Probing for Constituency Structure in Neural Language Models
This work addresses the problem of understanding internal representations in language models for researchers in NLP, providing incremental insights into the separation of syntactic and semantic knowledge.
The paper investigates whether neural language models implicitly learn syntactic constituency structure, finding that four pretrained transformer models achieve high accuracy on probing tasks even with semantically manipulated data, indicating that syntactic knowledge is learned and separable from semantic information.
In this paper, we investigate to which extent contextual neural language models (LMs) implicitly learn syntactic structure. More concretely, we focus on constituent structure as represented in the Penn Treebank (PTB). Using standard probing techniques based on diagnostic classifiers, we assess the accuracy of representing constituents of different categories within the neuron activations of a LM such as RoBERTa. In order to make sure that our probe focuses on syntactic knowledge and not on implicit semantic generalizations, we also experiment on a PTB version that is obtained by randomly replacing constituents with each other while keeping syntactic structure, i.e., a semantically ill-formed but syntactically well-formed version of the PTB. We find that 4 pretrained transfomer LMs obtain high performance on our probing tasks even on manipulated data, suggesting that semantic and syntactic knowledge in their representations can be separated and that constituency information is in fact learned by the LM. Moreover, we show that a complete constituency tree can be linearly separated from LM representations.