We present a setup for training, evaluating and interpreting neural language models, that uses artificial, language-like data. The data is generated using a massive probabilistic grammar (based on state-split PCFGs), that is itself derived from a large natural language corpus, but also provides us complete control over the generative process. We describe and release both grammar and corpus, and test for the naturalness of our generated data. This approach allows us to define closed-form expressions to efficiently compute exact lower bounds on obtainable perplexity using both causal and masked language modelling. Our results show striking differences between neural language modelling architectures and training objectives in how closely they allow approximating the lower bound on perplexity. Our approach also allows us to directly compare learned representations to symbolic rules in the underlying source. We experiment with various techniques for interpreting model behaviour and learning dynamics. With access to the underlying true source, our results show striking differences and outcomes in learning dynamics between different classes of words.
Language models (LMs) exhibit and amplify many types of undesirable biases learned from the training data, including gender bias. However, we lack tools for effectively and efficiently changing this behavior without hurting general language modeling performance. In this paper, we study three methods for identifying causal relations between LM components and particular output: causal mediation analysis, automated circuit discovery and our novel, efficient method called DiffMask+ based on differential masking. We apply the methods to GPT-2 small and the problem of gender bias, and use the discovered sets of components to perform parameter-efficient fine-tuning for bias mitigation. Our results show significant overlap in the identified components (despite huge differences in the computational requirements of the methods) as well as success in mitigating gender bias, with less damage to general language modeling compared to full model fine-tuning. However, our work also underscores the difficulty of defining and measuring bias, and the sensitivity of causal discovery procedures to dataset choice. We hope our work can contribute to more attention for dataset development, and lead to more effective mitigation strategies for other types of bias.
Transformers have become a key architecture in speech processing, but our understanding of how they build up representations of acoustic and linguistic structure is limited. In this study, we address this gap by investigating how measures of 'context-mixing' developed for text models can be adapted and applied to models of spoken language. We identify a linguistic phenomenon that is ideal for such a case study: homophony in French (e.g. livre vs livres), where a speech recognition model has to attend to syntactic cues such as determiners and pronouns in order to disambiguate spoken words with identical pronunciations and transcribe them while respecting grammatical agreement. We perform a series of controlled experiments and probing analyses on Transformer-based speech models. Our findings reveal that representations in encoder-only models effectively incorporate these cues to identify the correct transcription, whereas encoders in encoder-decoder models mainly relegate the task of capturing contextual dependencies to decoder modules.
In recent years, many interpretability methods have been proposed to help interpret the internal states of Transformer-models, at different levels of precision and complexity. Here, to analyze encoder-decoder Transformers, we propose a simple, new method: DecoderLens. Inspired by the LogitLens (for decoder-only Transformers), this method involves allowing the decoder to cross-attend representations of intermediate encoder layers instead of using the final encoder output, as is normally done in encoder-decoder models. The method thus maps previously uninterpretable vector representations to human-interpretable sequences of words or symbols. We report results from the DecoderLens applied to models trained on question answering, logical reasoning, speech recognition and machine translation. The DecoderLens reveals several specific subtasks that are solved at low or intermediate layers, shedding new light on the information flow inside the encoder component of this important class of models.
We study feature interactions in the context of feature attribution methods for post-hoc interpretability. In interpretability research, getting to grips with feature interactions is increasingly recognised as an important challenge, because interacting features are key to the success of neural networks. Feature interactions allow a model to build up hierarchical representations for its input, and might provide an ideal starting point for the investigation into linguistic structure in language models. However, uncovering the exact role that these interactions play is also difficult, and a diverse range of interaction attribution methods has been proposed. In this paper, we focus on the question which of these methods most faithfully reflects the inner workings of the target models. We work out a grey box methodology, in which we train models to perfection on a formal language classification task, using PCFGs. We show that under specific configurations, some methods are indeed able to uncover the grammatical rules acquired by a model. Based on these findings we extend our evaluation to a case study on language models, providing novel insights into the linguistic structure that these models have acquired.
Self-attention weights and their transformed variants have been the main source of information for analyzing token-to-token interactions in Transformer-based models. But despite their ease of interpretation, these weights are not faithful to the models' decisions as they are only one part of an encoder, and other components in the encoder layer can have considerable impact on information mixing in the output representations. In this work, by expanding the scope of analysis to the whole encoder block, we propose Value Zeroing, a novel context mixing score customized for Transformers that provides us with a deeper understanding of how information is mixed at each encoder layer. We demonstrate the superiority of our context mixing score over other analysis methods through a series of complementary evaluations with different viewpoints based on linguistically informed rationales, probing, and faithfulness analysis.
As Natural Language Processing (NLP) technology rapidly develops and spreads into daily life, it becomes crucial to anticipate how its use could harm people. However, our ways of assessing the biases of NLP models have not kept up. While especially the detection of English gender bias in such models has enjoyed increasing research attention, many of the measures face serious problems, as it is often unclear what they actually measure and how much they are subject to measurement error. In this paper, we provide an interdisciplinary approach to discussing the issue of NLP model bias by adopting the lens of psychometrics -- a field specialized in the measurement of concepts like bias that are not directly observable. We pair an introduction of relevant psychometric concepts with a discussion of how they could be used to evaluate and improve bias measures. We also argue that adopting psychometric vocabulary and methodology can make NLP bias research more efficient and transparent.
Detecting and mitigating harmful biases in modern language models are widely recognized as crucial, open problems. In this paper, we take a step back and investigate how language models come to be biased in the first place. We use a relatively small language model, using the LSTM architecture trained on an English Wikipedia corpus. With full access to the data and to the model parameters as they change during every step while training, we can map in detail how the representation of gender develops, what patterns in the dataset drive this, and how the model's internal state relates to the bias in a downstream task (semantic textual similarity). We find that the representation of gender is dynamic and identify different phases during training. Furthermore, we show that gender information is represented increasingly locally in the input embeddings of the model and that, as a consequence, debiasing these can be effective in reducing the downstream bias. Monitoring the training dynamics, allows us to detect an asymmetry in how the female and male gender are represented in the input embeddings. This is important, as it may cause naive mitigation strategies to introduce new undesirable biases. We discuss the relevance of the findings for mitigation strategies more generally and the prospects of generalizing our methods to larger language models, the Transformer architecture, other languages and other undesirable biases.
We investigate the extent to which modern, neural language models are susceptible to syntactic priming, the phenomenon where the syntactic structure of a sentence makes the same structure more probable in a follow-up sentence. We explore how priming can be used to study the nature of the syntactic knowledge acquired by these models. We introduce a novel metric and release Prime-LM, a large corpus where we control for various linguistic factors which interact with priming strength. We find that recent large Transformer models indeed show evidence of syntactic priming, but also that the syntactic generalisations learned by these models are to some extent modulated by semantic information. We report surprisingly strong priming effects when priming with multiple sentences, each with different words and meaning but with identical syntactic structure. We conclude that the syntactic priming paradigm is a highly useful, additional tool for gaining insights into the capacities of language models.