Narrative tasks

In chapter 3, we considered how listeners made sense of expressions used to refer to features in a static visual display where the only salient relations were spatial relations. To understand adequately the instructions given by the A-role speaker, a B-role listener had to locate a feature which was appropriately characterised by As description and ensure that it was in a location situated within a plausible search field. As we saw again in chapter 4, it was frequently the case that B could make sense of an underspecified message from A by locating a feature plausibly characterised by the denotation of the referring expression within a constrained search field, where the location is viewed by the listener from a particular perspective which is (largely) shared with the speaker.

We turn now to a different type of task, one which looks initially simple, but which turns out to be far more demanding for participants than the Map task. In the temporally structured task which I shall describe here, there are no enduring external aids to memory, no external model world whose stable features can be consulted, as in the Map task. Participants watch short episodes from a silent narrative which they see only once, enacted before them on a video screen. They then learn details of a further episode of the narrative which they were not shown, but which help them to interpret their own experience as they hear fellow participants talk about them.

This book is primarily concerned to give an account of how listeners behave as they participate in dialogues in which information is exchanged. What listeners have understood from what a previous speaker has said is frequently revealed in what listeners say themselves when they next take a turn at speaking. We shall examine what listeners say in their turn as speakers, looking for evidence of what they have understood previous speakers as saying. Detailed knowledge of such behaviour is fundamental to the development of our understanding of the cognitive basis of language use.

If we want to make claims about the nature of comprehension processes, we had better be sure that such claims will be widely applicable within the population. The subjects who participated in the dialogues illustrated here are drawn from a wide range of backgrounds and are diverse in their academic ability. They are drawn from Edinburgh schools and the University of Edinburgh, as well as from schools in the county of Essex and Essex University. They include young adults and adolescents of demonstrated intellectual ability, as well as 14-16 year-olds who are deemed by their schools to be in the bottom third of the academic ability range.

I am not concerned here to make quantitative claims. Rather, I draw attention to patterns of behaviour which occur repetitively in the data. The intention is to seek to explore the nature of the behaviour rather than to examine its detailed distribution.

In this chapter, I shall introduce the Map task method, and begin to explore some of its benefits and limitations as a window on the processes of interpretation. First, I shall set the approach in context by rapidly reviewing existing well-established research paradigms.

Methods and problems with methods

There are two perennial issues which confront anyone who is engaged in a theoretical enquiry into the nature of interpretation, or even in an empirical enquiry which is simply trying to determine whether or not a listener has understood an utterance. The first lies in defining what will count as adequate interpretation. The second is the problem of method. I am going to put aside the first question for the moment, simply assuming a commonsense view of adequate understanding which I shall attempt to refine in later chapters. In this chapter I shall concentrate on the second issue. How do you decide whether or not an utterance has been understood? Since we have no access to what is going on inside people's heads as they work out what an utterance means, we can only observe their behaviour after having interpreted all or some part of an utterance as they put the utterance to use. Alternatively, or additionally, we may have resort to imaginative reconstruction of what we think that the listener has understood, based on an empathetic assessment of how, if at all, we would have understood the utterance ourselves, had we been the intended listeners.

We have explored a range of basic questions in communication, examining the notions of communication, reference and context, and emphasising the role of the listener in communication rather than concentrating on that of the speaker.

We have seen, in the data presented here, examples of apparently fully correct interpretation by listeners, as well as a wide range of examples of apparently adequate and apparently inadequate interpretation, and we have speculated on the reasons for this variation. The Map task method has permitted us to address the points raised in 2.1 as possible reasons for a listener failing to respond adequately to what a speaker has said. It allows us to distinguish between occasions when a listener has apparently not heard what the speaker said, has heard but not interpreted what was said, has heard but not interpreted until later on what a speaker said, or has heard but not understood (all of) what a speaker said. We have seen many occasions when a listener selects just some part of the speaker's utterance to pay attention to, because only that part was relevant to the listener's current intentions, since for the listener, understanding what the speaker said is only a means to an end — and for the listener, it is the listener's end which is at issue.

We have repeatedly raised the question of what, in the Stolen letter task, constitutes the equivalent of the search field in the Map task. In chapter 4, we noted that deictic expressions play a potentially crucial role in situating the listener so that the listener adopts the same perspective on the search field as the speaker. We saw in chapter 3 how listeners who failed to locate themselves in the same vantage point as the speaker failed to understand adequately what the speaker said. Does deixis perform the same central role in the narrative task as it did in the spatially structured task? If so, is temporal deixis the crucial element here?

This chapter begins by laying out the broad outlines of the temporal structures which are referred to by subjects who undertake the Stolen letter task. I move on to examine how temporal deixis plays a part in restricting the contextual search field for the listener and how this interrelates with person/entity deixis. I then consider the part played by spatial deixis and person/entity deixis. Finally, I discuss how the Map task search field, viewed from a particular anchor point, can be related to the Stolen letter task search field and in this context I examine the function of what Fauconnier (1985) calls ‘space-builders’.

Before we turn to consider the use of deixis in the narrative Stolen letter task, we should briefly recall the use of deixis in the spatially structured Map task. In chapter 4, we noted the extensive use of spatial deictic expressions used to guide the listener through the landscape.

Introduction

In this chapter, I illustrate the kind of treatment I propose for switch-reference systems by presenting a formal account of switch-reference in Amele. At the end of the chapter, some suggestions are made about how such an account could be used to handle switch-reference in Eastern Porno, Lenakel and other languages.

Two factors have dictated the choice of Amele as an illustrative language. First, it is a clause-chaining language, and switch-reference languages of this kind have received little formal attention (cf. Finer 1985 a,b; Tsujimura 1987). Second, it exhibits the full range of nonreferential functional extensions of switch-reference marking described in chapter 2, including unexpected uses of SS marking with impersonal controlling clauses, and unexpected uses of DS marking triggered by changes other than in reference of the subject NPs.

Unification Categorial Grammar (UCG) will be used to formalise the account. Since UCG incorporates a semantics based on Discourse Representation Theory (DR Theory), the proposed account illustrates the way DR Theory can be used to handle switch-reference phenomena, both in Amele and more generally. It was shown in chapter 4 that the semantic representations of UCG can readily be translated into more familiar DR Theory representations.

The basis for the account is the idea developed in chapter 3 that switch-reference should be regarded as a kind of interclausal agreement. We saw in chapter 4 that in event versions of DR Theory, and in particular in UCG, it is assumed that the universe of discourse contains discourse markers or indices representing the eventualities introduced by clauses.

The subject of this book is switch-reference and closely related constructions. The aim was to see how far it was possible to give a compositional morphosyntactic/semantic account for the phenomena, but at the same time take seriously the range of data to be explained, both for the particular languages examined in detail and cross-linguistically, in a way rarely found in formal treatments. The book can thus be seen as a case study in marrying the concerns of linguistic description and formal theory.

Of particular concern both for the cross-linguistic survey and the discourse-semantic account was the functional basis for certain prototypical characteristics of switch-reference marking. Wilkins (1988: 172) says: ‘Notions such as “sameness” and “difference” through which switch-reference has been defined, have themselves been left virtually unexplored.’ In this book I begin to explore them. In so doing my main purpose is to encourage a substantially different way of thinking about switch-reference phenomena.

Along the way arguments are presented which bear on a number of other current issues, including the relation between the locus of morphologisation and its function, the nature of linguistically marked referential relations, the analysis of logophoric phenomena, and the semantic analysis of eventualities in particular with respect to their aspectual characteristics.

This book is a revised version of my 1988 doctoral dissertation (Centre for Cognitive Science and Department of Linguistics, University of Edinburgh).

Introduction

Switch-reference and logophoricity are both types of anaphoric linkage across clause boundaries which cannot be adequately accounted for by the Binding Theory (of Chomsky 1981 and later work). In central cases of switch-reference, a marker on the verb of one clause is used to indicate whether its subject has the same or different reference from the subject of an adjacent, syntactically related clause. In central cases of logophoricity, a special pronoun form is used within a reported speech context, to indicate co-reference with the source of the reported speech. This book gives a detailed examination of the two phenomena and proposes an account for them which is formalised in Discourse Representation Theory.

The major concern of the book is the functional complexity of switch-reference systems. Switch-reference markers have a much richer range of functions than just indicating obligatory co/disjoint reference. I will show that switch-reference systems are inextricably linked with the marking of temporal meaning as well as nominal meaning. I will then consider a range of apparently aberrant uses of switch-reference markers which have been reported for many languages. I will argue that unexpected uses of ‘same subject’ marking can be explained if we revise our definition of the switch-reference pivot and the switch-reference relation, in particular to take account of agentivity. The unexpected uses of ‘different subject’ marking which occur in fact represent common and systematic functional extensions of switch-reference systems which should be accounted for by any comprehensive theory.

Introduction

In chapters 5 and 6 formal accounts are proposed for switch-reference and for logophoricity. The proposals are made within the framework of a semantics based on Discourse Representation Theory, which is described in 4.1. The semantics is part of a grammar formalism called Unification Categorial Grammar, outlined in 4.2. In 4.3, the choice of this theoretical framework is justified.

Discourse Representation Theory

Discourse Representation Theory (DR Theory) is a formal semantic theory which has its origins in a desire to formulate a model-theoretic semantics for natural language which would be applicable to discourse phenomena, specifically anaphoric and tense phenomena. It was developed by Hans Kamp (1981a), though closely related ideas are presented in Karttunen (1976), Heim (1982) and Kamp (1983), and others, in particular Kempson (1984), have addressed the question of how to provide a unified account of different kinds of anaphoric dependency. DR Theory departs from Fregean semantics in taking discourse rather than the sentence to be the unit over which truth conditions are defined. Apart from this it does not represent a radical departure from a standard formal semantics, based on first-order predicate calculus and model-theoretic interpretation.

The extensions proposed by Kamp (1981a) in order to treat natural language discourse phenomena reflect his particular goals of accounting for the anaphoric behaviour of personal pronouns, and formulating a plausible account of the truth conditions of so-called ‘donkey sentences’ such as those in (la, b) (see Geach 1962, Evans 1977,1980). He took the latter task to involve giving general accounts of the conditional, and of the meaning of indefinite descriptions, as well as of pronominal anaphora.

Introduction

Logophoricity was introduced in chapter 1, where I tried to establish criteria for deciding whether or not some given set of linguistic phenomena constitutes a switch-reference system. There, I presented data from the African languages Igbo and Gokana, taken from Hyman & Comrie (1981) and Comrie (1983). Hyman & Comrie (1981) identify the Gokana data as a logophoric system, but Comrie (1983: 32, 36) calls it a ‘young switch-reference system’. He thereby draws attention to the similarities between switch-reference and logophoricity, and also raises the possibility of logophoric systems developing into switch-reference systems. In chapter 1, I argued that despite these similarities, and potential diachronic analyses of the data, we should maintain a clear distinction between switch-reference and logophoricity, and analyse the Gokana system as an instance of the latter rather than the former.

Recently, logophoricity has been the focus of renewed interest by linguists, due to analogies which have been drawn between the logophoric systems described for African languages, and ‘non-clause-bounded’, or ‘long-distance’ reflexives, which have received considerable discussion for some Scandinavian languages (Thráinsson 1976, Maling 1984, Barnes 1985, Hellan 1987), Japanese (Kuno 1972, 1988) and Italian (Giorgi 1984), and which have also been reported for a wide range of other languages, including Northern Porno (O'Connor 1986), Malayalam (Monahan 1982), and the Caucasian languages Chechen and Ingush (Nichols 1985) (see also the collection of papers in Koster & Reuland 1991). The basis for these comparisons is that the use of reflexive pronouns with clause-external antecedents in these languages seems to be restricted to just those semantically defined contexts in which logophoric pronouns may occur.

Introduction

Chapters 1 and 2 have given a comprehensive picture of switch-reference phenomena and some discussion of the relationship between switch-reference and comparable phenomena such as logophoricity. It should be clear from what has been said so far that the question ‘how can one be sure when one is dealing with a switch-reference system?’ (Munro 1980b: 2) has been an important one for workers in this area. Researchers have had a fairly concrete notion of what a ‘canonical’, ‘classical’ or ‘true’ switch-reference system looks like, and have been concerned to produce a ruling on languages which fail to conform to this profile in certain respects, such as Yup'ik Eskimo, the languages of the Northeast Caucasus, Warlpiri or Gokana. Thus, Munro (1980b: 2) notes that ‘most of the participants [at the conference, L.S.] felt that the Eskimo “fourth person” system was not a canonical switch-reference system.’ Such decisions are not unrelated to the other classificatory task which researchers have undertaken; that of trying to situate (canonical) switch-reference with respect to other linguistic phenomena, such as for example logophoricity or syntactic binding phenomena.

Why have these been such persistent preoccupations? After all, it would seem very reasonable to take the position that the way some function is realised formally in a language will be due to numerous factors which may themselves be amenable to systematic description, such as the morphosyntactic typology of the language, its historical development, and functional or semantic universals (for example, the animacy hierarchy, or links between nominal and temporal cohesive devices).

Introduction

In chapter 1, I said that the canonical conception of switch-reference assumes that it satisfies the Functional Condition, which (in the context of a further assumption made, that of the Subject Condition) stipulates that the function of switch-reference markers is to indicate obligatory co/disjoint reference between the two surface syntactic subject NPs of the related clauses. Thus canonical switch-reference assumes both a particular definition of the switch-reference pivot (as surface syntactic subject, or A + S in Dixon's 1979 terms) and a particular conception of the nature of the anaphoric link indicated by the switch-reference marker as holding between the pivot NPs (as a stipulated reference relation). However, I pointed out that in at least some languages with switch-reference systems, there is not an absolute correlation between SS marking and co-referential subjects, DS marking and subjects with disjoint reference. Although a few of these apparent anomalies of marking are the result of historical changes and other peculiarities, most can be understood as resulting from functional extensions of the system. Clearly, this presents problems for an account in terms of syntactic binding such as is proposed by Finer (1985a, b). If we take it to be a defining characteristic of a switch-reference system that it should satisfy the Functional Condition, then a number of otherwise absolutely prototypical systems fail to meet this criterion. In this chapter, we will explore switch-reference systems which manifest particular types of violations of the Functional Condition.