Domain Restriction in Generic Statements

ABSTRACT

Domain restriction is a pervasive if often neglected part of discourse comprehension. Speakers and authors implicitly limit the domain of discourse of quantifiers (e.g., everyone) and noun phrases (e.g., the girls). Our previous research shows that an initial temporal or locative prepositional phrase (PP), which introduces a topic situation, is the preferred context for interpreting the domain of subject noun phrases in the same sentence or a subsequent sentence. In this research, the sentences and minidiscourses were about concrete nongeneric situations. The present article reports three experiments designed to explore domain restriction in generic statements. Experiment 1 shows that indefinite determiners and the presence of a modal (would) favor generic interpretations. In Experiments 2 and 3 a sentence initial (vs. sentence final) PP increased interpretation of the subject as being restricted to the topic situation as well as increasing the frequency of generic interpretations. We propose that in sentences with modals (would) the initial PP acts as the antecedent of a conditional. Furthermore, based on differences in responses depending on whether the genericity of the subject or the temporal restriction on the subject are posed, we suggest that domain restriction is an inference from the representation constructed during discourse comprehension while a decision about genericity is a part of core sentence interpretation.

Acknowledgments

We are grateful to Ana Arregui and Vincent Homer for talking with us about this research.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1. The authors did a post-hoc classification of the Experiment 1 items into those whose predicates seemed likely to be episodic versus stative and into items whose predicates seemed more or less stereotypically related to their subject. There seemed to be a hint that GEN interpretations were more frequent for stative than for episodic predicates and for items with stereotypical relations to the subject. However, since this was not actually manipulated in any principled fashion, we simply note it as a possible avenue for further study and a possible factor to control in future experiments.

2. Because of the failure of non-anticonservative maximal glmers to converge, we conducted F1/F2 ANOVAs on empirical logit transformations of the data (Jaeger, 2008), substituting proportions of 0.99 and 0.01 for observed proportions of 1 and 0. The results were congruent with the reported glmers: for effect of modal, F1(1,54) = 90.09, F2(1,14) = 88.66; effect of article F1(1,54) = 36.57, F2(1,14) = 39.02; and interaction F1(1,54) = 13.85, F2(1,14) = 14.18, all p < .001.

3. In the postexperiment questionnaire, 44 of 56 participants gave some indication that they believed, after reading the description of our hypotheses about the presence of the modal influencing the in general versus particular interpretation and the effect of the sentence-initial temporal PP on the BW answers, that their answers conformed to these hypotheses. In contrast, only 20 participants gave any indication that they were aware of the effects we were looking for during the experiment itself (and in many of these cases, their answers indicated they were not really answering our question, e.g., Yes. I think so. Some questions were to my ears ambiguous). Similar results were obtained in the postexperiment questionnaires of Experiments 2 and 3, which are not presented.

4. A definite determiner may receive a generic interpretation (The labradoodle has a sweet disposition). So there is no literal inconsistency between the generic-biasing modal and the definite determiner. But it may be more difficult to determine what interpretation the speaker is trying to convey when the determiner is definite (not generic biasing) and a modal is present (generic biasing). By comparison, if the determiner is indefinite and a modal is present postulating, a GEN operator is sufficient to explain the speaker’s decision to use an indefinite and the decision to include a modal. This raises interesting questions that have to be deferred to future research.

5. An additional 29 participants were tested, but due to an apparent host system failure their data were lost before they could be analyzed.

6. As in Experiment 1, these results were generally confirmed by empirical logit ANOVAS. The effect of article was fully significant, F1(1,39) = 60.87, p < .001; F2(1,15) = 83.01, p < .001. The effect of PP position reached significance (F1(1,39) = 5.36, p < .03; F2(1,15) = 7.34, p <. 02). The interaction did not approach significance.

7. Again, this analysis was checked by performing empirical logit ANOVAs. The by-subjects (F1) analyses confirmed the mixed models reported in the text, but the by-items analyses did not all reach conventional levels of significance, calling the results into some question. For article type, F1(1,39) = 19.09, p < .001; F2(1,15) = 6.34, p < .03; for PP position, F1(1,39) = 9.27, p < .01; F2(1,15) = .2.28, p < .20. We can only conclude that the effect of PP position is inconsistent across items.

8. Once again, these results were checked by performing empirical logit F1/F2 ANOVAs. All results were strongly confirmed except for the by-items analysis of the main effect of temporal PP position on the proportion of GEN responses. Here, the difference failed to reach significance, F2 (1,15) = 3.68, p = 0.074.