BERTatDE at SemEval-2020 Task 6: Extracting Term-definition Pairs in Free Text Using Pre-trained Model

Proceedings of the 14th International Workshop on Semantic Evaluation, pages 690–696 Barcelona, Spain (Online), December 12, 2020. 690 BERTatDE at SemEval-2020 Task 6: Extracting term-definition pairs in free text using pre-trained model Huihui Zhang and Feiliang Ren Computer science and Engineering, Northeastern University Key Laboratory of Data Analytics and Optimization for Smart Industry (Northeastern University),Ministry of Education, Shenyang, China renfeiliang@mail.neu.edu.cn Abstract Definition extraction is an important task in Nature Language Processing, and it is used to identify the terms and definitions related to terms. The task contains sentence classification task (i.e., classify whether it contains definition) and sequence labeling task (i.e., find the boundary of terms and definitions). The paper describes our system BERTatDE in sentence classification task (subtask 1) and sequence labeling task (subtask 2) in the definition extraction (SemEval-2020 Task 6). We use BERT to solve the multi-domain problems including the uncertainty of term boundary that is, different areas have different ways to definite the domain related terms. We use BERT, BiLSTM and attention in subtask 1 and our best result achieved 79.71% in F1 and the eighteenth place in subtask 1. For the subtask 2, we use BERT, BiLSTM and CRF to sequence labeling, and achieve 40.73% in Macro-averaged F1. 1 Introduction Definition extraction refers to extracting term-definition pairs from real texts. It is of great significance in technology and application development of information processing. Because it can be applied to many scenarios. For instance, it can be used to construct domain dictionary, knowledge graph and automatic question answering system, etc. And it also can improve the efficiency of search engine to a certain extent. As we all known, nature language is always complex, for example, polysemy is a pervasive phenomenon for some words. There are two complicate aspects. First, term-definition pairs appears in different sentences. Second, there are lack of explicit definitions in some sentences. We can’t find obvious sign of definable words, like is, means, is defined as, etc. Considering these real situations, the DEFT corpus (Spala et al., 2019) collects some data that can reflect the reality. SemEval-2020 task 6 (Spala et al., 2020) and it is called by DeftEval for short. DeftEval sets up three subtasks on the DEFT corpus. Subtask 1 is a sentence classification task. For a giving sentence, subtask 1 is used to classify whether it contains a definition. If the definition is included, the corresponding sentence is marked as ‘1’. Subtask 2 is a sequence labeling task. For each words including punctuation in a sentence, subtask 2 is used to label them with BIO tags according to the corpus’ tag specification. Subtask 3 is a relation classification task. Given the tag sequence labels, subtask 3 is used to label the relations between each tags according to the corpus’ relation specifications. In the DeftEval, there are seven fields, the proportion of these fields are different. And different fields have different terms. This will bring some challenges such as the boundary of terms is uncertain. As each domain has different ways …

Content truncated to 500 words