Mayfest 2016: Context

9:00-9:30, Breakfast
9:30-9:45, Opening remarks
9:45-10:55, Ernest Lepore, Psychology and grammar
10:55-12:05, Robyn Carston, Word meanings and contextual assumptions
12:05-2:00, Midday break
2:00-3:10, Craige Roberts, Pragmatics in context
3:10-4:20, Lyn Frazier, Context effects on interpretation: The QUD
4:20-4:40, Coffee break
4:40-5:50, Petra Schumacher, Context as a guide for expectation and dynamic discourse construction

9:30-10:00, Breakfast
10:00-11:10, Mandy Simons, Questions as context
11:10-12:20, Shevaun Lewis, Context in language acquisition
12:20-2:30, Midday break
2:30-3:40, Andrew Kehler, Conversational Eliciture
3:40-4:50, Stephen Neale and Daniel Harris, Meaning intentions and the composition of sentence meaning
4:50-5:00, Coffee break
5:00-5:30, Discussion

In our book Imagination and Convention, Matthew Stone and I challenge the standard philosophical picture of communication, according to which in addition to grasping the linguistic meaning of an utterance, hearers invokes a variety of extra-semantic mechanisms, ‘pragmatic mechanisms’, in order to retrieve information beyond the semantic content that gets communicated. For example, according to Grice, hearers expect speakers to be cooperative, and so, to make contributions suitable to the purposes of the conversation and apt to the stage of the conversation. When an utterance with a certain semantic content is produced, the hearer presumes the speaker is being cooperative and this requires the hearer to impute various beliefs to the speaker. By these means, speakers succeed in communicating all sorts of information beyond conventional meaning. There are interesting differences between this Gricean orthodoxy and later departures but what they all share is a relatively thin conception of meaning, with most information transfer via linguistic communication relying on extra-semantic mechanisms.

Stone and I argue that in many paradigm cases where theorists have invoked pragmatic mechanisms to explain information transfer, the correct explanation is semantic. To show why we believe so, we defend a variety of linguistic conventions ignored by philosophers, but which replace standard pragmatic account of communicative patterns. These conventions concern tense and aspect, discourse coherence, and information structure. I will rehearse some of these in my presentation.

Lastly, we argue that even cases of communication that don’t seem conventionally governed, e.g., metaphorical interpretation, it is wrong to think that any communicative success is to be explained by cooperation or general interpretative principles. In these cases, effective communication relies on distinctive mechanisms of imaginative engagement.

It is a matter of contention whether the standing (encoded) meaning of substantive words (most nouns, verbs, adjectives) is fully semantic/conceptual or is semantically/conceptually underspecified. Either way, though, the concept a speaker communicates by a word in a context of utterance is typically distinct from (albeit related to) the standing meaning of that word. Using the relevance-theoretic (RT) framework (Sperber & Wilson 1995; Carston 2002, 2015), I will sketch a pragmatic account of how a hearer recovers the concept the speaker communicated by her utterance of a word on a particular occasion.

The operative notion of context here is that of a set of contextual assumptions (truth-evaluable mental representations), which are accessed by the addressee, as intended by the speaker, and used as premises in the inferential process of utterance interpretation. This is clearly quite different from the formal semantic notion of context as an index consisting of values of parameters such as speaker, addressee, time/place of utterance. It is also different, but less obviously so, from the notion of context as interlocutors’ common ground. I will attempt to spell out some of these differences and to indicate how the RT conception of context meshes with psycholinguistic work on language processing.

If we want to study how context influences interpretation, it behooves us to have a theory of what a context is. I offer some desiderata for such a theory, based on observations about the contextual factors that play a role in anaphora resolution, association with focus, speech act determination, and intrusive implicature. Each of these pragmatic phenomena gives independent evidence for a conception of context as a complex database (cf. Lewis’ 1979 scoreboard) whose central components are bodies of information about the interlocutors’ common ground and their evident goals, including discourse goals (the questions under discussion and their associated strategies of inquiry). Moreover, we see that this database must be dynamic, capturing Heim’s (1982) distinction between global contexts and local contexts; context change at the merely local level comes to bear in a truth conditionally important way on compositional interpretation.

Once we characterize context in this way, it facilitates satisfactory, non-ad hoc accounts of a broad range of prima facie unrelated pragmatic phenomena, including those pertaining to the information structure of utterances as reflected in their lexical content and surface form (presupposition, focus, topic, clefting, etc.), as well as conversational implicature. A theory of context of the sort proposed here appropriately drives and constrains pragmatic enrichment, yielding implicatures, while illuminating relationships between contextual information and surface form. This is in contrast to accounts that focus on the generation of conversational implicatures using free pragmatic enrichment or implicit operators at LF, accounts which are hard-pressed to account for such a wide range of phenomena and show their inter-relationships.

Context plays many roles in language interpretation. Drawing on work conducted jointly with Charles Clifton, I will focus on the Question Under Discussion (QUD) and its effects during adult language comprehension. I will begin with the role of overt questions, looking at how and when an overt question influences the processing of its answer. Among other things, a question influences assignment of focus in a direct answer, which in turn influences attention allocation (Cutler and Fodor 1979) and may trigger exhaustification of the new constituent in the answer (Frazier and Clifton in progress).

The role of implicit potential QUDs will then be addressed. One of the central claims of the QUD approach to discourse is that the QUD organizes discourse generally, not just when an overt question is present. For linguistic accounts of what makes a discourse coherent, it may suffice to identify an implicit QUD at the end of a sentence or discourse. But for processing accounts, the QUD is not particularly helpful if it is identified only AFTER processing the current sentence and its relation to discourse. Instead, it will be proposed, following Clifton and Frazier (2012), that potential implicit QUDs are identified during processing of a sentence when alternatives are postulated for independent reasons (in computing the denotation of a question, contrastive focus, or certain implicature alternatives), favoring prominent sets of alternatives, i.e., those focused or introduced in prominent positions. This approach allows potential QUDs to guide sentence and discourse processing as the discourse unfolds.

Time permitting, I will take up the question of how a general QUD comprehension principle (Favor analyses that relate new material to the QUD) relates to other processing principles. One question is whether prominence-based principles such as the Main Assertion Principle (Favor ellipsis antecedents that are prominent/part of the main assertion) can be reduced to the general QUD comprehension principle.

Context plays many roles in language interpretation. Drawing on work conducted jointly with Charles Clifton, I will focus on the Question Under Discussion (QUD) and its effects during adult language comprehension. I will begin with the role of overt questions, looking at how and when an overt question influences the processing of its answer. Among other things, a question influences assignment of focus in a direct answer, which in turn influences attention allocation (Cutler and Fodor 1979) and may trigger exhaustification of the new constituent in the answer (Frazier and Clifton in progress).
The role of implicit potential QUDs will then be addressed. One of the central claims of the QUD approach to discourse is that the QUD organizes discourse generally, not just when an overt question is present. For linguistic accounts of what makes a discourse coherent, it may suffice to identify an implicit QUD at the end of a sentence or discourse. But for processing accounts, the QUD is not particularly helpful if it is identified only AFTER processing the current sentence and its relation to discourse. Instead, it will be proposed, following Clifton and Frazier (2012), that potential implicit QUDs are identified during processing of a sentence when alternatives are postulated for independent reasons (in computing the denotation of a question, contrastive focus, or certain implicature alternatives), favoring prominent sets of alternatives, i.e., those focused or introduced in prominent positions. This approach allows potential QUDs to guide sentence and discourse processing as the discourse unfolds.

Time permitting, I will take up the question of how a general QUD comprehension principle (Favor analyses that relate new material to the QUD) relates to other processing principles. One question is whether prominence-based principles such as the Main Assertion Principle (Favor ellipsis antecedents that are prominent/part of the main assertion) can be reduced to the general QUD comprehension principle.

The notion of context is typically used as a cover term for many different sources of information – it is used with respect to the sentential context (including lexical and grammatical cues), the context of utterance (including the situation, the setting or context, the speaker, the speaker­-hearer relation), etc. Turning to language comprehension, context conceived in this very broad sense contributes to two basic functions: on the one hand, it drives the expectations that are generated during incremental processing. On the other hand, it engages in the dynamic construction of discourse representation with the aim of building a meaningful representation that often reflects speaker meaning (or rather what the hearer assumes to be speaker meaning for the time being). I will present data on the time course of language comprehension from the recording of event-­related brain potentials that support this functional distinction. These data show that i) context information is used while the interpretation unfolds; ii) context information can suspend very basic grammatical constraints (e.g., animacy requirements); iii) context impacts the dynamic updating of discourse representations. From the perspective of language processing, the functional contribution of context to very basic operations of the language architecture should thus be featured more prominently in models of meaning constitution.

In this talk, I will present a set of case studies, each of which demonstrates that the question which an utterance is understood to address has a very concrete effect on the interpretation of that utterance. These case studies will motivate the claim that Questions Under Discussion are a proper component of context.

I will first present simple examples that show that where an assertoric utterance is made in response to an explicit question, the assertion made depends on the question; and I will provide evidence that the same effects can be observed where the question is merely implicit.

I will then move to the familiar case of focus sensitive adverbs like only, demonstrating that the alternatives to which such adverbs are sensitive are not straightforwardly given by the compositionally determinable focus semantic value of the utterance; a correct account of the interpretation of sentences with only and its ilk requires reference to a contextually given set of alternatives, namely, a question. The case of focus sensitivity will also provide a first illustration of how the identification of the relevant question is guided by the interaction of utterance level constraints with discourse level constraints.

Finally, I will turn to the issue of projection – cases where an implication of an utterance is understood as a commitment of the speaker despite the presence in the utterance of an entailment canceling operator such as negation or an epistemic modal. I will examine two cases – sentences with a clausal complement, including standard cases of so-called factive and semi-factive predicates; and negated evaluative adjective sentences. Here too, I will illustrate how interpretation is dependent on the question under discussion; and will also show how the question itself is dependent on constraints from multiple sources.

I will conclude with some discussion of the status of questions as elements of context.

Every human utterance, no matter how simple or straightforward, is constrained by its context and intended purpose. Since children learn their first language through communicative interactions (not from, say, a text corpus of child directed speech), they must constantly engage with contextual information as well as the linguistic input in order to disentangle semantic and pragmatic levels of “meaning”. To fully explain language acquisition, we must determine what kinds of pragmatic competence are required to meet this challenge.
So what do we know about children’s pragmatic competence? The growing experimental pragmatics literature has produced some apparently contradictory results. In some tasks young infants seem to be impressively sensitive to context and speaker intention. In other tasks even older children are surprisingly insensitive to non-literal meaning. I argue that overall the evidence (from comprehension) suggests that core “pragmatic principles” are available to children early in development: if they manage to access all the relevant pieces of information, they know how to compute an appropriate meaning. However, in many cases children fail to access all the relevant pieces of information. That is, given the same context, they draw different conclusions than adults would about the purpose of the conversation and the information relevant to interpreting a certain utterance.

I discuss three general types of explanations for why children’s understanding of context changes over time:

(1) Children refine their expectations about the content and course of conversations as they gain experience participating in conversation in their culture.
(2) Children’s language processing mechanisms develop over time with experience comprehending language input. They become more efficient at representing and accessing contextual information for the purpose of real-time interpretation.

(3) Children’s language processing mechanisms become more efficient as their memory and control resources grow.

These different explanations are not in conflict; all of these factors may play a role. I discuss how each hypothesis fares with the existing evidence, and suggest paths for future research.

Whereas sentence (1a) states that the employee was fired and was embezzling money, it also strongly invites the inference that the employee was fired because of the embezzling. An analogous inference is lacking in (1b), however: one does not normally infer that the firing was caused by the employee’s hair color. And sentence (1c) leads to a counter­to­expectation inference, leading one to wonder why an employee with so many accolades would be let go.
(1a) The boss fired the employee who was embezzling money.
(1b) The boss fired the employee who has red hair.
(1c) The boss fired the employee who won numerous corporate awards.

We posit that these inferences do not follow directly from the procedures that have been argued to underlie other sorts of pragmatic enrichment, such as from a violation of communicative (e.g., Gricean) norms based on principles of rationality/cooperativity (as in IMPLICATURE), or the need to complete/expand a proposition so as to appropriately fix truth­conditional content (as in Bach's IMPLICITURE). We argue instead that they follow from more basic, general cognitive (not specifically linguistic) strategies for building mental models of the world that draw on types of experiential knowledge and associative principles that are known to be used to establish the coherence of passages across clauses. For want of a term of art, we brand the phenomenon as ELICITURE, selected to capture the fact that a speaker, by choosing a particular form of reference, intends to elicit such associative inferences on the part of her hearer. The importance of accounting for such inferences goes beyond the recovery of implicit communicated content; it is also crucial for the interpretation of explicit linguistic expressions. As an example, we present a case study that examines pronoun interpretation. A passage completion experiment was conducted using stimuli like (1a­b) as context sentences, presented to participants with or without an additional pronoun prompt. Whereas accounts of pronoun interpretation that appeal primarily to surface­level contextual factors find little to distinguish contexts (1a­b), a Bayesian analysis (Kehler et al. 2008; Kehler & Rohde 2013) predicts a difference, through an interconnected chain of referential and coherence-­driven dependencies. The results confirm that pronoun interpretation biases, but not production biases, are sensitive to whether or not an implicit cause can be inferred from a relative clause, revealing precisely the asymmetry predicted by the Bayesian analysis.

It is often said that context plays a role in constitutively determining what is said when a sentence is used. One common idea is that relative to contexts expressions have composable meanings (usually called “semantic contents”), with the composed meaning of a sentence usually being a proposition. Many semanticists appeal to formalized notions of contexts in working out such ideas. On some accounts, contexts are ordered n-tuples of entities, for others they are bodies of shared representations. The first two points we argue for in this talk, are:

(1) The general idea of contexts being partial determinants of what is said is wrong-headed;

(2) The only theoretically relevant notion of what is said is the Gricean notion of what a speaker says in uttering a sentence on a given occasion, which is analyzable in terms of speakers’ meaning intentions.

For Grice, what S means by uttering x is wholly determined by the meaning intentions with which S uttered x. And what S says in uttering x is wholly determined by those intentions together with the linguistic meaning of x. We demonstrate that

(3) The Gricean position correctly assigns no role to context in determining what is said, the roles of context being exhausted by their roles in the formation of (a) speakers’ meaning intentions and (b) hearers’ hypotheses about the contents of those intentions.

We then explore the “super-Gricean” position that

(4) What is said is wholly determined by a subclass of the meaning intentions with which S produced x that can be isolated without appealing to the linguistic meaning of x.

The meaning of a sentence x can be viewed as a type, a property of meaning intentions. As such, it provides hearers with incomplete evidence about the meaning intentions of a speaker uttering x. Since we reject the traditional method of composing semantic contents—meanings partially determined by context—we are left with the problem of specifying (a) the fixed composable meanings of subsentential expressions and (b) the mechanisms of meaning composition, which together yield of properties of meaning intentions as the composed meanings of sentences. In the final part of the talk we illustrate a method for transposing and extending standard intensional semantic theories to do what is needed.