Representations from large language models (LLMs) are known to be dominated by a small subset of dimensions with exceedingly high variance. Previous works have argued that although ablating these outlier dimensions in LLM representations hurts downstream performance, outlier dimensions are detrimental to the representational quality of embeddings. In this study, we investigate how fine-tuning impacts outlier dimensions and show that 1) outlier dimensions that occur in pre-training persist in fine-tuned models and 2) a single outlier dimension can complete downstream tasks with a minimal error rate. Our results suggest that outlier dimensions can encode crucial task-specific knowledge and that the value of a representation in a single outlier dimension drives downstream model decisions.

Recent work has shown that infusing layout features into language models (LMs) improves processing of visually-rich documents such as scientific papers. Layout-infused LMs are often evaluated on documents with familiar layout features (e.g., papers from the same publisher), but in practice models encounter documents with unfamiliar distributions of layout features, such as new combinations of text sizes and styles, or new spatial configurations of textual elements. In this work we test whether layout-infused LMs are robust to layout distribution shifts. As a case study we use the task of scientific document structure recovery, segmenting a scientific paper into its structural categories (e.g., “title”, “caption”, “reference”). To emulate distribution shifts that occur in practice we re-partition the GROTOAP2 dataset. We find that under layout distribution shifts model performance degrades by up to 20 F1. Simple training strategies, such as increasing training diversity, can reduce this degradation by over 35% relative F1; however, models fail to reach in-distribution performance in any tested out-of-distribution conditions. This work highlights the need to consider layout distribution shifts during model evaluation, and presents a methodology for conducting such evaluations.

In Transformer-based language models (LMs) the attention mechanism converts token embeddings into contextual embeddings that incorporate information from neighboring words. The resulting contextual hidden state embeddings have enabled highly accurate models of brain responses, suggesting that the attention mechanism constructs contextual embeddings that carry information reflected in language-related brain representations. However, it is unclear whether the attention weights that are used to integrate information across words are themselves related to language representations in the brain. To address this question we analyzed functional magnetic resonance imaging (fMRI) recordings of participants reading English language narratives. We provided the narrative text as input to two LMs (BERT and GPT-2) and extracted their corresponding attention weights. We then used encoding models to determine how well attention weights can predict recorded brain responses. We find that attention weights accurately predict brain responses in much of the frontal and temporal cortices. Our results suggest that the attention mechanism itself carries information that is reflected in brain representations. Moreover, these results indicate cortical areas in which context integration may occur.

We present a method for constructing taxonomic trees (e.g., WordNet) using pretrained language models. Our approach is composed of two modules, one that predicts parenthood relations and another that reconciles those pairwise predictions into trees. The parenthood prediction module produces likelihood scores for each potential parent-child pair, creating a graph of parent-child relation scores. The tree reconciliation module treats the task as a graph optimization problem and outputs the maximum spanning tree of this graph. We train our model on subtrees sampled from WordNet, and test on nonoverlapping WordNet subtrees. We show that incorporating web-retrieved glosses can further improve performance. On the task of constructing subtrees of English WordNet, the model achieves 66.7 ancestor F1, a 20.0% relative increase over the previous best published result on this task. In addition, we convert the original English dataset into nine other languages using Open Multilingual WordNet and extend our results across these languages.