Recent studies have shown that as Transformer-based language models become larger and are trained on very large amounts of data, the fit of their surprisal estimates to naturalistic human reading times degrades. The current work presents a series of analyses showing that word frequency is a key explanatory factor underlying these two trends. First, residual errors from four language model families on four corpora show that the inverse correlation between model size and fit to reading times is the strongest on the subset of least frequent words, which is driven by excessively accurate predictions of larger model variants. Additionally, training dynamics reveal that during later training steps, all model variants learn to predict rare words and that larger model variants do so more accurately, which explains the detrimental effect of both training data amount and model size on fit to reading times. Finally, a feature attribution analysis demonstrates that larger model variants are able to accurately predict rare words based on both an effectively longer context window size as well as stronger local associations compared to smaller model variants. Taken together, these results indicate that Transformer-based language models’ surprisal estimates diverge from human-like expectations due to the superhumanly complex associations they learn for predicting rare words.

Recent psycholinguistic studies have drawn conflicting conclusions about the relationship between the quality of a language model and the ability of its surprisal estimates to predict human reading times, which has been speculated to be due to the large gap in both the amount of training data and model capacity across studies. The current work aims to consolidate these findings by evaluating surprisal estimates from Transformer-based language model variants that vary systematically in the amount of training data and model capacity on their ability to predict human reading times. The results show that surprisal estimates from most variants with contemporary model capacities provide the best fit after seeing about two billion training tokens, after which they begin to diverge from humanlike expectations. Additionally, newly-trained smaller model variants reveal a ‘tipping point’ at convergence, after which the decrease in language model perplexity begins to result in poorer fits to human reading times. These results suggest that the massive amount of training data is mainly responsible for the poorer fit achieved by surprisal from larger pre-trained language models, and that a certain degree of model capacity is necessary for Transformer-based language models to capture humanlike expectations.

This work presents a linguistic analysis into why larger Transformer-based pre-trained language models with more parameters and lower perplexity nonetheless yield surprisal estimates that are less predictive of human reading times. First, regression analyses show a strictly monotonic, positive log-linear relationship between perplexity and fit to reading times for the more recently released five GPT-Neo variants and eight OPT variants on two separate datasets, replicating earlier results limited to just GPT-2 (Oh et al., 2022). Subsequently, analysis of residual errors reveals a systematic deviation of the larger variants, such as underpredicting reading times of named entities and making compensatory overpredictions for reading times of function words such as modals and conjunctions. These results suggest that the propensity of larger Transformer-based models to ‘memorize’ sequences during training makes their surprisal estimates diverge from humanlike expectations, which warrants caution in using pre-trained language models to study human language processing.

While there is much recent interest in studying why Transformer-based large language models make predictions the way they do, the complex computations performed within each layer have made their behavior somewhat opaque. To mitigate this opacity, this work presents a linear decomposition of final hidden states from autoregressive language models based on each initial input token, which is exact for virtually all contemporary Transformer architectures. This decomposition allows the definition of probability distributions that ablate the contribution of specific input tokens, which can be used to analyze their influence on model probabilities over a sequence of upcoming words with only one forward pass from the model. Using the change in next-word probability as a measure of importance, this work first examines which context words make the biggest contribution to language model predictions. Regression experiments suggest that Transformer-based language models rely primarily on collocational associations, followed by linguistic factors such as syntactic dependencies and coreference relationships in making next-word predictions. Additionally, analyses using these measures to predict syntactic dependencies and coreferent mention spans show that collocational association and repetitions of the same token largely explain the language models’ predictions on these tasks.

Transformer-based large language models are trained to make predictions about the next word by aggregating representations of previous tokens through their self-attention mechanism. In the field of cognitive modeling, such attention patterns have recently been interpreted as embodying the process of cue-based retrieval, in which attention over multiple targets is taken to generate interference and latency during retrieval. Under this framework, this work first defines an entropy-based predictor that quantifies the diffuseness of self-attention, as well as distance-based predictors that capture the incremental change in attention patterns across timesteps. Moreover, following recent studies that question the informativeness of attention weights, we also experiment with alternative methods for incorporating vector norms into attention weights. Regression experiments using predictors calculated from the GPT-2 language model show that these predictors deliver a substantially better fit to held-out self-paced reading and eye-tracking data over a rigorous baseline including GPT-2 surprisal.

This paper describes Team Ohio State’s approach to the CMCL 2021 Shared Task, the goal of which is to predict five eye-tracking features from naturalistic self-paced reading corpora. For this task, we fine-tune a pre-trained neural language model (RoBERTa; Liu et al., 2019) to predict each feature based on the contextualized representations. Moreover, motivated by previous eye-tracking studies, we include word length in characters and proportion of sentence processed as two additional input features. Our best model strongly outperforms the baseline and is also competitive with other systems submitted to the shared task. An ablation study shows that the word length feature contributes to making more accurate predictions, indicating the usefulness of features that are specific to the eye-tracking paradigm.

Expectation-based theories of sentence processing posit that processing difficulty is determined by predictability in context. While predictability quantified via surprisal has gained empirical support, this representation-agnostic measure leaves open the question of how to best approximate the human comprehender’s latent probability model. This work presents an incremental left-corner parser that incorporates information about both propositional content and syntactic categories into a single probability model. This parser can be trained to make parsing decisions conditioning on only one source of information, thus allowing a clean ablation of the relative contribution of propositional content and syntactic category information. Regression analyses show that surprisal estimates calculated from the full parser make a significant contribution to predicting self-paced reading times over those from the parser without syntactic category information, as well as a significant contribution to predicting eye-gaze durations over those from the parser without propositional content information. Taken together, these results suggest a role for propositional content and syntactic category information in incremental sentence processing.

Recent evidence supports a role for coreference processing in guiding human expectations about upcoming words during reading, based on covariation between reading times and word surprisal estimated by a coreference-aware semantic processing model (Jaffe et al. 2020).The present study reproduces and elaborates on this finding by (1) enabling the parser to process subword information that might better approximate human morphological knowledge, and (2) extending evaluation of coreference effects from self-paced reading to human brain imaging data. Results show that an expectation-based processing effect of coreference is still evident even in the presence of the stronger psycholinguistic baseline provided by the subword model, and that the coreference effect is observed in both self-paced reading and fMRI data, providing evidence of the effect’s robustness.

Unsupervised PCFG induction models, which build syntactic structures from raw text, can be used to evaluate the extent to which syntactic knowledge can be acquired from distributional information alone. However, many state-of-the-art PCFG induction models are word-based, meaning that they cannot directly inspect functional affixes, which may provide crucial information for syntactic acquisition in child learners. This work first introduces a neural PCFG induction model that allows a clean ablation of the influence of subword information in grammar induction. Experiments on child-directed speech demonstrate first that the incorporation of subword information results in more accurate grammars with categories that word-based induction models have difficulty finding, and second that this effect is amplified in morphologically richer languages that rely on functional affixes to express grammatical relations. A subsequent evaluation on multilingual treebanks shows that the model with subword information achieves state-of-the-art results on many languages, further supporting a distributional model of syntactic acquisition.

While the use of character models has been popular in NLP applications, it has not been explored much in the context of psycholinguistic modeling. This paper presents a character model that can be applied to a structural parser-based processing model to calculate word generation probabilities. Experimental results show that surprisal estimates from a structural processing model using this character model deliver substantially better fits to self-paced reading, eye-tracking, and fMRI data than those from large-scale language models trained on much more data. This may suggest that the proposed processing model provides a more humanlike account of sentence processing, which assumes a larger role of morphology, phonotactics, and orthographic complexity than was previously thought.

This paper describes the OSU submission to the SIGMORPHON 2019 shared task, Crosslinguality and Context in Morphology. Our system addresses the contextual morphological analysis subtask of Task 2, which is to produce the morphosyntactic description (MSD) of each fully inflected word within a given sentence. We frame this as a sequence generation task and employ a neural encoder-decoder (seq2seq) architecture to generate the sequence of MSD tags given the encoded representation of each token. Follow-up analyses reveal that our system most significantly improves performance on morphologically complex languages whose inflected word forms typically have longer MSD tag sequences. In addition, our system seems to capture the structured correlation between MSD tags, such as that between the “verb” tag and TAM-related tags.