Do large multilingual language models learn language universals? We consider a candidate universal much-discussed in the linguistics literature, the Final-over-Final Condition (Sheehan et al., 2017b). This Condition is syntactic in the sense that it can only be stated by reference to abstract sentence properties such as nested phrases and head direction. A study of typologically diverse “mixed head direction” languages confirms that the Condition holds in corpora. But in a targeted syntactic evaluation, Gemini Pro only seems to respect the Condition in German, Russian, Hungarian and Serbian. These relatively high-resource languages contrast with Basque, where Gemini Pro does not seem to have learned the Condition at all. This result suggests that modern language models may need additional sources of bias in order to become truly human-like, within a developmentally-realistic budget of training data.

Automatic speech recognition (ASR) has the potential to accelerate the documentation of endangered languages, but the dearth of resources poses a major obstacle. Čakavian, an endangered variety spoken primarily in Croatia, is a case in point, lacking transcription tools that could aid documentation efforts. We compare training a new ASR model on a limited dataset using the Kaldi-based ASR pipeline Elpis to using the same dataset to adapt the transformer-based pretrained multilingual model Whisper, to determine which is more practical in the documentation context. Results show that Whisper outperformed Elpis, achieving the lowest average Word Error Rate (WER) of 57.3% and median WER of 35.48%. While Elpis offers a less computationally expensive model and friendlier user experience, Whisper appears better at adapting to our collected Čakavian data.

Language models (LMs) are a meeting point for cognitive modeling and computational linguistics. How should they be designed to serve as adequate cognitive models? To address this question, this study contrasts two Transformer-based LMs that share the same architecture. Only one of them analyzes sentences in terms of explicit hierarchical structure. Evaluating the two LMs against fMRI time series via the surprisal complexity metric, the results implicate the superior temporal gyrus. These findings underline the need for hierarchical sentence structures in word-by-word models of human language comprehension.

Humans understand sentences word-by-word, in the order that they hear them. This incrementality entails resolving temporary ambiguities about syntactic relationships. We investigate how humans process these syntactic ambiguities by correlating predictions from incremental generative dependency parsers with timecourse data from people undergoing functional neuroimaging while listening to an audiobook. In particular, we compare competing hypotheses regarding the number of developing syntactic analyses in play during word-by-word comprehension: one vs more than one. This comparison involves evaluating syntactic surprisal from a state-of-the-art dependency parser with LLM-adapted encodings against an existing fMRI dataset. In both English and Chinese data, we find evidence for multipath parsing. Brain regions associated with this multipath effect include bilateral superior temporal gyrus.

To assist in the documentation of Čakavian, an endangered language variety closely related to Croatian, we test four currently available ASR models that are trained with Croatian data and assess their performance in the transcription of Čakavian audio data. We compare the models’ word error rates, analyze the word-level error types, and showcase the most frequent Deletion and Substitution errors. The evaluation results indicate that the best-performing system for transcribing Čakavian was a CTC-based variant of the Conformer model.

Pro-drop is commonly seen in many languages, but its discourse motivations have not been well characterized. Inspired by the topic chain theory in Chinese, this study shows how character-verb usage continuity distinguishes dropped pronouns from overt references to story characters. We model the choice to drop vs. not drop as a function of character-verb continuity. The results show that omitted subjects have higher character history-current verb continuity salience than non-omitted subjects. This is consistent with the idea that discourse coherence with a particular topic, such as a story character, indeed facilitates the omission of pronouns in languages and contexts where they are optional.

Hierarchical sentence structure plays a role in word-by-word human sentence comprehension, but it remains unclear how best to characterize this structure and unknown how exactly it would be recognized in a step-by-step process model. With a view towards sharpening this picture, we model the time course of hemodynamic activity within the brain during an extended episode of naturalistic language comprehension using Combinatory Categorial Grammar (CCG). CCG has well-defined incremental parsing algorithms, surface compositional semantics, and can explain long-range dependencies as well as complicated cases of coordination. We find that CCG-derived predictors improve a regression model of fMRI time course in six language-relevant brain regions, over and above predictors derived from context-free phrase structure. Adding a special Revealing operator to CCG parsing, one designed to handle right-adjunction, improves the fit in three of these regions. This evidence for CCG from neuroimaging bolsters the more general case for mildly context-sensitive grammars in the cognitive science of language.

The Alice Datasets are a set of datasets based on magnetic resonance data and electrophysiological data, collected while participants heard a story in English. Along with the datasets and the text of the story, we provide a variety of different linguistic and computational measures ranging from prosodic predictors to predictors capturing hierarchical syntactic information. These ecologically valid datasets can be easily reused to replicate prior work and to test new hypotheses about natural language comprehension in the brain.

We present the Le Petit Prince Corpus (LPPC), a multi-lingual resource for research in (computational) psycho- and neurolinguistics. The corpus consists of the children’s story The Little Prince in 26 languages. The dataset is in the process of being built using state-of-the-art methods for speech and language processing and electroencephalography (EEG). The planned release of LPPC dataset will include raw text annotated with dependency graphs in the Universal Dependencies standard, a near-natural-sounding synthetic spoken subset as well as EEG recordings. We will use this corpus for conducting neurolinguistic studies that generalize across a wide range of languages, overcoming typological constraints to traditional approaches. The planned release of the LPPC combines linguistic and EEG data for many languages using fully automatic methods, and thus constitutes a readily extendable resource that supports cross-linguistic and cross-disciplinary research.

Domain-specific training typically makes NLP systems work better. We show that this extends to cognitive modeling as well by relating the states of a neural phrase-structure parser to electrophysiological measures from human participants. These measures were recorded as participants listened to a spoken recitation of the same literary text that was supplied as input to the neural parser. Given more training data, the system derives a better cognitive model — but only when the training examples come from the same textual genre. This finding is consistent with the idea that humans adapt syntactic expectations to particular genres during language comprehension (Kaan and Chun, 2018; Branigan and Pickering, 2017).

Language exhibits hierarchical structure, but recent work using a subject-verb agreement diagnostic argued that state-of-the-art language models, LSTMs, fail to learn long-range syntax sensitive dependencies. Using the same diagnostic, we show that, in fact, LSTMs do succeed in learning such dependencies—provided they have enough capacity. We then explore whether models that have access to explicit syntactic information learn agreement more effectively, and how the way in which this structural information is incorporated into the model impacts performance. We find that the mere presence of syntactic information does not improve accuracy, but when model architecture is determined by syntax, number agreement is improved. Further, we find that the choice of how syntactic structure is built affects how well number agreement is learned: top-down construction outperforms left-corner and bottom-up variants in capturing non-local structural dependencies.

Recurrent neural network grammars (RNNGs) are generative models of (tree , string ) pairs that rely on neural networks to evaluate derivational choices. Parsing with them using beam search yields a variety of incremental complexity metrics such as word surprisal and parser action count. When used as regressors against human electrophysiological responses to naturalistic text, they derive two amplitude effects: an early peak and a P600-like later peak. By contrast, a non-syntactic neural language model yields no reliable effects. Model comparisons attribute the early peak to syntactic composition within the RNNG. This pattern of results recommends the RNNG+beam search combination as a mechanistic model of the syntactic processing that occurs during normal human language comprehension.

Typological differences between English and Chinese suggest stronger reliance on salience of the antecedent during pronoun resolution in Chinese. We examined this hypothesis by correlating a difficulty measure of pronoun resolution derived by the activation-based ACT-R model with the brain activity of English and Chinese participants listening to a same audiobook during fMRI recording. The ACT-R model predicts higher overall difficulty for English speakers, which is supported at the brain level in left Broca’s area. More generally, it confirms that computational modeling approach is able to dissociate different dimensions that are involved in the complex process of pronoun resolution in the brain.

The current study examined the role of syntactic structure during pronoun resolution. We correlated complexity measures derived by the syntax-sensitive Hobbs algorithm and a neural network model for pronoun resolution with brain activity of participants listening to an audiobook during fMRI recording. Compared to the neural network model, the Hobbs algorithm is associated with larger clusters of brain activation in a network including the left Broca’s area.

Multiword expressions have posed a challenge in the past for computational linguistics since they comprise a heterogeneous family of word clusters and are difficult to detect in natural language data. In this paper, we present a fMRI study based on language comprehension to provide neuroimaging evidence for processing MWEs. We investigate whether different MWEs have distinct neural bases, e.g. if verbal MWEs involve separate brain areas from non-verbal MWEs and if MWEs with varying levels of cohesiveness activate dissociable brain regions. Our study contributes neuroimaging evidence illustrating that different MWEs elicit spatially distinct patterns of activation. We also adapt an association measure, usually used to detect MWEs, as a cognitively plausible metric for language processing.

Using the Entropy Reduction incremental complexity metric, we relate high gamma power signals from the brains of epileptic patients to incremental stages of syntactic analysis in English and French. We find that signals recorded intracranially from the anterior Inferior Temporal Sulcus (aITS) and the posterior Inferior Temporal Gyrus (pITG) correlate with word-by-word Entropy Reduction values derived from phrase structure grammars for those languages. In the anterior region, this correlation persists even in combination with surprisal co-predictors from PCFG and ngram models. The result confirms the idea that the brain’s temporal lobe houses a parsing function, one whose incremental processing difficulty profile reflects changes in grammatical uncertainty.

The relative contributions of meaning and form to sentence processing remains an outstanding issue across the language sciences. We examine this issue by formalizing four incremental complexity metrics and comparing them against freely-available ROI timecourses. Syntax-related metrics based on top-down parsing and structural dependency-distance turn out to significantly improve a regression model, compared to a simpler model that formalizes only conceptual combination using a distributional vector-space model. This confirms the view of the anterior temporal lobes as combinatory engines that deal in both form (see e.g. Brennan et al., 2012; Mazoyer, 1993) and meaning (see e.g., Patterson et al., 2007). This same characterization applies to a posterior temporal region in roughly “Wernicke’s Area.”

While both spoken and written language processing stand to benefit from parsing, the standard Parseval metrics (Black et al., 1991) and their canonical implementation (Sekine and Collins, 1997) are only useful for text. The Parseval metrics are undefined when the words input to the parser do not match the words in the gold standard parse tree exactly, and word errors are unavoidable with automatic speech recognition (ASR) systems. To fill this gap, we have developed a publicly available tool for scoring parses that implements a variety of metrics which can handle mismatches in words and segmentations, including: alignment-based bracket evaluation, alignment-based dependency evaluation, and a dependency evaluation that does not require alignment. We describe the different metrics, how to use the tool, and the outcome of an extensive set of experiments on the sensitivity.