processing knowledge to focus the current focus on dialogue a model of each participant’s current beliefs the goal driven character of dialogue the rule of conversation shared by all participants Modeling Individual beliefs p.419 Model logic : Three belief spaces Figure 15.20 p.420 temporal logics allow to talk about the truth of the set of proposition at current state of the real world, in the pat, and in the future as well. conditional logic allow to talk about the truth or falsehood under some circumstances 323-670 Artificial Intelligence Lecture36-40 Page 26