SEMANTIC INTERPRETATION Semantic interpretation is a large process as it involves various components (representation of text that can be fed into a computer to allow other computational manipulations in order to understand any language). Katz and Fodor in their 1963 paper "The Structure of a Semantic Theory” put forward what properties a semantic theory should posses, Asemantic theory should be able to : 1. Explain sentences having ambiguous meanings. For example, it should account forthe fact that the word bilin the sentence The bil is large is ambiguous in the sense that it could represent money or the beak ofthe bird. 2, Resolve the ambiguities of words in context. For example, ifthe same sentence is extended to form The bill is large but need to be paid, then the theory should be able to disambiguate the monetary ‘meaning of the bill 3. Ideny meaningless bur bu sraccaly wel formed sentences, such she famous example by __ Chomsky: Colorless green ideas sleep fu1. Explain sentences having ambiguous meanings. For example, it should account forthe fact that the Word bilin the sentence The bil s large is ambiguous in the sense that it could represent money or the beak of the bird, 2. Resolve the ambiguities of words in context. For example, if the same sentence is extended to form The bills large but need to be paid, then the theory should be able to disambiguate the monetary meaning of the bill, 3. Identify meaningless but but syntactically well-formed sentences, such as the famous example by Chomsky: Colorless green ideas sleep furiously. 4, Identify syntactically or transformationally unrelated paraphrases of a concept having the same semantic content. Requirements for achieving a semantic representation: 1, Structural Ambiguity # 2. Word sense 3. Entity and Event resolution 4, Predicate-Argument Structure 5, Meaning RepresentationSTRUCTURAL AMBIGUITY Structural ambiguity arises when a sentence has more than one meaning due to the way words are arranged in that sentence, For example, The sentence Sara caught he butterfly by the tree's structurally ambiguous because it has 2 meanings: 1, Sara caught the butterfly while she was standing by the tree, 2. Sara caught the butterfly which was fluttering near the tree,WORD SENSE The same word type, or word lemma, is used in different morphological variants to represent different entities or concepts in the world. For example, Take a word nail (human anatomy & metallic object) 1, He nailed the loose arm of the chair. 2. He got a box of metallic nails. 3. This nails are growing too fast. 4, He went to manicure to remove his nails.EVENT AND ENTITY RESOLUTION Entity: The process of identifying enttes ie people, organization, location and more. Example: ‘Steve Jobs was the co-founder of Apple, which is headquartered in Cupertino, Event: The actions described with associated entities. Example: Elon Musk announced that Tesla will build @ new factory in Texas, + Identigying and linking references tothe same entity across the text.PREDICATE-ARGUMENT STRUCTURE After identifying the word senses, entities and events we have to identify the participants of the entities in these events. Generally, this process can be defined as the identification of who did what to whom, when, where, why and how. Be Adc Corp. tad tw anu tf Cotte Dat Crys amy mite ne Figure 4-1: A regesenttion of who id what to whom, when, where, why. and howMEANING REPRESENTATION ‘The final process of the semantic interpretation is to build a semantic representation or meaning ‘epresentation that canbe manipulated by algorithms, This process is sometimes called as deep representation, For example, 1. If our player 2 has the ball, then position our player 5 in the midfield, (bower(player 2) (dfplayer 5) posited)‘SYSTEM PARADIGMS + Researches have examined meaning representations and methods to recover from different levels of granularity (the level of detail) and generality (how broad or general the information is). + For many of the potential experimental conditions, no hand-annotated data is available. + Therefore, itis important to get a perspective on the various primary dimensions on which the problem of ‘semantic interpretation has been tackled. ‘The historic approaches which are more prevalent and successful generally fall into 3 categories: 1. SYSTEM ARCHITECTURES (a) Knowledge based: As the name suggests, these systems use a predefined set of rules or knowledge base to obtain a solution to a new problem. (b) Unsupervised: These systems tend to require minimal human intervention to be functional by using existing resources that can be bootstrapped for a particular application or domain. (c) Supervised: These systems require some manual annotation. Typically, researches create feature functions. A model is trained to use these features to predict labels, and then it is applied to unseen data. (¢) Semi-supervised: Manual annotation is very expensive and does not yield enough data. In such instances, researches can automatically expand the dataset on which their models are trained either *2. SCOPE (a) Domain Dependent: These systems are specific to certain domains. (b) Domain independent: These systems are general enough that the techniques can be applicable to multiple domains. 3. COVERAGE (a) Shaliows These systems tend to produce an intermediate representation that can then be converted to ‘one that machine can base its actions on, (b) Deep: These systems usually create a terminal representation that is directly consumed by a machine or application,Word Sense Systems Researchers have explored various system architectures to address the sense disambiguation problem. ‘We can classify these systems into four main categories: 1, Rule based or Knowledge based 2 Supervised 3. Unsupervised 4, Semi-supervised1. Rule-based: -Rule-based systems for word sense disambiguation are among the earliest methods developed to tackle problem of determining the correct meaning of a word based on its context. These systems rely heavily on dictionaries, thesauri, and handcrafted rules, Algorithms and Techniques: {() Lesk Algorithm: -One of the simplest and oldest dictionary-based algorithms. The algorithm assigns the sense of a word that has the most overlap in terms of words with the words in its context. Example: If the word "bank" appears in a context with words like “money” and “deposit,” the financial sense of "bank" is chosen. {il Enhanced Lesk Algorithm: -Banerjee and Pedersen extended the Lesk algorithm to include synonyms, hypernyms (more general terms), hyponyms (more specific terms), and meronyms (part-wwhole relationships). This increases the accuracy of overlap measurement and improves disambiguation performance.(i) Enhanced Lesk Algorithm: Banerjee and Pedersen extended the Lesk algorithm to include synonyms, hypernyms (more general terms), hyponyms (more specific terms), and meronyms (part-whole relationships). “This increases the accuracy of overlap measurement and improves disambiguation performance. (li) Structural Semantic Interconnections (SSI) Proposed by Navigli and Velardi, Constructs semantic graphs using resources like WordNet, domain labels, and annotated corpora.(ii Structural Semantic Interconnections (SS}): “Proposed by Navigli and Velardi, -Constructs semantic graphs using resources like WordNet, domain labels, and annotated corpora. Uses an iterative algorithm to match the semantic graphs of context words with the target word until the best matching sense is identified, Working of Rule-based: hanes caecion ay Ape -2 Dictionary/Thesaurus Matching 3. Weight Computation 4, Sense SelectionSupervised Systems: -Supervised systems for woud sense disambiguation use machine learning to train classifiers on manually annotated datasets. These systems typically perform better than unsupervised methods when tested on annotated data but require significant manual effort for annotation and a predetermined sense -Use various machine learning models, ike Support Vector Machines (SVMs) and Maximum Entropy (MaxEnt) Features Used in Supervised Systems: “Lexical Context This feature comprises the words and lemmas of words occuring inthe entre paragraph, *Parts of Speech: POS tags for words in the contest window. *Local Collocatons: ‘Sequences of nearby words that provide semantic context. ‘Example: For the word "nail" in “He bought a box of nails,” collocation features might include "box.of* and “of mais.” ‘Topic Feature: ‘The broad topic or domain ofthe text can indicate the likely sense of a word.Features Used in Supervised Systems: “Lexical Context: This feature comprises the words and lemmas of words occuring inthe enti paragraph, *Parts of Speech: feStagiber rds tn cst vedon *Local Collocatons: ‘Sequences of nearby words that provide semantic context. Example: For the word “nail” in “He bought a box of nals” colcation features might include "box_of” and “of.nais” *Topic Feature: ‘The broad topic or domain ofthe text can indicate the Bkely sense ofa word. “Additional Rich Features: ove of the Sentence Ate, passive, or semi-passve, Presence of Subyec/Objecr Whether the word has a subject or object.