A Learning Methodology for English Grammar-based Ontologies

• Main Authors: Jia-Cing Ruan, 阮家慶
• Other Authors: Daniel J. Buehrer
• Format: Others
• Language: en_US
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/92741307771641681808

博士 === 國立中正大學 === 語言學研究所 === 103 === This preliminary study tries to introduce a new concept of grammar-based ontology through building up an ontology with patterns composed by conjoining syntactic and semantic structures. The background refers to Hanks’ (2004, 2013) thought that any word, especially a polysemous word, does not have its clear meaning unless it occurs in a context. Consequently, phraseological patterns and collocations enable the possibilities to disambiguate word meaning. Since WordNet (Fellbaum, 1998) is unreliable for not providing contrastive analyses of word senses (Ide and Wilks, 2006), using patterns instead of words to achieve the same goal seems more reasonable. With the Corpus Pattern Analysis (CPA) (Hanks, 2004), Hanks has realized a pattern dictionary for English verbs. This dictionary contains pattern information composed within a corpus, and is complementary to Construction Grammar (Goldberg, 1995) or FrameNet (Baker, Fillmore and Lowe, 1998). Hanks indicates Construction Grammar studies need corpus data as evidence, instead of using fictitious examples. FrameNet provides numerous frame structures while the Corpus Pattern Analysis focuses on systematic analyses between patterns and verbs. Not only the Corpus Pattern Analysis and its one of the foundations, Theory of Norms and Exploitations (TNE) (Hanks, 2013), but also Construction Grammar, Construction Morphology (Booij, 2010) and Distributed Morphology (Halle and Marantz, 1993; Harley and Noyer, 1999; Embick and Noyer, 2007) admit the existence of meanings of patterns or constructions themselves independently. Thus, an ontology ought not to be only lexicon-based, but grammar-based or pattern-based. The needs for a grammar-based ontology include three dimensions. First, according to Construction Grammar, whatever words, phrases or sentences have their form-meaning pairs, which exhibits that a meaning corresponding to a form to any grammatical (syntactic or semantic) element itself exists. The grammatical meanings which show implicit abstract knowledge of the world have long been ignored in building up an ontology. Second, according to Construction Morphology or Distributed Morphology, the meaning of a word formation (especially compound or complex words) is predetermined by its syntactic pattern or construction. This indicates ‘patterns’ contributes to human knowledge. Third, without the grammatical knowledge (whether in awareness or not) a language may not be used properly, which points out that traditional lexicon-based ontologies fail to reflect the knowledge of real language usage, and thus further fail to reflect meanings from language usage interfaced to the objects of the world. Through implementing the queries of a grammar-based ontology and word-sketch-like systems made from a grammar-based ontology, either the systems or the query results are incapable to be reproduced by traditional ontologies, which implies the need of a grammar-based ontology. Furthermore, the grammar-based ontology has been applied to try doing the machine sentence generation, which cannot be handled by traditional ontologies, either. Additionally, a grammar-based ontology can be applied to study typological issues in linguistics, and can also be applied to analyze reference books of learning languages in educational purposes. The evaluation of a grammar-based ontology is much more straightforward due to the mixture of several systems with clear precisions, compared to traditional ontologies. However, the precision or quality of a grammar-based ontology is changed if the tags are changed in constituency parse results, dependency parse results or semantic tagging results. Applying different tagsets result in different performance. The limitations of the grammar-based ontology are stated below. In the design of a grammar-based ontology proposed in the present dissertation, the pragmatic information cannot be handled due to the lack of non-taxonomic relations. Using English as an example, the coordinate and subordinate conjunctions play important roles to be the non-taxonomic relations between two concepts. Second, the grammar-based ontology does not provide much information for evaluating CPA.