How Brain-Like is an LSTM’s Representation of Nonsensical Language Stimuli?

Abstract

The representations generated by many models of language (word embeddings, recurrent neural networks and transformers) correlate to brain activity recorded while people listen. However, these decoding results are usually based on the brain’s reaction to syntactically and semantically sound language stimuli. In this study, we asked: how does an LSTM (long short term memory) language model, trained (by and large) on semantically and syntactically intact language, represent a language sample with degraded semantic or syntactic information? Does the LSTM representation still resemble the brain’s reaction? We found that, even for some kinds of nonsensical language, there is a statistically significant relationship between the brain’s activity and the representations of an LSTM. More exceptional, a character-based LSTM’s representation of pseudoword sentences is significantly correlated to EEG collected while people listened to those sentences - even though the pseudowords were not in the LSTM training data. This indicates that, at least in some instances, LSTMs and the human brain handle nonsensical data similarly.