Agenda
Nino Grillo – Retrieval Interference or ease of thematic integration?
Joint work with Andrea Santi (UCL), Fani Karageorgou and Shayne Sloggett (University of York)
Many current models of sentence comprehension employ a content addressable memory architecture which does not privilege structural information [1–5]. A seminal finding comes from [2], who suggest that even nouns outside the current sentential context interfere with the formation of intra-sentential dependencies. In this study, we explore the possibility that these results do not reflect similarity-based interference, but instead show facilitated thematic integration for semantically associated dependents.
In a self-paced reading experiment, [2] tested sentences like (1), manipulating verb type (plausible, implausible) and an extraneous Memory load (load, no-load). In the load condition, participants memorized a list of three nouns (e.g. table–sink–truck) to be recalled after the sentence. The verb-type manipulation determined whether the memory-load words were plausible arguments of the main verb.
(1) a. Implausible: It was the boat/ that the guy/ who lived/ by the sea/ sailed/ on a sunny day.
b. Plausible: It was the boat/ that the guy/ who lived/ by the sea/ fixed/ on a sunny day.
At the clefted verb, [2] found longer reading times (RTs) for plausible verbs in the load condition, but no RT difference across verb-type in the no-load condition. In a subsequent study, [6] found the same effect in comprehension question accuracy for low-working-memory participants, but no effect on RTs. These interference effects have been taken to suggest that dependency formation employs a cue-based retrieval mechanism which does not exclusively probe the current sentential context.
In contrast, we propose that these results do not reflect similarity-based interference, but instead demonstrate facilitated thematic integration when the verb and the filler are semantically related (boat-sail) compared to when they are not (boat-fix ). While [2] argue against this alternative, noting that a verb-type effect was not observed in their no load condition, we posit that the increased complexity of the load condition may have accentuated the contrast in (1). If true, this would mean that the memory list isn’t causing similarity-based interference, but is instead inducing a task-based effect.
We tested this alternative hypothesis in two experiments using self-paced reading (N=60 English native speakers) and the G(rammatical)-maze (N=48). In SPR we held the presence of a memory load constant and manipulating the contents of that memory load (Load Type) so that they were either possible (nouns), or impossible (verbs) arguments of the main verb in an otherwise direct replication of [2]. If the results of [2, 6] were due to similarity based interference, we expect an interaction between the original verb manipulation and Load Type, such that longer RTs and decreased accuracy are associated with noun-loads (ie, plausible arguments for the main verb), but no difference with a memory load of verbs (ie, implausible arguments). In contrast, if previous findings were due to facilitated integration for semantically related filler-verb pairs compared to pairs that are unrelated, we expect an effect of Relatedness across the Load Type manipulation. That is, a stronger semantic relation between the filler and the main verb should result in shorter RTs and/or fewer errors regardless of the words in the memory-load list. The G-maze study used the same sentences but with no memory load. If our hypothesis is correct, we reasoned that since the maze is more demanding than SPR, we should be able to observe the relatedness effect also in the absence of memory load, contrary to the predictions of the retrieval interference account.
Results from both experiments align with the predictions of the thematic integration account. In SPR, we found evidence that increased memory load leads to facilitated thematic integration for semantically related fillers/verb dependencies, independently on whether the load was a potential argument at retrieval site (noun-load) or not (verb-load). In the G-maze task we found shorter reading times at the verb and shorter response time to comprehension questions when the filler and the verb were semantically related (boat-sail) than when they were not (boat-fix ).
References
[1] Lewis, R. L., & Vasishth, S. (2005). Cognitive science, 29 (3).
[2] Van Dyke, J. A., & McElree, B. (2006). Journal of Memory and Language, 55 (2).
[3] Vasishth, S., Brüssow, S., Lewis, R. L., & Drenhaus, H. (2008). Cognitive Science, 32 (4).
[4] Bartek, B., Lewis, R. L., Vasishth, S., & Smith, M. R. (2011). Journal of Experimental Psychology: Learning, Memory, and Cognition, 37 (5).
[5] Parker, D., & Phillips, C. (2017). Journal of Memory and Language, 94.
[6] Van Dyke, J. A., Johns, C. L., & Kukona, A. (2014). Cognition, 131 (3).
Nino Grillo is affiliated with the University of York.
Teams
Meeting ID: 339 126 611 319 8
Passcode: Z8s6ku9M