Combining Statistical, Topical and Syntactic Information for Automatic Term Extraction

Automatic Term Extraction (ATE) is a branch of Natural Language Processing that aims to retrieve domain-specific expressions from academic or popular science documents. Extracted terminology is frequently used as one of preprocessing steps for a large number of other Natural Language Processing tasks, for example, in ontology development, glossary construction, text summarization and data visualisation. In this work, we apply Phrase Mining, Statistical measures, Topic Modelling and Syntactic Parsing in order to build a pipeline to extract terminology from academic and popular science texts. Whereas single-word terms (SWT) are relatively easy to extract, multi-word terms (MWT) still pose a problem for automatic systems. In our research, we pay special attention to multi-word terms extraction quality by using technologies that focus on dealing with multi-word phrases, namely, the ToPMine algorithm for Phrase Mining, C-Value measure for candidate terms scoring and syntactic rules that apply to multi-word candidates. We test our pipeline on the ACL RD-TEC 2.0 dataset, which is publicly available and has been used as a benchmark dataset for Automatic Term Extraction.