Alba and Xiaoyu in Geneva for SNL

The neurolinguistics of cataphora and quantification.

September 29-October 3, the Society for the Neurobiology of Language holds its annual meeting in Geneva, Switzerland, with two posters from current Terrapins. Alba Jorquera presents "He Who Must Not Be Named (Yet): Sustained Anterior Negativity as a Potential Neural Index of Active Cataphor Processing", work co-authored by Victor, Tessa, Philip and Ellen, and Xiaoyu Yang presents "Investigating determiner meanings using time-resolved decoding in EEG."

• He Who Must Not Be Named (Yet): Sustained Anterior Negativity as a Potential Neural Index of Active Cataphor Processing

  This EEG study tests the hypothesis that sustained anterior negativity (SAN) indexes the working memory maintenance of predictively generated syntactic representations (Cruz Heredia et al., 2022). We use cataphors (pronouns that precede their antecedents) as our test case. Previous work has found that SAN is associated with predictions involving long-distance dependencies, but exactly what computations it indexes remains unknown. Prior behavioral work shows that in sentences with cataphors (e.g. After he came to the birthday party, John cheerfully gave Mary a hug), comprehenders predict a cataphor's antecedent in the closest structurally available position before it is actually encountered (Kazanina et al., 2007). Here we look for a neural index of this predictive process by contrasting sentences with cataphors to sentences with negative quantifiers (e.g., After nobody came to the birthday party, the child cried for hours). This design is motivated by a prior EEG study (Cruz Heredia et al., SNL 2020) that found no SAN for cataphora (After he…John) compared to forward anaphora (After John…he). We hypothesize that their null result may reflect a problem with the control condition: comprehenders could be predicting coreference in both the forward anaphora and cataphora conditions, in which case both might induce an equivalent sustained negativity. In the current study, using a negative quantifier ensures that no coreference relation can be predicted in the control condition because no such relation (nobody-the child) is possible. If SAN indexes the maintenance of syntactic predictions in general, we should observe a SAN across the first clause following the cataphor, relative to controls. Absence of a SAN would suggest sustained activity does not track syntactic prediction more broadly. We include a standard wh-contrast (wh-questions vs. yes/no questions) to replicate prior SAN findings for wh-dependencies. We also include an additional cataphor condition to explore what happens when the cataphor is not resolved at the expected main subject position but only later (e.g., After he came to the birthday party, Mary cheerfully gave John a hug). We expect a mismatch effect in the ERPs at the expected position and potentially a SAN that continues until the cataphor is resolved. The cataphor contrast includes three conditions: sentences where the first name in the second clause is a grammatical antecedent (After he…John), sentences where it is not but a later antecedent is available (After he…Mary…John), and negative quantifier controls (After nobody…the child). Regions of interest (e.g., came to the birthday party) are identical across all three conditions. We average ERPs for the cataphor conditions and compare with the control. A SAN for cataphors relative to control would support our syntactic prediction hypothesis. We designed 96 items for the cataphor contrast (32 per condition) and 64 for the wh-contrast (32 per condition). Stimuli will be presented RSVP-style, word-by-word, followed by comprehension probes. Epochs for the cataphor contrast will be time-locked to the word following the cataphor/quantifier and span until the end of the first clause; epochs for the wh-contrast will be handled similarly. Data collection is underway.

• Investigating determiner meanings using time-resolved decoding in EEG

  In this study, we collect EEG data during comprehension of sentences like 'Each/Every circle is red' and 'All/Some of the circles are red' and use time-resolved decoding to test hypotheses about the roles of the determiner's first vs. second arguments (‘circle’ vs. ‘red’) in meaning construction. Knowlton et al. (2020, 2023a) found that when people evaluate sentences containing 'every' against visual stimuli, only objects specified by its first argument are represented as a group (e.g. the circles), but not those specified by its second argument (e.g. the red objects). Based on this evidence, they proposed that determiner meanings take the first argument to define a restrictive domain, and the second argument to provide a predicate to be applied to domain member(s). We hypothesize that comprehension of such sentences thus consists of two operations, (1) the formation of the restrictive domain and (2) the predication of domain members, which are carried out serially during the incremental processing of the noun phrase and the predicate respectively. Here we test this hypothesis by using neural responses as an index of the timing of these operations. We predict that differences in determiner meaning that affect the first operation should result in decodability of the determiner from neural responses to the first argument (‘circle’). For this we make use of the contrast between 'each' vs. 'every', which has been found to motivate individual vs. group representations of the specified objects (Knowlton et al., 2022a, 2022b, 2023b). In contrast, since 'all' and 'some' differ in quantificational force, i.e. quantities to be predicated of, our hypothesis predicts that this difference affects the second operation and leads to different neural responses and thus determiner decodability only during the predicate (‘red’). The experiment comprises four determiner blocks - each, every, all, some - each containing 80 sentences presented with RSVP using a 600 ms SOA. Sentences in the 'each' and 'every' blocks have the structure 'DET SHAPE is COLOR' (e.g. 'Each circle is red'), and those in the 'all' and 'some' blocks have the structure 'DET of the SHAPEs are COLOR' (e.g. 'All of the circles are red'). 20% of the trials are followed by a memory probe (sentence recognition), and another 20% by a visual quantification probe screen containing various shapes of different colors, where participants are asked to estimate the number of objects specified by either the noun phrase or the predicate. We conduct two pairwise multivariate decoding analyses, one for the each/every pair and one for the all/some pair. Each uses a binary logistic regression classifier trained at each time point throughout the noun phrase and the predicate to separate the two relevant determiner conditions. Data collection and analysis are underway.