|
Building temporal relations on-line Silvia Gennari How are temporal interpretations built during sentence processing? Previous research has shown that interpreters quickly establish temporal relations with discourse events when encountering tensed-verbs [1]. In this paper, I investigate how properties of the event structure denoted by verbs determine the temporal interpretation being built. Two verb types are often distinguished [2, 3, 4]: Stative verbs denote
facts that tend to persist in time and lack causal structure (e.g.,
being-German). Eventive verbs denote complex changes that are temporally
bounded by their cause-effect relations (e.g., building a house). These
properties determine the temporal relations that events and states establish
in discourse [5, 6, 7]: While eventive sentences are sequentially interpreted
relative previous events (e.g., the building (1) Bill said yesterday that [Hillary was sick] Experiment 1 tested whether stative sentences are preferably interpreted as overlapping the matrix event in (1). In an off-line study (25 subjects, 16 items, 30 fillers), speakers judged whether a target sentence (either consistent or inconsistent with temporal overlap) was true relative to the information previously provided, as in (3). (3) Information: Bill said yesterday that Hillary was sick. The mean proportion of responses consistent with overlapping interpretations was .80 for stative sentences versus .20 for eventive ones (p=.0001). The overlapping interpretation is available and preferred for states, but not events. Experiment 2 investigated whether the overlapping interpretation is built on-line, rather than inferred afterwards. If an overlapping temporal relation with previous events is established when "was-sick" is integrated (as in [1]), then later temporal information inconsistent with the already established overlapping relation should take longer to process than consistent information. A revision of the initially established relation would occur. (4) Bill said yesterday that Hillary was sick this week/last week.... To test this hypothesis, a self-paced reading experiment was conducted with 20 items like (4), 30 subjects, 100 fillers. Items were counterbalanced by temporal distance from the main event and frequency-matched. Mean reading time for overlapping phrases (this week) was shorter than that for non-overlapping ones (last week) (p= .01). Thus, an overlapping relation with the main event is already established at the verb-phrase. These findings elucidate contextual and semantic constraints acting during processing. Lexical event structure is fully integrated with previous temporal information in real time. References [1] Trueswell & Tanenhaus, 1991, "Tense, Temporal Context,
and Syntactic Ambiguity Resolution", Language and Cognitive Processes,
6(4), 303-338. |