Presentation

Alicia Parrish¹, Amilleah Rodriguez¹, Liina Pylkkänen^1,2; ¹New York University, ²NYUAD Institute, New York University Abu Dhabi

Language enables us to compose meaning from words regardless of whether the words are directly adjacent. In both the phrase ‘blue hat’ and ‘the blue color of that hat’, the words ‘blue’ and ‘hat’ compose to form a complex conceptual representation. We know that the left anterior temporal lobe (LATL) is sensitive to conceptual combination. But we do not know whether this sensitivity extends to long-distance instances of conceptual combination, that is, cases in which the two composing elements are not adjacent. Here, we addressed this by varying the locality of conceptual combination during an MEG measurement. Parrish & Pylkkänen (2022) showed that LATL composition does not require an accompanying syntactic merge, as long as the elements conceptually combine. But in that study, the combining elements were still adjacent. Here, we break that adjacency to further examine the independence of the LATL from syntax. Further, we use a decoding approach to measure the activation of lexical representations during composition, in particular, the activation of the first element (e.g., blue) during the presentation of the second element (e.g., hat) in both long-distance and local contexts. Our aim is to understand the format of lexical representations that serves as input to composition. [METHODS] So far, data from 12 participants have been collected. They read English sentences via RSVP in a picture-verification task during a magnetoencephalography recording. This study uses a 2x2x2 design of locality (local, non-local) by conceptual combination (combinatory, non-combinatory) by word order (noun-adjective, adjective-noun). The combinatory comparison varies whether the target word conceptually composes with an earlier adjective/noun: COMBINATORY: ‘the hat is a really pretty *blue* color’ / ‘the blue color of this *hat* is pretty’; NON-COMBINATORY: ‘the hat is near a pretty *blue* lamp’ / ‘the blue lamp near this *hat* is pretty’. [RESULTS] ~ROI analyses~ Using a cluster permutation test within pre-determined ROIs, we observe a marginally significant increase in activation for combinatory stimuli relative to non-combinatory stimuli in the adjective-noun order condition (but not noun-adjective order) in the LATL, measured on the target noun. This effect was present in both the long-distance stimuli and local controls, with no interaction of these factors, consistent with an account of a conceptual combination mechanism that is sensitive to conceptual compatibility but not to linear adjacency. ~Decoding analyses~ Our decoding analysis reveals greater decoding accuracy of the preceding adjective on the noun that it composes with in local combinatory contexts compared to non-combinatory contexts. However, we observe no difference between conditions in non-local contexts or noun-adjective word order. [CONCLUSION] By separating two conceptually composing words with intervening material, we show that the mechanism behind LATL conceptual combination tracks conceptual compatibility without regard to linear adjacency. Via our decoding analysis, we are also able to show activation of the context word while it combines with the currently presented word, but this was only observed for local composition, suggesting a different format of lexical representations for local and non-local conceptual combination.

Topic Areas: Meaning: Combinatorial Semantics, Syntax