The Cost of the Epistemic Step: Investigating Scalar Implicatures in Full and Partial Information Contexts

Abstract

We present the first ERP experiments that test the online processing of the scalar implicature some ⇝ not all in contexts where the speaker competence assumption is violated. Participants observe game scenarios with four open cards on the table and two closed cards outside of the table, while listening to statements made by a virtual player. In the full access context, the player makes a fully informed statement by referring only to the open cards, as cards on the table; in the partial access context, she makes a partially informed statement by referring to the whole set of cards, as cards in the game. If all of the open cards contain a given object X (Fullset condition), then some cards on the table contain Xs is inconsistent with the not all reading, whereas it is unknown whether some cards in the game contain X is consistent with this reading. If only a subset of the open cards contains X (Subset condition), then both utterances are known to be consistent with the not all implicature. Differential effects are observed depending on the quantifier reading adopted by the participant: For those participants who adopt the not all reading in the full access context, but not in the partial access context (weak pragmatic reading), a late posterior negativity effect is observed in the partial access context for the Fullset relative to the Subset condition. This effect is argued to reflect inference-driven context retrieval and monitoring processes related to epistemic reasoning involved in evaluating the competence assumption. By contrast, for participants who adopt the logical interpretation of some (some and possibly all), an N400 effect is observed in the partial access context, when comparing the Subset against the Fullset condition, which is argued to result from the competition between the two quantifying expressions some cards on the table and some cards in the game functioning in the experiment as scalar alternatives.

Keywords: N400; alternatives; competence assumption; epistemic step; late posterior negativity; primary and secondary implicature; scalar implicature.

Figure 1

The structure of a trial: The picture was presented 1,800 ms before the onset of the audio and until 600 ms after the audio offset. The audio files varied in length (range: 2,262–3,211 ms, the average duration of Table-sentences was 2,800 ms and the average duration of Game-sentences was 2,603 ms) and contained approximately 200 ms of an initial silence for a more natural sound onset. The onset of the critical word was between 1,857 and 2,664 ms after the onset of the audio, the average onset was 2,139 ms. Subsequently, a screen with the assignment of the buttons was presented (left vs. right hand is counterbalanced). For a given scene, the condition was dependent on the specific sentence played, as explained in Table 1.

Figure 2

Six example filler trials of various sorts. For each quantifier all three card configurations (closed/open/no additional cards) were used. The number of visible object categories 1–3 was counterbalanced per quantifier/context/card configuration. The two contexts (Table/Game) were cross-balanced for each filler category. Sentences with bare numerals are subject to similar pragmatic ambiguity as those with some, i.e., based on the so-called exactly vs. at least readings. Since all target trials with some always showed two additional cards face-down, we introduced more fillers with two closed cards than those with other card configurations, so that the ratio of trials with closed/open/no additional cards would be relatively balanced across quantifiers.

Figure 3

The mean response times (in ms) with standard deviations (±1 SD), for each group and condition.

Figure 4

Grand averages for all conditions for logicians (N = 33), Experiment 1. Topographical maps of the differences between the compared Set levels separately for the Table and Game context in consecutive 100 ms time-windows, between 250 and 950 ms post-onset.

Figure 5

Grand averages for all conditions for weak pragmatists (N = 9), Experiment 1. Topographical maps of the differences between the compared Set levels separately for the Table and Game context in consecutive 100 ms time-windows, between 250 and 950 ms post-onset.

Figure 6

Grand averages for all conditions for weak pragmatists (N = 15), Experiment 2. Topographical maps of the differences between the compared Set levels separately for the Table and Game context in consecutive 100 ms time-windows, between 250 and 950 ms post-onset.

Figure 7

The comparison of subject averages in all conditions for the strong pragmatists, including the subject who displayed a mixed weak/strong response pragmatic response pattern (N = 4).