Treat us like the sequences we are: Prepositional Paraphrasing of Noun Compounds using LSTM

Proceedings of the 27th International Conference on Computational Linguistics, pages 1827–1836 Santa Fe, New Mexico, USA, August 20-26, 2018. 1827 Treat us like the sequences we are: Prepositional Paraphrasing of Noun Compounds using LSTM Girishkumar Ponkiya†, Kevin Patel†, Pushpak Bhattacharyya† and Girish K Palshikar‡ †Indian Institute of Technology Bombay, Mumbai ‡TCS Research, Tata Consultancy Services, Pune {girishp,kevin.patel,pb}@cse.iitb.ac.in, gk.palshikar@tcs.com Abstract Interpreting noun compounds is a challenging task. It involves uncovering the underlying pred- icate which is dropped in the formation of the compound. In most cases, this predicate is of the form VERB+PREP. It has been observed that uncovering the preposition is a significant step towards uncovering the predicate. In this paper, we attempt to paraphrase noun compounds using prepositions. We consider noun compounds and their corresponding prepositional paraphrases as parallelly aligned sequences of words. This enables us to adapt different architectures from cross-lingual embedding literature. We choose the architecture where we create representations of both noun compound (source sequence) and its corresponding prepositional paraphrase (target sequence), such that their sim- ilarity is high. We use LSTMs to learn these representations. We use these representations to decide the correct preposition. Our experiments show that this approach performs considerably well on different datasets of noun compounds that are manually annotated with prepositions. 1 Introduction A noun compound is a sequence of two or more nouns which have a well-defined meaning when writ- ten together. For example, orange juice, colon cancer, research paper submission, paper submission deadline, etc. The fact that “juice is made from orange” is hidden in orange juice. Noun compound interpretation deals with uncovering such hidden relations. Noun compounds are usually interpreted in two ways: labelling and paraphrasing. Labelling involves assigning a semantic relation to a noun compound e.g., student protest: AGENT, orange juice: MADEOF, etc.. These relations come from a set of a predefined taxonomy of semantic relations (Lauer, 1995; Warren, 1978; Barker and Szpakowicz, 1998; Girju et al., 2003; Tratz and Hovy, 2010; Ponkiya et al., 2018). Such detailed, fine-grained information can be useful for downstream tasks such as machine translation (Baldwin and Tanaka, 2004; Balyan and Chatterjee, 2015), question answering (Ahn et al., 2005), text entailment (Nakov, 2013), etc. Unfortunately, there is a lack of standard taxonomy. There is no consensus on which set of labels should be uniformly used. On the other hand, paraphrasing involves rewriting the noun compound as a paraphrase which conveys its meaning explicitly, e.g., orange juice: “juice made from orange” or “juice with orange flavour”. Generic paraphrasing has been relatively less pursued (Butnariu et al., 2009; Hendrickx et al., 2013). Prepositional paraphrasing – paraphrasing using only prepositions, e.g., orange juice: “juice of or- ange” – is a simpler version of generic paraphrasing. The advantage of prepositional paraphrasing as compared to labelling is that the set of prepositions is finite, limited and pre-defined. However, the shortcoming is that the information is too coarse-grained for downstream tasks. Prepositional paraphrasing of noun compounds is a useful subtask to solve. Our preliminary analysis reveals a strong …

Content truncated to 500 words