The "resolution" of verb meaning in context
MetadataShow full item record
It is well-known that the meaning of a word often changes depending on the context in which the word is used. Determining the appropriate interpretation for a word occurrence requires a knowledge of the range of possible meanings for that word, and consideration of those possibilities given available contextual evidence. However, there is still much to be learned about the nature of our lexical knowledge, as well as how we make use of that knowledge in the course of language comprehension. I report on a series of three experiments that explore these issues. I begin with the question of how precise our perceptions of word meaning in context really are. In Experiment 1, I present a Magnitude Estimation study in which I obtain judgments of meaning-in-context similarity over pairs of intransitive verb occur- rences, such as The kid runs / The cat runs, or The cat runs / The lane runs. I find that participants supply a large range of very specific similarity judgments, that judgments are quite consistent across participants, and that these judgments can be at least partially predicted even by simple measures of contextual properties, such as subject noun animacy and human similarity ratings over pairs of subject nouns. However, I also find that while some participants supply a great variety of ratings, many participants supply only a few unique values during the task. This suggests that some individuals are making more fine-grained judgments than others. These differences in response granularity could stem from a variety of sources. However, the offline nature of Experiment 1 does not enable direct examination of the comprehension process, but rather focuses on its end result. In Experiment 2, I present a Speed-Accuracy Tradeoff study that explores the earliest stages of meaning-in-context resolution to better understand the dynamics of the comprehension process itself. In particular, I focus on the timecourse of meaning resolution and the question of whether verbs carry context-independent default interpretations that are activated prior to semantic integration. I find, consistent with what has previously been shown for nouns, that verbs do in fact carry such a default meaning, as can be seen in early false alarms to stimuli such as The dawn broke -- Something shattered. These default meanings appear to reflect the most frequent interpretation of the verb. While these default meanings are likely an emergent effect of repeated exposure to frequent interpretations of a verb, I hypothesize that they additionally support a shallow semantic processing strategy. Recently, a growing body of work has begun to demonstrate that our language comprehension is often less than exhaustive and less than maximally accurate -- people often vary the depth of their processing. In Experiment 3, I explore changes in depth of semantic processing by making an explicit connection to research on human decision making, particularly as regards questions of strategy selection and effort- accuracy tradeoffs. I present a semantic judgment task similar to that used in Experiment 2, but incorporating design principles common in studies on decision making, such as response-contingent financial payoffs and trial-by-trial feedback on response accuracy. I show that participants' preferences for deep and shallow semantic processing strategies are predictably influenced by factors known to affect decision making in other non-linguistic domains. In lower-risk situations, participants are more likely to accept default meanings even when they are not contextually supported, such as responding "True" to stimuli such as The dawn broke -- Something shattered, even without the presence of time pressure. In Experiment 3, I additionally show that participants can adjust not only their processing strategies but also their stimulus acceptance thresholds. Stimuli were normed for truthfulness, i.e. how strongly implied (or entailed) a probe sentence was given its context sentence. Some stimuli in the task posessed an intermediate degree of truthfulness, akin to implicature, as in The log burned -- Something was dangerous (truthfulness 4.55/7). Across 3 conditions, the threshold separating "true" from "false" stimuli was moved such that stimuli such as the example just given would be evaluated differently in different conditions. Participants rapidly learned these threshold placements via feedback, indicating that their perceptions of meaning-in-context, as expressed via the range of possible conclusions that could be drawn from the verb, could vary dynamically in response to situational constraints. This learning was additionally found to occur both faster and more accurately under increased levels of risk. This thesis makes two primary contributions to the literature. First, I present evidence that our knowledge of verb meanings is at least two-layered -- we have access to a very information-rich base of event knowledge, but we also have a more schematic level of representation that is easier to access. Second, I show that these different sources of information enable different semantic processing strategies, and that moreover the choice between these strategies is dependent upon situational characteristics. I additionally argue for the more general relevance of the decision making literature to the study of language processing, and suggest future applications of this approach for work in experimental semantics and pragmatics.