Looking inside Noun Compounds: Unsupervised Prepositional and Free Paraphrasing

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 4313–4323 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 4313 Looking inside Noun Compounds: Unsupervised Prepositional and Free Paraphrasing Girishkumar Ponkiya†, Rudra Murthy‡, Pushpak Bhattacharyya† and Girish K Palshikar⋆ †Indian Institute of Technology Bombay, Mumbai ‡IBM Research ⋆TCS Research, Tata Consultancy Services, Pune {girishp,pb,rudra}@cse.iitb.ac.in, gk.palshikar@tcs.com Abstract A noun compound is a sequence of contigu- ous nouns that acts as a single noun, although the predicate denoting the semantic relation be- tween its components is dropped. Noun Com- pound Interpretation is the task of uncover- ing the relation, in the form of a preposition or a free paraphrase. Prepositional paraphras- ing refers to the use of preposition to explain the semantic relation, whereas free paraphras- ing refers to invoking an appropriate predi- cate denoting the semantic relation. In this paper, we propose an unsupervised methodol- ogy for these two types of paraphrasing. We use pre-trained contextualized language mod- els to uncover the ‘missing’ words (preposition or predicate). These language models are usu- ally trained to uncover the missing word/words in a given input sentence. Our approach uses templates to prepare the input sequence for the language model. The template uses a special token to indicate the missing predicate. As the model has already been pre-trained to uncover a missing word (or a sequence of words), we exploit it to predict missing words for the input sequence. Our experiments using four datasets show that our unsupervised approach (a) performs com- parably to supervised approaches for preposi- tional paraphrasing, and (b) outperforms super- vised approaches for free paraphrasing. Para- phrasing (prepositional or free) using our un- supervised approach is potentially helpful for NLP tasks like machine translation and infor- mation extraction. 1 Introduction Noun compounds- contiguous sequences of nouns- are common linguistic constructs. A compound is called compositional if the meaning of the com- pounds can be derived from the meaning of its components. The component nouns are related through a semantic relation that is constituents de- pendent. For instance, ‘student protest’ and ‘univer- sity protest’ are protests. However, the student(s) are AGENT (doer of an event), whereas university is LOCATION of the protest. The task of identifying such relations between the components of a noun compound is called noun compound interpretation (NCI). Such interpreta- tion can help a wide variety of NLP tasks, like machine translation (Baldwin and Tanaka, 2004; Paul et al., 2010; Balyan and Chatterjee, 2015), question answering (Ahn et al., 2005), text entail- ment (Nakov, 2013), and semantic parsing (Tratz, 2011). For instance, to translate the English noun compound ‘cow milk’ to Hindi, a machine transla- tion system needs to generate the postposition kA (of) in addition to translating the individual nouns. The correct translation of the compound is ‘g¯aya ka d¯udha’ (lit. ‘cow -of milk’; ‘milk of cow’). Without understanding the underlying relation, a machine translation system might fail. Interpretation via abstract labels (representing se- mantic relations) is popular in the literature. Given a noun compound, …

Content truncated to 500 words