Abstract
As discussed in Chapter 1, tensed sentences are often interpreted referentially/presuppositionally, picking out a time described by the preceding sentence. This anaphoric connection is illustrated in (1) below.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Preview
Unable to display preview. Download preview PDF.
References
The formal mechanisms needed to get the second sentence’s temporal anaphor to take the interval introduced by the first sentence as an antecedent are fairly straightforward. All that is needed is a contextually-determined variable assignment function, one which will assign an index to the interval introduced by the first sentence. The presuppositional free variable in the second sentence is then simply subscripted with this familiar index, yielding coreference under the current variable assignment. The question of what interval(s) is introduced by the first sentence is much trickier, however. Webber (1988) provides evidence showing that an event-depicting sentence introduces multiple intervals, associated with different subparts of the event. Each of those subintervals (as well as the interval associated with the full runtime of the event) is available for the RefT of the following sentence to refer to. I have no clear idea of how the introduction of multiple intervals might be implemented formally, unfortunately. See section 6 below for some discussion of Webber’s data and insights.
The reason for this preference is uncertain, but it may be related to the intuition that Dutch native speakers have regarding the present perfect version of (4). That is, it could be that the current sentence must take the value for its temporal anaphor from the preceding sentence in order to be rhetorically related to it. This assumption is at least implicit in many treatments of tense in discourse (cf. Dowty, 1986, for example), and it is made explicit by Trueswell and Tanenhaus (1991). They argue that not taking a clause’s RefT from the preceding sentence has the effect of signalling a new discourse segment, indicating that the current sentence is not to be connected to the preceding one. If correct, this assumption may explain the difficulty of creating future-tense discourses of the sort used in the studies in 3.5 and 3.6 below. Future-tense sentences take the utterance time as their RefT, rather than a future time provided by the preceding sentence (see Kratzer, 1997, 1998). Therefore, a future-tense sentence cannot be directly rhetorically related to the preceding one. Connected discourses in the future tense are much more difficult to create as a result. For additional arguments that past tense sentences in narratives must take an interval from the preceding sentence as their RefT, rather than the utterance time, see Caenepeel and Oversteegen (1993).
Ogihara and Musan make their proposals in static frameworks, as does von Fintel (1994) for adverbial and modal quantification. For dynamic approaches to domain restriction and its effect on quantifier interpretation, see Roberts (1995) looking in particular at modal quantification. See also Groenendijk, Stokhof, and Veltman (1996) for a competing dynamic-semantic approach to the same phenomena.
See also van Geenhoven (1999) for a formalization of Klein’s (1994) approach to tense, which also makes use of implicit quantification over times and a restriction on that quantification which confines it to familiar times only.
Interestingly, there are examples similar to (6) (pointed out to me by Angelika Kratzer) in which lifetime effects persist even in a past-tense context. Take (i): (i) My sister was born in 1963. She had blue eyes. Barring the reading under which “have blue eyes” is coerced into a stage-level interpretation, the implicature that my sister is now dead remains for (i). I have no idea why (i) should differ from (6). I will leave aside this problem here.
The discussion here has been confined to static models of temporal interpretation, but I believe the same points should hold for dynamic approaches which also make use of quantifier restriction. See Roberts (1995) for some discussion and consideration of how such an approach would work for temporal anaphora phenomena. See also Dekker (1993) for how a view of tense involving simple existential quantification can be modelled in a dynamic framework. Dekker makes use of an operation he calls Existential Disclosure to “open up” temporal indices which have been quantified over and make them available for further manipulation by context-change algorithms, to model tense in discourse.
As Veerle van Geenhoven (p.c.) and Sandro Zucchi (p.c.) point out, there is nothing inherently incompatible between a quantificational and a referential approach to tense, at least not an indirectly referential one. In fact, the domain restriction approach to capturing the context-dependent portion of temporal interpretation could be considered simply a different formalization of approaches like Partee’s (1984) DRT-based analysis. This is how Roberts (1995) frames her discussion of Reference Times as domain restrictions, in fact: she describes domain restriction as one formal way of characterizing the semantic contribution of context-dependent elements like Reference Times. It is not clear to me whether the two alternatives are anything more than notational variants of one another, in their current forms.
The function of establishing a new discourse referent seems to be confined to definite NPs, however. See Garrod (1994) for discussion and Farurud (1990) for corpus-based evidence regarding the distribution of pronouns and definite NPs and the establishment of new referents.
Speelman and Kirsner also looked at the effect of topicaliztion on the resolution of anaphors, by manipulating whether the antecedent was introduced as the subject of an active-voice clause (in default topic position) or in an agentive by-phrase in a passive sentence (in non-topic position). This manipulation will not be discussed here. See Speelman and Kirsner (1990) for discussion of the effects topicalization has on pronoun resolution.
Additional integration-based evidence that both pronouns and definite NPs are resolved immediately comes from an auditory study by Marlsen-Wilson et al. (1993). They found that pronouns which were disambiguated by gender and were in attentional focus (their antecedents were the topic of the preceding passage) showed immediate plausibility effects like the ones Garrod, et al. found. That is, when the pronoun’s antecedent didn’t fit with the role the pronoun was playing in the critical sentence, subjects’ responses in naming the pronoun aloud were immediately slowed. Also, in a recent ERP study, van Berkum, Brown, and Hagoort (1999) found that subjects exhibited a long, low negativity starting 300–350 milliseconds after hearing a definite NP if the story they were listening to did not provide an unique antecedent for the NP. Interestingly, this negativity was similar to patterns associated with searches of long-term memory. This result suggests that subjects are immediately consulting some representation of preceding context to find an antecedent for the definite NP.
See also Heim (1982) and Kadmon (1987) for linguistic evidence that pronouns require a unique salient antecedent in preceding context in order to be felicitous.
This distinction presents an interesting parallel with the “discourse lifetime” effects described by Heim (1982), as Sungryong Koh (p.c.) points out. Heim notes that definite NPs have a longer “lifetime” in discourse than pronouns. They may be referred back to for longer than pronouns can, for example. This behavior goes along naturally with the immediacy of pronouns’ processing: pronouns trigger immediate deep processing because their natural lifespan in the discourse is shorter. This linguistic distinction may ultimately be tied to the different treatment of pronouns and definite NPs by the processor.
See Webber (1991) and in particular Maes (1997) for some evidence that more abstract referents are somehow naturally less ‘salient’ than concrete ones.
Frazier couches her claim in the representational assumption that at some level, a clause containing a quantifier is split up into a tripartite structure consisting of the quatifier, a restriction on it, and the nuclear scope. This view makes the quantifier restriction a distinct entity from the quantifier, either in a sentence-level semantic representation (Kratzer, 1989; Diesing, 1992) or in a discourse representation (Heim, 1982). A slightly different but potentially compatible view of this process should yield the same result. Von Fintel (1994) and others have argued that formally, the contextual dependence of quantifiers is best modelled by positing a free variable associated with the quantifier domain whose value is fixed by context. This characterization of domain restrictions on quantifiers should produce the same predictions as Frazier’s approach: the quantifier has an anaphor associated with it, which will cue the parser to look to context to resolve it. On both approaches, quantifiers serve as an (indirect) cue to consult context. The latter approach is the one assumed for the Anaphoric Cue Hypothesis proposed in Chapter 1, unifying domain restriction with other forms of anaphoric dependencies.
The [Hown] in the Specifier of CP in these representations is the element that actually does the questioning about the set of papers introduced in [tn-many papers]. See Cresti (1995) and Villalta (1999) for more details regarding the representation of how many questions, in both English and French (where the LF representations in (10a-b) correspond much more closely to the surface forms, particularly in split combien questions).
This description assumes that reduced relatives are tenseless, as do Trueswell and Tanenhaus.
Thanks to Alice ter Meulen for discussion of the potential significance of this fact.
A slightly different construal of this hypothesis, one which makes similar predictions, was suggested to me by Ken Drozd (p.c.). Under this alternative version, temporal anaphora resolution takes place early, with the temporal anaphor picking out the utterance time in the future-tense contexts. The content of the event description then indicates that a perfect operator (and the indefinite interpretation associated with it) is inappropriate for the particular event involved. This creates a clash effect at the end of the event description in the future-tense contexts. As will be described below, however, the predictions of Minimal Events aren’t borne out, on either this early-resolution version or the global-delay version described in the text.
These subevents/subintervals associated with the events in these examples are based on independently-motivated analyses of event structure (cf. Moens and Steedman, 1988; Pustejovsky, 1991, 1995). Roughly, a telic event has a preparatory phase, followed by a result state. A culmination or transition point may separate the preparatory phase from the result state. An atelic event has only a single phase, in contrast. Dowty (1979), Pustejovsky (1991), and others have argued that telic verbs are essentially atelic event descriptions with a result state added: the preparatory phase is an Activity (cf. Vendler, 1967), and the result state is just a State. Atelic verbs/event descriptions involve just an Activity. This level of detail is beyond the present concerns; see Pustejovsky (1991) and van Hout (1996) for particularly good summaries of the issues involved.
The fact that Villalta (1999) found effects of preceding context on fairly subtle interpretive decisions involving how many questions, also using a self-paced method, is further argument against this account of the delay. In fact, her results come from the same experiment as Study 3: her experimental items were run as fillers for the Study 3 items. The delay effect seen for Study 3 therefore seems even less likely to be the result of the insensitivity of the technique or the particular participants involved.
Interestingly, this picture of temporal relations makes them similar to the processes Garrod (1994) and Garrod, et al (1994) claim are involved in definite NP anaphora resolution. Garrod and colleagues argue that definite NPs are resolved in two steps: first, an antecedent is located via fast and automatic processes, and second, coherence relations are used to determine the exact relation which the definite NP bears to that antecedent. If the results reported here are genuine, there may be further points of comparison between definite NP resolution and the processor’s handling of temporal relations. For example, the kinds of part-whole or evoked relations that definite NPs can bear to their antecedents — and the manner in which those relations are processed — may have parallels in the temporal domain.
Author information
Authors and Affiliations
Rights and permissions
Copyright information
© 2001 Springer Science+Business Media Dordrecht
About this chapter
Cite this chapter
Dickey, M.W. (2001). Interpreting TenseP. In: The Processing of Tense. Studies in Theoretical Psycholinguistics, vol 28. Springer, Dordrecht. https://doi.org/10.1007/978-94-010-0568-5_3
Download citation
DOI: https://doi.org/10.1007/978-94-010-0568-5_3
Publisher Name: Springer, Dordrecht
Print ISBN: 978-1-4020-0185-7
Online ISBN: 978-94-010-0568-5
eBook Packages: Springer Book Archive