2021/11/30 by Avi Chawla, Nidhi Mulay, Chawla, Avi +7 · 1 citation
Computer Science · #Computation and Language (cs.CL) #FOS: Computer and information sciences #Natural Language Processing Techniques #Speech and dialogue systems #Topic Modeling
paper · pdf · doi:10.48550/arxiv.2111.15417
openalex publication_date 2021/11/30 · openalex created_date 2022/05/05 · openalex updated_date 2026/07/28
Recent years of research in Natural Language Processing (NLP) have witnessed dramatic growth in training large models for generating context-aware language representations. In this regard, numerous NLP systems have leveraged the power of neural network-based architectures to incorporate sense information in embeddings, resulting in Contextualized Word Embeddings (CWEs). Despite this progress, the NLP community has not witnessed any significant work performing a comparative study on the contextualization power of such architectures. This paper presents a comparative study and an extensive analysis of nine widely adopted Transformer models. These models are BERT, CTRL, DistilBERT, OpenAI-GPT, OpenAI-GPT2, Transformer-XL, XLNet, ELECTRA, and ALBERT. We evaluate their contextualization power using two lexical sample Word Sense Disambiguation (WSD) tasks, SensEval-2 and SensEval-3. We adopt a simple yet effective approach to WSD that uses a k-Nearest Neighbor (kNN) classification on CWEs. Experimental results show that the proposed techniques also achieve superior results over the current state-of-the-art on both the WSD tasks