Introducing Orthogonal Constraint in Structural Probes

TL;DR

This paper proposes a novel orthogonal constraint for structural probing of NLP models, improving interpretability and robustness by decomposing linear projections into rotations and scalings, and evaluating on syntactic and lexical tasks.

Contribution

It introduces an orthogonal constraint in structural probing, enabling better separation of linguistic information and reducing vulnerability to memorization.

Findings

Orthogonal constraint improves interpretability of embeddings.

Probes effectively separate lexical and syntactic information.

Method shows robustness against memorization in representations.

Abstract

With the recent success of pre-trained models in NLP, a significant focus was put on interpreting their representations. One of the most prominent approaches is structural probing (Hewitt and Manning, 2019), where a linear projection of word embeddings is performed in order to approximate the topology of dependency structures. In this work, we introduce a new type of structural probing, where the linear projection is decomposed into 1. isomorphic space rotation; 2. linear scaling that identifies and scales the most relevant dimensions. In addition to syntactic dependency, we evaluate our method on novel tasks (lexical hypernymy and position in a sentence). We jointly train the probes for multiple tasks and experimentally show that lexical and syntactic information is separated in the representations. Moreover, the orthogonal constraint makes the Structural Probes less vulnerable to memorization.