This paper describes an AI that uses construction grammar (CG)—a means of knowledge representation for deep understanding of text. The proposed improvements aim at more versatility of the text form and meaning knowledge structure, as well as for intelligent choosing among possible parses. Along with the improvements, computational CG techniques that form the implementation basis are explained. Evaluation experiments utilize a Winograd schema (WS)—a major test for AI—dataset and compare the implementation with state-of-the-art ones. Results have demonstrated that compared with such techniques as deep learning, the proposed CG approach has a higher potential for the task of anaphora resolution involving deep understanding of the natural language.
This paper describes using a finite-state automaton (FSA) to retrieve Japanese TV guide text. The proposed FSA application can be considered novel due to lack of research on the subject. The automaton has been implemented for matching and extracting all possible combinations of search query words in all possible word orders that may be present in the TV guide text. This implementation also sorts the extraction results by analyzing word semantic features (such as “being an object” or “being a property of an object”). The present paper also proposes a search system using the above implementation and compares it with a baseline system that matches query words (of multi-word queries) in exactly the same and exactly the opposite word orders only. Both systems use morphological parsing and apply a stop list to the query. A multi-parameter evaluation has shown advantages of the proposed system over the baseline one.
Abstract-This paper describes a system for searching the Webbased Japanese TV program guide. The system features using morphological parsing and part-of-speech analysis to locate words with nominal and attributive semantic features in the query. Such words are matched mandatorily when searching the TV program guide text, while other words are matched optionally. Moreover, certain words and morphemes are removed from the query as they are considered to have little semantic value. The system checks every query against a stop list of such words and morphemes. Other processing methods, e.g. reversing the search phrase word order and allowing "zero or more words" between the search target words, are also utilized. The present paper uses TV guide search examples to demonstrate how the proposed method can improve Japanese TV program data search results. The paper also contains a few ideas about ways the method could be used for other languages.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.