Marie Nilsenova, University of Amsterdam/LHSP

Title: Optimal Interpretation of Tense in Discourse

Abstract:

Like other anaphoric categories, tense in general cannot be interpreted solely from the compositional meaning of a single clause. The reason is that in most cases, the time of an event is interpreted with respect to another time, established in the context. It has been argued that temporal relations between events in a discourse can be derived from general discourse relations. In DICE, the framework proposed by Lascarides, Asher and Oberlander, this is done in a system of defeasible rules. In my presentation, I would like to show that a number of stipulations in their system can be removed and various problems can be solved if we treat tense interpretation in bidirectional OT (Blutner 2000). In particular, it is possible to account for temporal relations between eventualitites if we take them to be the result of a game between the speaker's and hearer's principles (Dekker & van Rooy 2000). I will focus on the interpretation of English pluperfect and propose that it can be treated as the marked form of expressing past. At the end of the presentation, some questions concerning markedness will be raised. It seems that purely linguistic knowledge of the structure of events is insufficient to determine the temporal structure of a discourse. A sequence of past tenses can sometimes be interpreted as temporally ordered, while other times there is no particular order inferrable (Lascarides & Asher 1994, de Swart & Verkuyl 1999). In a series of papers, Lascarides, Asher and Oberlander propose that temporal relations between events are, in fact, derived from general discourse relations. This happens in a system of defeasible rules (DICE) which are based both on the linguistic knowledge, as well as the world knowledge of conversation participants. The authors argue that their approach can account for a number of problems that traditional Reichenbachian accounts cannot solve. Like for other tenses, for pluperfect these are the Interaction Problem (sequence of pluperfects can be both temporally ordered and temporally undetermined) and the Relevance Problem (incoherent discourses), topped with the pluperfect-specific Perspective Problem (pluperfect can change the point of view [POV] in discourse). The Lascarides/Asher/Oberlander framework is inadequate in a number of aspects: (i) analyzed in detail, the proposed semantics for pluperfect turns out to be stipulated and unable to account for a number of examples, (ii) in general, the rules in DICE rely mainly on lexicalized "natural order of events" and fail to take into account the actual conversational situation. Lascarides&Asher (1993, 1994; henceforth L&A), propose that pluperfect is a discourse marker which indicates that only a limited set of discourse relations (Elaboration, Explanation, Parallel and Contrast) can be established between two sentences. The explanation is motivated by the backwards moving character of pluperfect ('Constraint When Changing Tense' in L&A). However, there are inconsistencies in the system. First, L&A contradict their own proposal when also Narration and Background turn out to be needed to solve the Interaction problem. Second, Background is derivable for most examples with pluperfect from the principles as they are formulated. In consequence, the Relevance Problem can no longer be accounted for (cf. (1) and (2)).

(1) ?Max poured himself a cup of coffee. He had entered the room.

(2) ?My car broke down. The sun had set.

In my opinion, it is arguable that (1) and (2) really create incoherent discourses, as claimed by L&A. Sperber & Wilson (1986) have shown that the maxim of relevance is hard to violate since the hearer will usually prefer to adjust her context set to accommodate the problematic utterance. But even so, the preferred DR for connecting the sentences in (1) and (2) does not appear to be Background, which is exactly the relation that gets derived by DICE (while L&A claim that no relation can be derived at all). Another set of L&A's counterexamples to the traditional approaches, specifically to Kamp & Rohrer (1983), is the Perspective Problem - cf. (3):

(3) a. The telephone rang.
b. It was Mme Dupont.
c. Her husband had eaten too many oysters for lunch.
c'. Her husband ate too many oysters for lunch.

While in (3c') the author's POV is preserved, in (3c) it is replaced with the perspective of the character (Mme Dupont), creating a free indirect style. L&A's account depends on postulating a specific lexical entry for the item 'telephone'. This explanation fails to capture the generality of the phenomenon, however, since it has been noted that pluperfect (if in the main clause) matches with character's POV as a rule (Wiebe 1994). Yet another problem for L&A is the fact that while simple past tense can often stand alone, as in (4a), the corresponding (4b) in pluperfect is not possible (unless a part of a larger discourse). DICE, however, cannot account for the questionability of (4b).

(4) a. We forgot to turn off the gas.
b. ?? We had forgotten to turn of the gas.

I will propose that the above encountered problems can be solved in a straightforward manner and temporal interpretation in general greatly simplified if we take it to be derived in a game of interpretation, as described in Dekker & van Rooy (2000). The game of interpretation is due to two conflicting principles, the Q-principle vs. the I-principle, which have been defined in various ways. I will be using the reformulation of Blutner (1998) according to whom the I-principle aims for the most coherent interpretation, while the Q-principle acts as a mechanism that blocks uneconomical/marked outputs. Building on Lascarides & Asher (1994), I assume that temporal interpretation can be governed by fout DRs, two of which cause backward movement of time (Explanation and Elaboration) and two of which preserve the time perspective in the order in which it is presented (Narration and Result). I will also take into consideration whether the ordering of events is in accordance with the structuring of the discourse segment. In the particular case of pluperfect, I will suggest that it is simply a marked way of expressing past tense and use the Dekker&van Rooy framework to account for its distribution. Consider the following examples:

(5) a. Max stood up. John greeted him.
b. Max stood up. John had greeted him.

The interpretation of (5a) as Narration is optimal because no causal connection can be established. In (5b), on the other hand, Narration is blocked by the use of a marked expression and a different interpretation has to be found. In this particular case, it will be one under which the eventuality in the second sentence preceded the culmination point of the event in the first clause (Explanation/Elaboration). The same reasoning can be applied to a number of other examples, including (1) and (2). The pluperfect's ability to shift POV from the author to a character again appears to be due to markedness of the form. As pointed out by Wiebe (1994), POV does not typically shift unless explicit information is provided. Since in (3a,b), events are presented from the POV of the author, the perspective in (3c) will remain the same if the unmarked simple past form is used. The fact that its use is blocked in (3c') indicates to the hearer that a different interpretation needs to be searched for. Finally, it is easy to show that in terms of discourse interpretation, if only one pair is available in (4), it has to be the optimal candidate (and hence the discourse topic). If this candidate is blocked and the marked form of past tense used instead, as in (4b), there is no possible meaning that can be ascribed to the sentence. Clearly, the notion of markedness plays a central role in the suggested account which raises at least the following issues:

(i) Exactly how does the least-effort explanation account for markedness phenomena? For example, Horn (1984) shows that periphrastic forms are not always more costly than single lexical forms (idiomaticity also plays a role). Also, it is not obvious whether we should consider more economical the use of (5b) for the Explanation/Elaboration case, rather than the use of a sentence connective such as 'because'.

(ii) Zeevat (2000) suggests that examples of marked expressions treated by Blutner (2000) in his weak bidirectionality can be recast as constraints on generation, namely PARSE-MARKED >> ECONOMY. While ECONOMY tells the speaker to use the expression that requires least effort, the stronger constraint PARSE-MARKED requires that markedness be preserved. However, applied on the above examples with temporal anaphora, it is clear that whether a certain interpretation is marked can only be decided once hearer's perspective is taken into consideration. It thus appears that the case of tense interpretation poses a difficulty to Zeevat's proposal.

References:
Blutner, R. (1998) Lexical Pragmatics. Journal of Semantics 15:115-162.
Blutner, R. (2000) Some Aspects of Optimality in Natural Language Interpretation. In H.de Hoop (ed.): Papers on Optimality Theoretic Interpretation. Utrecht University.
Dekker, P. & R.van Rooy (2000) Optimality Theory and Game Theory: Some Parallels. In H.de Hoop (ed.): Papers on Optimality Theoretic Interpretation. Utrecht University.
Horn, L.R. (1984) Towards a new taxonomy for pragmatic inference: Q-based and R-based implicatures. In D.Schiffrin (ed.): Meaning, Form, and Use in Context. pp. 11-42, Washington, Georgetown UP.
Kamp, H. & C.Rohrer (1983) Tense in texts. In Bauerle, R., Schwarze, C. and von Stechow, A. (eds.): Meaning, Use and Interpretation of Language, pp. 250-269. Berlin: de Gruyter.
Lascarides, A. & N. Asher (1993) A Semantics and Pragmatics for the Pluperfect. In Proceedings of the 6th European Chapter of the Association of Computational Linguistics, Utrecht.
Lascarides, A. & N. Asher (1994) Temporal Interpretation, Discourse Relations and Common Sense Entailment. L&P 16:437-493.
Oberlander, J. & A. Lascarides (1991). Discourse Generation, Temporal Constraints, and Defeasible Reasoning. ms., University of Edinburgh.
Sperber, D. & D.Wilson (1986) Relevance. Blackwell Publishers.
de Swart, H. & H.Verkuyl (1999) Tense and Aspect in Sentence and Discourse. ms. Utrecht University.
Wiebe, J. (1994) Tracking Point of View in Narrative. Computational Linguistics 20:2, pp.233-288.
Zeevat, H. (2000) The Asymmetry of Optimality Theoretic Syntax and Semantics. ms., University of Amsterdam.