Structural Persistence in Language Models: Priming as a Window into Abstract Language Representations

Structural Persistence in Language Models: Priming as a Window into Abstract Language Representations Arabella Sinclair1,2∗ Jaap Jumelet2∗ Willem Zuidema2 Raquel Fern´andez2 1School of Natural and Computing Sciences 2Institute for Logic, Language and Computation University of Aberdeen, United Kingdom University of Amsterdam, The Netherlands arabella.sinclair@abdn.ac.uk {j.w.d.jumelet|zuidema|raquel.fernandez}@uva.nl Abstract We investigate the extent to which modern neural language models are susceptible to structural priming, the phenomenon whereby the structure of a sentence makes the same structure more probable in a follow-up sen- tence. We explore how priming can be used to study the potential of these models to learn abstract structural information, which is a prerequisite for good performance on tasks that require natural language understanding skills. We introduce a novel metric and re- lease PRIME-LM, a large corpus where we control for various linguistic factors that in- teract with priming strength. We find that Transformer models indeed show evidence of structural priming, but also that the gen- eralizations they learned are to some extent modulated by semantic information. Our ex- periments also show that the representations acquired by the models may not only en- code abstract sequential structure but involve certain level of hierarchical syntactic infor- mation. More generally, our study shows that the priming paradigm is a useful, additional tool for gaining insights into the capacities of language models and opens the door to fu- ture priming-based investigations that probe the model’s internal states.1 1 Introduction It has become increasingly clear that modern, neural language models (LMs) are capable of rep- resenting and learning a broad range of linguistic phenomena (Gulordava et al., 2018; Hewitt and Manning, 2019; Tenney et al., 2019a; Rogers et al., 2020; Warstadt et al., 2020). However, many open questions remain about the extent to which specific LMs have indeed acquired specific ∗Equal contribution. 1Our code and data can be found at https://github .com/dmg-illc/prime-lm. linguistic constructions, about whether these mod- els encode an abstract notion of structure in their representations, and about the best ways to even assess the syntactic abilities of these models. A rich literature has emerged in the last few years ad- dressing these questions, often taking inspiration from methodologies developed in theoretical lin- guistics, psycholinguistics, neurolinguistics, and language acquisition research (Futrell et al., 2019; Ettinger, 2020; Boleda, 2020; Gauthier et al., 2020; Baroni, 2022), where the same questions have been asked about the human mind/brain for centuries. Building on this tradition, this paper turns to structural priming to investigate the degree to which LMs encode abstract structural information independent from the concrete words that make up sentences. This phenomenon refers to the fact that humans are more likely to produce—or to more easily comprehend—a sentence of a certain structure X (the target) when they have been exposed before to a sentence of a similar structure X (the prime), than if they had been prompted with a sentence of a different structure Y . For example, a native speaker of English will be more inclined to produce the target sentence with a prepositional object …

Content truncated to 500 words