Implementing a Reverse Dictionary, based on word definitions, using a Node-Graph Architecture

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 2797–2806, Osaka, Japan, December 11-17 2016. Implementing a Reverse Dictionary, based on word definitions, using a Node-Graph Architecture Sushrut Thorat Center for Mind/Brain Sciences University of Trento Rovereto, TN 38068, Italy sushrut.thorat94@gmail.com Varad Choudhari Department of Computer Science Rajarambapu Institute of Technology Islampur, MH 415414, India varad.choudhari@gmail.com Abstract In this paper, we outline an approach to build graph-based reverse dictionaries using word defi- nitions. A reverse dictionary takes a phrase as an input and outputs a list of words semantically similar to that phrase. It is a solution to the Tip-of-the-Tongue problem. We use a distance-based similarity measure, computed on a graph, to assess the similarity between a word and the input phrase. We compare the performance of our approach with the Onelook Reverse Dictionary and a distributional semantics method based on word2vec, and show that our approach is much better than the distributional semantics method, and as good as Onelook, on a 3k lexicon. This simple approach sets a new performance baseline for reverse dictionaries.1 1 Introduction A forward dictionary (FD) maps words to their definitions. A reverse dictionary (RD) (Sierra, 2000), also known as an inverse dictionary, or search-by-concept dictionary (Calvo et al., 2016), maps phrases to single words that approximate the meaning of those phrases. In the Oxford Learner’s Dictionary2, one definition of ‘brother’ is ‘a boy or man who has the same mother and father as another person’. A reverse dictionary will map not only this phrase to ‘brother’, but also phrases such as ‘son of my parents’. A reverse dictionary is primarily a solution to the Tip of the Tongue problem (Schwartz and Metcalfe, 2011) which regularly plagues people when they want to articulate their thoughts. It can also be used in the treatment of word selection anomic aphasia (Rohrer et al., 2008), a neurological disorder in which patients can identify objects and understand semantic properties, but cannot name the object or produce one word to describe the concept. Popular languages let us create a multitude of phrases from a finite number of words. A static database of all possible phrases is unbound, if not infinite (Ziff, 1974). We need to dynamically compute the output word(s) from the input phrase. To map a phrase to a word, we have to compute the meanings of the phrase and the word (Fromkin et al., 2011). The principle of compositionality states that the meaning of an expression is composed of the meaning of its parts and the way they are combined structurally. The most basic parts, words, can be defined in terms of word definitions, references to objects, or lexical relations and hierarchies. Computing the meaning of a phrase requires constructing the constituent tree and recognising the relationship between the constituents, which is a complex, open problem. Compositional Distributional Semantic Models have been used towards computing the meaning of a phrase, with partial success (Baroni, 2013; Erk, 2012). Recurrent neural networks show promise in …

Content truncated to 500 words