Probabilistic language models in cognitive neuroscience: Promises and pitfalls

Abstract

A R T I C L E I N F O A B S T R A C TCognitive neuroscientists of language comprehension study how neural computations relate to cognitive computations during comprehension. On the cognitive part of the equation, it is important that the computations and processing complexity are explicitly defined. Probabilistic language models can be used to give a computationally explicit account of language complexity during comprehension. Whereas such models have so far predominantly been evaluated against behavioral data, only recently have the models been used to explain neurobiological signals. Measures obtained from these models emphasize the probabilistic, information-processing view of language understanding and provide a set of tools that can be used for testing neural hypotheses about language comprehension. Here, we provide a cursory review of the theoretical foundations and example neuroimaging studies employing probabilistic language models. We highlight the advantages and potential pitfalls of this approach and indicate avenues for future research.

Hierarchical and sequential processing of language

“…However, effects of word-level statistics are (nearly) unavoidable: Word frequencies are for a large part responsible for word recognition times (Gardner, Rothkopf, Lapan, & Lafferty, 1987; among many others) and word reading times closely follow word probability conditional on the sentence context (Smith & Levy, 2013). Similar word-probability effects have been observed in neural activity during sentence or text comprehension (Brennan, Stabler, Van Wagenen, Luh, & Hale, 2016;Frank, Otten, Galli, & Vigliocco, 2015;Nelson et al, 2017;Willems, Frank, Nijhof, Hagoort, & Van den Bosch, 2016 for review, see Armeni, Willems, & Frank, 2017). Such probabilistic (i.e.…”

Lexical Predictability During Natural Reading: Effects of Surprisal and Entropy Reduction

“…Importantly, this study serves to link information complexity metrics to mechanistic accounts of cognition in a more naturalistic way. Metrics such as surprisal and entropy reduction are useful in that they can be generated word‐by‐word according to the specifications of a language model, which can then be related to behavioral or neural processing data to assess the viability of that model (Armeni, Willems, & Frank, ; Brennan, ; Hale, ). As noted above, though, these metrics tend to be estimated using computational parsers or other statistical language models.…”

M/EEG analysis of naturalistic stories: a review from speech to language processing

“…Building on the work of Hale (2001) and Levy (2008), who proposed mappings from natural language processing methods (parsers and Shannon entropy), the use of NLP models as regressors has expanded in fMRI research (cf. Brennan, Stabler, Van Wagenen, Luh, & Hale, 2016) and is slowly making the transition towards M/EEG (see Armeni, Willems, & Frank, 2017;Brennan, 2016 for review, as well as van Schijndel, Murphy, & Schuler, 2015 for an early attempt in the time-frequency domain with only three participants).…”