Abstract
This article examines the basis of Natural Language Understanding of transformer based language models, such as BERT. It does this through a case study on idiom token classification. We use idiom token identification as a basis for our analysis because of the variety of information types that have previously been explored in the literature for this task, including: topic, lexical, and syntactic features. This variety of relevant information types means that the task of idiom token identification enables us to explore the forms of linguistic information that a BERT language model captures and encodes in its representations. The core of this article presents three experiments. The first experiment analyzes the effectiveness of BERT sentence embeddings for creating a general idiom token identification model and the results indicate that the BERT sentence embeddings outperform Skip-Thought. In the second and third experiment we use the game theory concept of Shapley Values to rank the usefulness of individual idiomatic expressions for model training and use this ranking to analyse the type of information that the model finds useful. We find that a combination of idiom-intrinsic and topic-based properties contribute to an expression's usefulness in idiom token identification. Overall our results indicate that BERT efficiently encodes a variety of information from topic, through lexical and syntactic information. Based on these results we argue that notwithstanding recent criticisms of language model based semantics, the ability of BERT to efficiently encode a variety of linguistic information types does represent a significant step forward in natural language understanding.
Highlights
There is a large body of existing work that is focused on probing BERT representations
We confirm that distributed representations are suitable for this task, but we go beyond improving the state of the art and perform a variety of experiments designed to investigate what types of information BERT uses for the task, and in this way we explore what information is useful for creating a general idiom token identification model
Contributions: (a) we report a new state of the art for general idiom token identification, using BERT sentence embeddings; (b) we demonstrate that the game theory concept of Shapley values provides a basis for analysing idiomatic usage; and (c) we explain the strong performance of BERT embeddings in terms of their ability to model idiom-intrinsic and topic-based properties
Summary
The last 5 years of natural language processing research has been a record of remarkable progress in the state-of-the-art across a range of tasks (Wang et al, 2019). In the intervening period more rich and advanced embedding techniques, such as BERT (Devlin et al, 2018), have since been developed and report vastly improved performance on many NLP tasks It is worth investigating the application of these new embedding models to the problem of general idiom token identification. We build on the work of Salton et al (2016) and look at how well a contemporary sentence embedding model performs on the task of general idiom token identification, on the example of English Verb-Noun Idiomatic Combinations (VNICs). Contributions: (a) we report a new state of the art for general idiom token identification, using BERT sentence embeddings; (b) we demonstrate that the game theory concept of Shapley values provides a basis for analysing idiomatic usage; and (c) we explain the strong performance of BERT embeddings in terms of their ability to model idiom-intrinsic and topic-based properties
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have