Mira Ariel

The long experimental road from semantic meaning to pragmatic Truth-Compatible Inferences

Following Grice (1989) and Horn (1989), scalar quantifiers, as well as or constructions, have been assumed to carry only lower-bounding semantic meanings. Most then means ‘more than half (and possibly all)’, and X or Y means ‘at least one of X and Y (and possibly both)’. Their respective upper-bounded actual interpretations (commonly assumed to be ‘more than half but not all’, ‘one of X and Y, but not both’) were derived by routinely adding on pragmatic upper-bounding scalar implicatures (‘not all’ for most, ‘not both’ for or).

I identified three problems with this analysis in the early 2000s. First, a lower-bounding meaning denotes a complex concept, which should more likely be expressed compositionally (e.g., at least above half, X and/or Y), rather than by a single lexeme (Note that we don’t have a single lexeme for the only lower-bounding ‘less then all/half’ or the only upper-bounding ‘up to half/all’). Second, it is not reasonable to assume that it is the default (upper-bounded) interpretation that is pragmatically derived, while the lower-bounded interpretation, which is barely (if ever) attested in discourse, is the semantic core. Third, pragmatic implicatures are speaker-intended messages, but ‘not all’ and ‘not both’ don’t function as speaker-intended messages, except in contrastive contexts (based on the Santa Barbara Corpus, Du Bois et al., 2000-2005). For example, most Xs…, supports a discourse argument based on the majority set of Xs. But ‘not all’ potentially supports the opposite argumentative direction, based on the minority complement set. The consensus analysis has to assume that speakers routinely generate implicatures which counter their own discourse points.

The typical discourse use of most, I discovered, is indeed lower and upper-bounded, but the upper bound does not come about from an ‘all’ rejection implicature, because the speaker’s argumentative point is better served when the minority complement set is not at all profiled. Just like typical lexemes, most too, I proposed, has a circumbounded semantic meaning (developed out of meaning of ‘the lion’s share’): ‘a proper subset larger than 50%’, where both the lower- and the upper-bounded aspects are lexically specified.

I then had to account for attested True judgments by participants for e.g., most Xs in the face of an ‘all’ state of affair. Such judgments seem to testify to the availability of the lower-bounding meaning after all. Developing Koenig‘s (1991) distinction between semantic meanings and states of affairs merely compatible with such meanings I proposed that most (and later some) have circumbounded meanings, which can (but crucially, need not) be considered as compatible with the subsuming ‘all’ state of affairs.

I therefore needed to show a differential participant behavior re meaning versus re truth judgments. Specifically, that the latter do not actually tap into the purely speaker-intended meaning, because they may involve Truth-Compatible Inferences (TCIs) in addition. These potentially bridge the gap between the (circumbounded) speaker-intended meaning and the ‘all’ state of affairs. Thus began my long road from my initial theoretical hunch to experimental substantiation. I needed to show that participants distinguish between the speaker-intended meaning and states of affairs potentially compatible with those meanings.

Ariel (2004) presented differential results from two experimental tasks. I have later dubbed these Participant-Control tasks and After-the-fact tasks (Shetreet and Ariel, 2023). In a Participant-Control task, it is the participant who controls the correspondence between the target expression and the state of affairs intended by the speaker. Picture matching tasks are one such task. These tasks, I predicted, would overwhelmingly yield circumbounded responses. Indeed, in such a task participants refused to confirm that the speaker could have intended ‘100%’ for most, even when no proper upper-bounded option was provided (all other options were less than 51%). Participants consistently preferred a ‘none of the above’ response over the supposedly perfectly legitimate semantic lower-bounded ‘100%’. At the same time, when presented with a target most… against an ‘all’ state of affairs (the correspondence here established by the experimenter) the same participants were sometimes willing to confirm as True the most sentence against an ‘all state of affairs. Crucially, however, the rate of these lower-bounded responses were lower than the upper-bounded responses in the Participant-control task, and they were not at all consistent.

This is exactly what I predicted. When the participant controls the correspondence between the target expression and the speaker-intended state of affairs, we only get circumbounded responses. When the task allows participants to consider after-the-fact TCIs some of them mobilize those and judge the sentence true.

In the following 20 some years I have employed this two-task examination in some and in or studies too. Results for all three expressions have been consistent: Circumbounded readings are consistently and overwhelmingly chosen when the participant has control over the correspondence between the expression and the state of affairs. Lower-bounded responses only show up in After-the-fact tasks, and only inconsistently so.   

References

Ariel, Mira. 2004. Most. Language 80:658–706.

Du Bois, John W., Wallace L. Chafe, Charles Meyer, Sandra A. Thompson, Robert Englebretson and Nii Martey. 2000-2005. Santa Barbara corpus of spoken American English, Parts 1-4: Philadelphia: Linguistic Data Consortium.

Grice, H. Paul. 1989. Studies in the way of words. Cambridge, Mass.: Harvard University Press.

Horn, Laurence R. 1989. A natural history of negation. Chicago: University of Chicago Press.

Koenig, Jean-Pierre. 1991. Scalar predicates and negation: Punctual semantics and interval interpretations. In Chicago Linguistic Society 27. Chicago: Chicago Linguistics Society, 140-155.

Shetreet, Einat and Mira Ariel. 2023. Taking control over weak scalar expressions. Tel Aviv University.