Terminological paraphrase extraction from scientific literature based on predicate argument tuples

Abstract

Terminological paraphrases (TPs) are sentences or phrases that express the concepts of terminologies in a different form. Here we propose an effective way to identify and extract TPs from large-scale scientific literature databases. We propose a novel method for effectively retrieving sentences that contain a given terminological concept based on semantic units called predicate-argument tuples. This method enables effective textual similarity computations and minimized errors based on six TP ranking models. For evaluation, we constructed an evaluation collection for the TP recognition task by extracting TPs from a target literature database using the proposed method. Through the two experiments, we learned that scientific literature contain many TPs that could not have been identified so far. Also, the experimental results showed the potential and extensibility of our proposed methods to extract the TPs.

Exploring the Unexplored: Identifying Implicit and Indirect Descriptions of Biomedical Terminologies Based on Multifaceted Weighting Combinations

“…As aforementioned, another line of research tried to identify various types of linguistic expressions apart from noun phrases by converting term definitions as well as all the target texts into predicate argument tuples (PATs) [ 6 ]. A PAT can be defined as a syntactical unit of predicate argument structure (PAS) [ 19 ] that consists of one or two arguments and a grammatical relation (a.k.a.…”

“…In some cases, they think that employing existing standard terms is insufficient to describe their evolved technical concepts that could be partially presented by using them. In addition, even if this case is relatively unusual, some authors do not know the existence of the technical terms representing their verbose expressions [ 6 ].…”

“…In order to retrieve or analyze scholarly information effectively, however, it is vital to tackle this problem first instead of trying to find important but vague textual expressions. As one of the initial obvious trials to resolve the problem above, Choi and Myaeng [ 6 ] introduced a new notion called “ terminological paraphrase ( TP )” to indicate textual expressions denoting technical terms. In addition, they developed 6 ranking (in general, the ranking model involves the computation of index (term and PAT) weights as well as the similarities between the two textual chunk vectors; refer to [ 6 ] for more details) models to retrieve and extract them from biomedical abstracts.…”

“…This paper introduces an enhanced TP extraction system in which multiple term and PAT (predicate argument tuple) weighting schemes can be complementarily combined to extract more hidden TPs from texts. Based on the previously proposed 6 weighting schemes in [ 6 ], the system can apply multiple weighting schemes simultaneously to the TP extraction procedure for more achievement of the diversity of the extracted terminological paraphrases. In other words, given that each scheme could extract different TPs from others, it is reasonable to assume that their combinations could extract more diversified TPs without harming the accuracy of the extraction results [ 10 ].…”

Learning Kernels for Semantic Clustering: A Deep Approach

“…As training data they used a POS annotated corpus of sentences that contain noun phrases. It was trained a binary SVM aimed at both paraphrase detection and multi-word term equivalence assertion (Choi and Myaeng, 2012;Abend et al, 2014). More complex constructions were not considered, but their feature mixure performs very well.…”