L2F/INESC-ID at SemEval-2019 Task 2: Unsupervised Lexical Semantic Frame Induction using Contextualized Word Representations

Abstract

Building large datasets annotated with semantic information, such as FrameNet, is an expensive process. Consequently, such resources are unavailable for many languages and specific domains. This problem can be alleviated by using unsupervised approaches to induce the frames evoked by a collection of documents. That is the objective of the second task of SemEval 2019, which comprises three subtasks: clustering of verbs that evoke the same frame and clustering of arguments into both frame-specific slots and semantic roles. We approach all the subtasks by applying a graph clustering algorithm on contextualized embedding representations of the verbs and arguments. Using such representations is appropriate in the context of this task, since they provide cues for word-sense disambiguation. Thus, they can be used to identify different frames evoked by the same words. Using this approach we were able to outperform all of the baselines reported for the task on the test set in terms of Purity F 1 , as well as in terms of BCubed F 1 in most cases.

SemEval-2019 Task 2: Unsupervised Lexical Frame Induction

“…Task B.2 employed hand crafted features, a method to encode syntactic information, and again an agglomerative clustering method. Ribeiro et al (2019) also reported results for all subtasks using similar techniques to those reported in the other two submitted papers. Ribeiro et al (2019) used the bidirectional neural language model BERT, which also used.…”

“…Ribeiro et al (2019) also reported results for all subtasks using similar techniques to those reported in the other two submitted papers. Ribeiro et al (2019) used the bidirectional neural language model BERT, which also used. Task A employed contextualized word representations proposed in (Ustalov et al, 2018), and Biemann's clustering algorithm (Biemann, 2006).…”

“…Task A employed contextualized word representations proposed in (Ustalov et al, 2018), and Biemann's clustering algorithm (Biemann, 2006). Compared to the two other systems, Ribeiro et al (2019) exploited input structures, weighted them, and used them elegantly in its algorithm. With the same method but different hyper-parameters for B.2 along with combining results from Task A, Ribeiro et al (2019) offered a solution to B.1.…”

“…Compared to the two other systems, Ribeiro et al (2019) exploited input structures, weighted them, and used them elegantly in its algorithm. With the same method but different hyper-parameters for B.2 along with combining results from Task A, Ribeiro et al (2019) offered a solution to B.1. Table 2 reports the BCF scores for system submissions along with a baseline for each task.…”

How does BERT capture semantics? A closer look at polysemous words

“…While this allows to compare the average displacement due to semantic change across words, it does not give us a good sense of the overall structure of word vectors. In addition, the embedding space produced by BERT has been analyzed in terms of syntactic features, such as parse-trees (Coenen et al, 2019;Jawahar et al, 2019), part-of-speech, verbs and arguments (Shi et al, 2019;Ribeiro et al, 2019).…”