Semantic Domains in Computational Linguistics

Ambiguity and variability are two basic and pervasive phenomena char-acterizing lexical semantics. Their pervasiveness imposes the developmentof Natural Language Processing systems provided by computational modelsto represent them in the application domain. In this work we introducea computational model for lexical semantics based on Semantic Domains.This concept is inspired by the “Theory of Semantic Fields”, proposed instructural linguistics to explain lexical semantics. The main property ofSemantic Domains is lexical coherence, i.e. the property of domain relatedwords to co-occur in texts. This allows us to define automatic acquisitionprocedures for Domain Models from corpora, and the acquired models pro-vide a shallow representation for lexical ambiguity and variability. DomainModels have been used to define a similarity metric among texts and termsin the Domain Space, where second order relations are reflected. Topicsimilarity estimation is at the basis of text comprehension, allowing us todefine a very general domain driven methodology. The basic argument weput forward to support our domain-based approach is that the informationprovided by the Domain Models can be profitably used to boost the perfor-mances of supervised Natural Language Processing systems for many tasks.In fact, Semantic Domains allows us to extract domain features for texts,terms and concepts. The obtained index, adopted by the Domain Kernel toestimate topic similarity, preserves the original information while reduc-ing the dimensionality of the feature space. The Domain Kernel is used to define a semi-supervised learning algorithm for Text Categorization thatachieves the state-of-the-art results while decreasing by one order the quan-tity of labeled texts required for learning. We also apply Domain Modelsto approach a Term Categorization task, improving noticeably the predic-tion accuracy on domain specific terms. The property of the Domain Spaceto represent together terms and texts allows us to define an IntensionalLearning schema for Text Categorization, in which categories are describedby means of discriminative words instead of labeled examples, achievingperformances close to the human agreement. Then we investigate the roleof domain information in Word Sense Disambiguation, developing both anunsupervised and a supervised approaches that strongly rely on the notionof Semantic Domain. The former is based on the lexical resource Word-Net Domains and the latter exploits both sense tagged and unlabeled datato model the relevant domain distinctions among word senses. Our super-vised approach improves the state-of-the-art performance in many tasks fordifferent languages, while reducing appreciably the amount of sense taggeddata required for learning. Finally, we present a multilingual lexical ac-quisition procedure to obtain Multilingual Domain Models from compara-ble corpora. We exploit such models to approach a Cross Language TextCategorization task, achieving very promising results largely surpassing abaseline. KeywordsLexical Semantics, Word Sense Disambiguation, Text Categorization, Mul-tilinguality, Kernel Methods