You are viewing the SNL 2017 Archive Website. For the latest information, see the Current Website.

Poster B13, Wednesday, November 8, 3:00 – 4:15 pm, Harborview and Loch Raven Ballrooms

Incremental working memory effects across consecutive nominal constituents: An ERP study

Alicia Parrish1, Kaylin Smith1, Alan Beretta1;1Michigan State University

Models of incremental sentence processing describe the integration of new words into a sentence, but many experimental tests of these models have needed to constrain themselves to testing differences in words between two different sentences. Exploiting the ability to control for phrase boundary and word category effects by using nominal compounds, this study uses Gibson’s (1998) well-established SPLT model to investigate the working memory effects as the parser moves from the first (N1) to the second (N2) to the third (N3) nominal constituent in triple-noun compounds. We created 35 sets of triple-noun compounds in Icelandic, a language with rich inflectional morphology and highly productive compounding, in a 2x2 design of number/case/gender agreement between the determiner and N1 and N2 such that there were the following four conditions: (i) Agree-Agree: determiner agrees with N1 and N2; (ii) Nonagree-Agree: determiner does not agree with N1 but agrees with N2; (iii) Agree-Nonagree: determiner agrees with N1 but not with N2; (iv) Nonagree-Nonagree: determiner does not agree with N1 or N2. In all conditions, the compound was placed in a sentence frame, and N3 agreed with the determiner to ensure that sentences were grammatical. There are three main components of the relevant predictions from Gibson’s model: (i) when there is a lack of agreement, it is not possible to integrate the noun/compound with the determiner; (ii) new discourse referents are only introduced through integrations; and (iii) there are multiple integrations on nominal constituents after N1 in order to create the [N1 N2] compound and then integrate with the determiner. Forty-seven native Icelandic speakers completed this ERP study. They read sentences presented via RSVP and we measured responses at the presentation of N1, N2, and N3. Higher working memory load is indexed by a sustained anterior negativity (Fiebach et al., 2005; Phillips et al., 2005). All results reported are for the anterior region across the 300-700ms window. We find that the predictions made by the SPLT model are borne out in the ERP responses. The Agree-Agree condition shows significant sustained increases in negativity for N2 compared to N1 and for N3 compared to N2. The Nonagree-Agree condition shows similar increases in negativity as later nominal constituents are encountered, but to a lesser extent than in the Agree-Agree condition because there is no integration at N1 and no new discourse referents are introduced between the determiner and N2. In the Agree-Nonagree condition, N1 and N2 show no difference because there is no new integration possible on N2. N3 shows an increase in negativity, reflecting two integrations and an intervening discourse referent. In the Nonagree-Nonagree condition, N2 and N3 show no difference because they both integrate constituents with no intervening discourse referents. Both N2 and N3 are more negative than N1, at which no integration could occur. This study provides strong support for a model of sentence processing that relies on working memory costs at integration points and, crucially, takes into account the cost of intervening discourse referents.

Topic Area: Grammar: Morphology

Back to Poster Schedule