Building Concept Graphs from Monolingual Dictionary Entries

Building Concept Graphs from Monolingual Dictionary Entries G´abor Recski Research Institute for Linguistics Hungarian Academy of Sciences Bencz´ur u. 33 1068 Budapest, Hungary recski@mokk.bme.hu Abstract We present the dict to 4lang tool for processing entries of three monolingual dictionaries of English and mapping definitions to concept graphs following the 4lang principles of semantic representation introduced by (Kornai, 2010). 4lang representations are domain- and language-independent, and make use of only a very limited set of primitives to encode the meaning of all utterances. Our pipeline relies on the Stanford Dependency Parser for syntactic analysis, the dep to 4lang module then builds directed graphs of concepts based on dependency relations between words in each definition. Several issues are handled by construction-specific rules that are applied to the output of dep to 4lang. Manual evaluation suggests that ca. 75% of graphs built from the Longman Dictionary are either entirely correct or contain only minor errors. dict to 4lang is available under an MIT license as part of the 4lang library and has been used successfully in measuring Semantic Textual Similarity (Recski and ´Acs, 2015). An interactive demo of core 4lang functionalities is available at http://4lang.hlt.bme.hu. Keywords: semantics, lexicon, knowledge representation 1. Introduction We present the dict to 4lang tool for automati- cally building graphs in the style of the 4lang concept dictionary (Kornai et al., 2015) using entries from var- ious explanatory dictionaries of English. Our pipeline maps the output of a state-of-the-art dependency parser to subgraphs over concept nodes corresponding to the words of each definition. The resulting graphs have been used successfully for measuring semantic similar- ity (Recski and ´Acs, 2015) and also allows us to map virtually all English text to the 4lang representation. The full pipeline is available for download under an MIT license at http://github.com/kornai/4lang, graphs created from three major dictionaries (Long- man, Collins, en.wiktionary) are also freely accessible at http://people.mokk.bme.hu/ ˜recski/4lang/graphs. An online demo of core 4lang functionalities is available at http://4lang.hlt.bme.hu. This paper is structured as follows: Section 2 provides a short introduction to graph-based representations of mean- ing, followed by an overview of the 4lang formalism and its basic principles of semantic representation in Section 3. Section 4 presents the dict to 4lang tool, including the mapping from Stanford dependencies to 4lang configura- tions, and reports some figures characterizing the graphs created from each dataset. Section 5 mentions some errors typical in the output and discusses possible solutions, Sec- tion 6 presents the results of manual evaluation. Section 7 presents a method for reducing the vocabulary of newly built 4lang-graphs by replacing nodes with their defini- tions. Finally, Section 8 discusses some applications of the pipeline. 2. Background Directed graphs of concepts have been used to represent the meaning of words, phrases, and utterances by several influential systems in the second half of the 20th century, including the Semantic Memory Model of (Quillian, 1968) or the KL-ONE system (Brachman and Levesque, 1985) and its descendants (Moser, 1983; Brachman et al., 1983). More recently, Abstract Meaning …

Content truncated to 500 words