Journal of Pragmatics 12 (1988) 519-534 North-Holland
519
THEMATIC SUBJECTHOOD AND COGNITIVE CONSTRAINTS ON DISCOURSE STRUCTURE
Simon GARROD and Tony SANFORD*
In narratives throughout the world it is possible to distinguish between two types of character: the Thematic Subjects of the narrative and the subsidiary characters. This paper presents th[ argument that the distinction reflects a fundamental cognitive constraint on how writers and readers can represent information about different characters in a story. A number of experimental studies are cited which suggest that Thematic Subjects are represented and processed in a quite different way from other eh~r~,~,ters ir~ narrative. The difference seems to reflect the fact that such characters are held in the f ~ u s of the r~der's attention and this in turn has :onsequences for how information about them is mentally represented. To this extent the structure of narrative seems to reflect constraints on the structure of mental representations used in a variety of cognitive activities.
1. |ntroduction The cognitive psychology of discourse is concerned with the nature of the mental processes which occur in understanding spoken or written language, and the problem of how these processes might be realised in the minds of the communicator and audience given the limits on attention and immediate memory which we know to obtain. To take a cognitive approach to discourse is to take seriously the importance of such cognitive, as opposed to purely linguistic or cultural, constraints in explaining discourse structure. In this paper, we explore what we believe to be a structural consequence of such a constraint, namely the phenomenon of discourse topic or thematic subjecthood. It is widely recognised in both the psychological and the text linguistic literature that narrative texts tend to revolve around a principal character or set of characters (e.g. Rumelhart (1975), Grimes (1975)). Furthermore there is a general consensus, at least in the more recent literature, that the thematic status of a character has consequences for how that character may be introduced into the story and be referred to subsequently. Unfortunately, no
* Authors' address" S. Garrod and T. Sanford, Depart:aent of Psychology, Glasgow University, Adam Smith Building, Glasgow 8QT 9YR, Scotland. 0378-2166/88/$3.50 © 1988, Elsevier Science Publishers B.V. (North-Holland)
520
S. Garrod, T. Sanford / Thematic subjecthood and cognitive constraints
such consensus exists when it comes to the exact definition of thematic status or even the best terminology to use in discussing it (Hirst (1979)). However, for our present purposes it is sufficient to note that narratives throughout the world seem to be built around a small number of major participants with one being singled out to play a central role within any stretch of the discourse. This character we will refer to as the Thematic Subject of the discourse segment. 1 This apparently ubiquitous property of narrative could simply be treated as an historical accident, but it will be suggested here that it reflects fundamental processing constraints in readers or listeners which arise in part from their reliance on limited 'focused' memory systems for representing and interpreting text (see Sanford and Garrod (1981), Garrod and Sanford (1982)). From a reader or listener's point of view, text processing presents a number of problems in what might be called memory management. In order to impose relevent interpretations or. the words and sentences encountered, the reader requires rapid and selective access to representations of what has been previously interpreted. Without this, it is difficult to imagine how many commonplace processes in reading could come about. For instance, such problems arise whenever an anaphoric reference is encountered or when interpretation depends upon drawing a bridging inference between different sentences in the text (Clark (1975)). We will suggest that Thematic Subjects play an important role in this whole process of memory management. However, before developing the details of the proposal, we need to consider more precisely what we mean by Thematic Subject of discourse, and how the thematic status of a narrative character relates to its accessibility during reading. The notion of Thematic Subject (TS), like t~at of Sentence Topic, is a relational and pragmatic one. Whereas a Sentence Topic can be described as 'what the sentence is about' (see Reinhart (1981) for zn excellent discussion) so the T$ can be described as 'who the discourse is about'. Furthermore, as a sentence can contain only one topic, so a discourse segment can have only one TS at a time (though conjoined TSs are possible). In fact, as will become apparent in our discussion, the notion of a TS is in many ways analogous to ~hat of a Sentence Topic except that its domain is a segment of discourse r:~ther than an isolated sentence. It also shares with Sentence Topic the i, herent problems of 'objective' linguistic definition :,,h':ch arise from such a . ~agmatic description, but this does not mean that it cannot be adequately ci~aracterised. Rather, its characterisation depends as much'upon psychologic::l criteria as it does on linguistic ones. Hence we will rely mainly on various We could have chosen alternative descriptions such as discours6 topic, principal protagonist or ~ain character. However, we feel that the term Thematic Subject is less ambiguous than the first ai~ernative and more appropriate than the other two.
S. Garrod, T. Sanford / Thematic subjecthood and cognitive constraints
521
methods ibr assessing the perceived prominence of the character, its availability for reference and such like, as a means of establishing th: thematic status of the character in the experimental studies to be considered. In exploring this notion, the first question we ask is how a re~der determines who is to be the TS in the story, and whether this is associated with specific linguistic marking of such characters. Having established at least one major cue to ~hematic status, it is then possible to demonstrate a number of quite specific processing consequences of recognising the distinction between a Tg and other participants in the ~_arr~tive, all stemming from the way the TS holds the focus of the reader's attention. These will be considered under the headings of ;eYerence, mental persistence and attributive inference.
2, Establishing the thema~c status of characters in short narrative
If the manner in which narrative is processed depends upon the distinction between principal and secondary characters, then one would expect readers to be sensitive to any linguistic cues encountered early in a text which might indicate the thematic status of the characters introduced. We have carried out a number of studies to try and establish what these cues might be in short narrative texts. In these studies a story continuation procedure was used, where readers are asked to supply plausible continuation sentences for short texts that they have been given to read. The continuations are then analysed accord;rig to which of the characters mentioned in the story are included and how they are referred to. The rationale behind this procedure is simple. If the Thematic Subject of a narrative is that character at the centre of the reader's attention, then one would expect the reader to continue the text fragment with a reference to this cLaracter. Hence if two characters have been previously introdu~d into the story, the one most frequently referred to in the continuations can be taken as the Thematic Subject of the previous story fragment. Sanford, Moar an0 Garrod (in preparation) have used this procedure to explore the efficacy of a number of potential cues to thematic subjecthood including: primacy of mention in the story, °scenario-dependence' (as defined in Anderson, Garrod and Sanford (1983)) and the type of description used to introduce the character (proper name or role description). The results suggest that type of description is the predominant cue that readers used. If only one character is introduced by proper name then this character will be used, in almost every instance, as the subject of the continuation sentence. The other factors had negligible effects on the contipaations (see table I for a summary of the results).
522
S. Garrod, T. Sanford / Thematic subjecthood and cognitive constraints
Table 1 Example of materials used in part one of the Sanford; Moat and Garrod experiment_ (here the customer is judged to be less scenario-dependent than the waiter). Below are shown the results of the continuation experiment as a function of primacy, situation dependence and naming of antecedent. [Masie/the customer] entered the restaurant and sat down. [Alphonso/the waiter] we..rily limped over to take her order. versus
2nd 46%
dependent 44%
versus
independent 48%
named 61%
versus
not aamed 30% •
1st
43%
• p < 0.001 (sign test).
~n fact, there are very good reasons why proper names zhould be taken as indicators of thematic subjecthood. Kripke (1972) ct~im.~ that r~rcbnar nnrnee usually serve as rigid designators. That is, they designate the same individual in all possible worlds. From the point of view of narratives this can be readii contrasted with the case of role descriptions. If a character (John say) =, depicted as going for lunch somewhere, then going to a concert, and then going to dinner, he will be considered to be the same individual in each of these episodes. Yet the definite role description the waiter could be used in both the lunch and dinner episodes without readers supposing it to refer to tile same individual on both occasions. Thus a proper name is an ideal means of introducing a character to whom one will want to keep referring in the f u t u r e - it effectively fixes the reference. On the basis of this prelh~ainary study wc can therefore take means of introduction as at least one major cue for thematic subjecthood, early in a story. This makes i~ possible to examine the extent to which characters introduced in this fashion are treated differently in subsequent processing of the narrative. Three phenomena w~ll be considered- (1) referential consequences of thematic subjecthood, (2) the 'mental persistence' of Thematic Subjects, and (3) the way in which they attract attributional inferences. These will be considered in turn. .
.
.
.
.
.
l.-~
vl.,~.~
s,Ll.~,a~
hJ
3. Referr":ng to Thematic Subjects: Their accessibility and mental persistence There are now a number of studies reported in the literature which suggest that readers find it easier to resolve an anaphoric reference to a Thematic Subject than to other types of discourse antecedent. Differences in the ease of resolution have typically been measured by the relative time taken to read a
S. Garrod, T. Sanford / Thematic subjecthood and cognitive constraints
523
seqtence containing anaphofic references (Garrod and Sanford (1982)). However, it seems that this phenomenon is particularly associated with pronominal reference. ~f a proper name is the most effective way of introducing a TS into a story, then a sentence initial pronoun seems to be the most effective way of maintaining reference to this character. Hence, when one controls for the textual distance ~tween antecedent mention and anaphor, the reading time advantage is only associated with sentences containing pronouns; no such distinction emerges with anaphoric proper names or other definite descriptions. An example of the materials and reading time measures from such an experiment are illustrated in table 2. Table 2 (a)
Example of the materials used in the Sanford, Moar and Garrod reading time experiment
Order of exposure (1) (2) (3) (4) (5) (6)
Mr. Bloggsfrhe manager was dictating a letter. Claire/The se,zret~ry was taking shorthand. It was getting to be late in the afternoon. He/She/Mr. Bloggs/Claire was feeling hungry, i QUESTION Was Claire/The secretary doing the filing?
(b) Reading time in msecs, for the sentence containing the anaphor (4 above) as a function of anaphor and antecedent Anaphor Pronoun Antecedent I st 2"a
named 2170 2174
Full description not named 2456 b 2466 b
named ?631 2633
not named 2626 2626
a = target sentence. b Min F'¢x.ot~ = 9.47 (p < 0.01).
Convergent evidence for the special status of Thematic Subjects as preferred antecedents for referential pronouns also comes from a recent study using spelling error detection latency as an alternative and more immediate mea,;ure of processing difficulty (Garrod and Sanford (1985)). The experiment can best be illustrated in relation to a sample set of" materials used in a previous study on the interpretation of anaphoric proper names and definite descriptions. An example of the materials is shown in table 3. In these materials two characters are introduced into the story, 'Elizabeth' and 'the lifeguard'. A critical target sentence follows which can occur in one of four conditions. It either contains a reference to 'Elizabeth' or 'the lifeguard' which may be followed by a verb which is either consistent or inconsistent with the full contextual interpretation of the previous anaphoric noun-phrase. So for instance (1) and (2) below are consistem, while (3) and (4) are contextually anomalous:
S. Garrod, T. Sanford / Thematic subjecthood and cognitive constraints
524
(1) (2) (3) (4)
Within Within Within Within
seconds, seconds, seconds, seconds,
Elizabeth sank beneath the surface. the lifeguard jumped into the pool. Elizabeth jumped into the pool. the lifeguard sank beneath the surface.
Furthermore, the contextual anomaly, where it occurs, depends upon establishing the contextual significance of the prior anaphor. For h-~stance, in (3) the fact that it is anomalous for Elizabeth to jump at this point in the story comes from our knowledge of her physical state: of being suspended in water, out of her depth. This knowledge can only become available after we have identified the discourse referent for Elizabeth and then recovered the relevant contextual knowledge about the referent in question. Table 3 Materials used in spelling error detection study (Garrod and Sanford (1985)). Below are shown the magnitude of the consistency effects as a function of antecedent and anaphor.
Title:
A dangerous incident at the pool
Context: Elizabeth was a very inexperienced swimmer and wouldn't have gone into the pool if the lifeguard hadn't been standing nearby. But as soon as she was out of her depth she started to panic and wave her hands about in a frenzy. Target sentences: (1) Within seconds (2) Within seconds (3) Within seconds (4) Within seconds
Elizabeth jumped ~ into the pool. the lifeguard jumped into the pool. Elizabeth sank" beneath the surface. the lifeguard sank beneath the surface.
Consistency effects (msecs. ) : . Proper paine Definite description Mah, character pronoun Secondary character pronoun
813 500 551 - 159
Misspellings: jumped - jimped, sank - senk.
In the original experiment the time to detect misspellings on the critical verbs was recorded on the assumption that such judgements would be affected by the consistency of the verb in its context (see Cole and Jakimik (1980), for evidence of such effects in detection of mispronunciations in speech). In fact, using materials of the sort illustrated, just such an effect was observed; spelling errors on contextually inconsistent verbs took about 550msecs. longer to detect than those on consistent verbs. This result indicates that the readers have been able to establish the contextual significance of the anaphoric noun-phrase just as soon as they have encountered it. Given this initial result it was possible to use the same procedure to investigate what would happen with pronominal anaphors in this situation, simply by replacing the fuller descriptions with pronouns. Of course the materials were also amended to ensure that the pronoun would be quite
S. Garrod, T. Sanford / Thematic subjecthood and cognitive constraints
J25
unambiguous at that point in the story on the basis of number or gender agreement with its intended antecedent. Th~ ~e~,ond experiment produced a very striking result since the consistency effect was only observed when the subject of the verb was a pronoun referring to the character introduced with the proper name, the TS of the story. The results are shown in table 3 where the magnitude of the consistency effect is illustrated against the nature of the immediately precedii~g anaphor. In effect, the experiment demonstrates that only pronouns that refer back to the TS are resolved immediately during the course of reading these materials. Presumably, interpretation of the other pronouns is suspended until the reader has processed subsequent information in the sentence, and probably depends upon interpretation of the main verb (see Marslen-Wilson, Levy and Tyler (i982) for a relevant discussion). Further evidence for the special referential status of Thematic Subjects can also be found in certain studies of narrative production, which suggest that pronominal reference, at least in sentence subject position, is reserved for the current thematic subject of the story being told (Karmi!off-gmith (1980), Marslen-Wilson et al. (1982)). All of this evidence is consistent with the view that the character who fills the role of Thematic Subject is most prominent in the mind of tbe reader, and hence a preferred candidate for any textual device which signals reference maintenance, such as a pronoun or null anaphor. Another consequence of the prominence of thematic subjects is in what we will caU their 'mental persistence'. Anderson, Garrod and Sanford (1983) demonstrated that as stories pass from one episode to the next, certain characters seem to ieave the focus of the reader's attention, while others persist. They employed materials like those shown in table 4 where two characters are introduced in a stereotypie setting (one named and the other described by their role in the setting) and then a sentence is given which may or may not Table 4 Sample of materials and continuation results from Anderson, Garrod and Sanford (1983). At the cinema Jenny found the film rather boring. The projectionist had to keep changing the reels. It was supposed to be a silent classic. Ten minutes later the film was forgotten. Seven hours ) Frequency of continuations mentioning either the Thematic Subject (Jenny) or the Subsidiary Character (the projectionist) as a function of episode shift.
Same episode Different episode
Thematic Subject
~ubsidiary Character
17.5%
8%
2,~.5%
4%
526
S. Garrod, T. Sanford / Thematic subjecthood and cognitive constraints
cue a shift in scene. These sentences were constructed to indicate shifts in discourse time which were either within the normal time span associated with the main activity in the scene (e.g. the time taken to eat a meal) or well beyond it. In this way the sentence could cue a shift in episode and be used to set up conditions for testing the relative prominence and accessibility of the two characters. A humbler of measures were used for character accessibility following shifts in text ep,;,oode • story continuation, time to answer questions about the two characters, and reading time for a subsequent sentence containing a pronoun which referred back to either of t~e characters. In all these cases there was evidence that the named character, the TS of the story fragment, remained in the reader's focus of attention following shifts in scene where the subsidiary character disappeared. For inz*,ance in table 4 the frequencies of mention of the two characters in continuations is shown both in the case where there is an episode shift and where there is none. The same pattern emerges for the reading time and question answering time data shown in table 5. Table 5 Sample of the materials and the question answering time and reading time results (in msecs.) from Anderson, Garrod and Sanford (1983). In th,, restaurant The Browns were eating a meal in the restaurant. The waiter was hovering around the table. This restaurant was well kr own for its food.
Five hours } later the restaurant was empty. Forty minutes _ He
[ serving
all the good food.
Question answering times TS a SC t' 1725 2075 1723 2190
Target sentence reading times TS SC 1760 1850 1730 1900
within episode between episode
a TS = Thematic Subject reference,
i' SC = Subsidiary Character reference.
All of the experimental evideace seems to confirm one's intuition that Thematic Subjects have a special attentional status for the reader as ,:ompared to other, subsidiary, characters in a narrative. First, they seem to influence control over pronominal assignment, possibly acting as the default assignment for any sentence initial pronoun that is encountered. Secondly, they seem to retain this special status over transitions from one episode to the next in the narrative. Of course, if our observations about the thematising role of proper names are correct one might expect that any current Thematic Subject could be displaced or possibly held in abeyance if some new character is introduced using a proper name, but this would not occur otherwise.
S. Garrod, T. Sanford / Thematic subjecthood and cognitive constraints
527
The final processing phenomenon which we have observed in relation to Thematic Subjects is that of attributional inl~rence. Thematic Subjects seem to attract certain types of inference in a way that subsidiary characters do not, and we now turn to this phenomenon in the next section.
4. Thematic Subjects and at'tTlhutional inference It is now clearly established that readers tend to draw a wide variety of inferences about the various characters introduced in a text, which go well beyond the information given (see for instance Thorndyke (1976)). Many of these inferences are motivated by attempts to find reasons why some character did what he or she did or they may be more mundane inferences which simply fill in the details which the writer has left unstated. Recently, we have carried out some experiments on a special type of ~nference which is commonly associated with narratives. The inference concerns the interpretation of what we will call psychological atmosphere statements, an example of which is given below: Mary hated going to the dentist and felt anxious when the receptionist called out her name. The atmosphere was hot and sticky ...
The final sentence in this example is an apparently neutral remark about the atmosphere in the dentist's office. However, it has the interesting property that it can only be verified by some character in that situation who finds 'the atmosphere hot and sticky' since this is ultimately a subjective judgement, hence the description psychological atmosphere statement. From the present point of view such statements are interesting because if you give readers such a text and subsequently ask them if either 'Mary' or 'the receptionist' found the atmosphere hot and sticky they will characteristically only attribute this inference to Mary (the T$), even though the statement is in some sense a~tributionally neutral. This observation arose from a recent study (Sanford and AI-Ahmar (in preparation)) where subjec::" were presented with texts like the following: At the restaurant
Juliet entered the restaurant.
There was a table in the comer The waiter took the order. Things seemed to go well ~hat night. In one condition, subjects read passages of this type under the self-paced reading-time conditions described earlier. The sentence of interest in the present example is the final one. 'Things' cannot 'seem to go well' without
528
S. Garrod, T. Sanfgrd / Thematic subjecthood and cognitive constraints
assuming that there is some person for whom this is true. Our guess was that this person would be the TS, and that the subjects would be inclined to answer 'Yes' to question (1) and 'No' or 'Don't know' to question (2): (1) Did things go well for Juliet that night? (2) Did things go well for the waiter that night? Subjects were presented with a mixture of questions after reading each passage which included those like (1) and (2) above, and they had to respond by pressing the appropriate one of three buttons labelled 'Ye_~', 'No' or 'Don't know'. Nearly 90% of responses to these questions were affirmative when the question referred to the TS while only 50% were affirmative when the question referre~i to the other character in the story. Thes,~ results suggest that psychological atmosphere statements ('things seemed to go well that night', 'the atmosphere was hot and sticky', etc.) are . . . . . . as. though . . . .r "a,~,,;,.t -'"r''" ~1,,. judgements . . . . . . ~n , "~ . , feelings of the ~'~ l o and interpreted th,.~, ,h,~ not of the other characters in the story. However, it might be argued that the results from this experiment do not necessarily imply that the reader is drawing this inference during initial interpretation of the text, but is doing so only when required to come up with an answer to the question. This potential objection to the study motivated a further experiment which was designed to explore the more irp ~ediate consequences of drawing an attributional inference. The rationale behind this additional study was straightforward and depended upon previous observations (see Haviland and Clark (1975), Keenan (1978)) that readers take additional time to understand sentences that do not: directly follow from the previous text, presumably because it takes time to establish a bridge between the sentence under interpretation and any representation of the prior text. Consider for instance a situation where someone has read the text about Mary going to the dentist given at the beginning of this section and then encounters the following sentence: Mary/The receptionist mopped her brow. which refers to an action which would be a natural consequence of finding the atmosphere hot and sticky. Presumably such a sentence would cause no prc,blem to readers so long as they had drawn the original inference that Mary or the receptionist had found the atmosphere hot and sticky. On the other hand if no such inference had been drawn then one would expect this to lead to an increase in comprehension time for the critical sentence. This prediction was tested using a self-paced reading time study of the sort described earlier but with materials like those shown in table 6. The materials co ~tain a brief initial section where two characters are introduced: the
S. Garrod, T. Sanford / Thematic subjecthood and cognitive constraints
529
Thematic Subject and a subsidiary character. Following this a psychological atmosphere statement occurs and then a filler sentence which contains a reference to either the TS or a subsidiary character. This sets the seer: for the critical inference sentence which always refers to the character mentioned in the previous filler sentence. The key manipulation with these materials concerned the presence or absence of the atmosphere statement and its effect on the reading time for the critical inference (target) sentence. Table 6 Example of the materials and reading time results for the "atmosphere statement" inference experiment.
Lunch at the cafeteria Alistair hung up his coat and picked a tray. The waitress smiled as she poured the coffee. The atmosphere was hot and sticky: tHet°°k lthecup. t She offered j He/She mopped his/her brow. b Reading time results (in msecs~) for the inference target sentence as a function of the antecedent referent and presence or absence of the atmosphere statement. With atmosphere statement Without atmosphere statement
TS 1379 1650~
SC 1430 1463
• = atmosphere statement, b = inference target sentence. Min F' = 5.44, p < 0.01 for the interaction. TS - Thematic Subject reference. SC = Subsidiary Character reference.
The results of this manipulation are also shown in table 6. It can be seen that the time taken to read the target sentences about the principal character depended upon whether or not they were preceded by the appropriate 'atmosphere' statements, while no such effect was observed when the sentences contained references to a secondaD: character. The result therefore confirms our earlier finding and suggests that readers draw the attfibutional inference only about the TS of the story and do so at the time they read the atmosph:-:. statement. The inferehce is indeed made 'on-line'.
5. Discussion: Thematic Subjects as controlle~ of focus
Taken together, these various experimental findings on the referential consequences, mental persistence and attributional inferences associated wi*.h Thematic Subjects in n~.rrative all sugge~,t that they are assi~',~ed ~, :~ i-~i status during text processing. Thematic Subjects seem to hold the foc~ ~::~.~":,ttention during reading, they are privileged candidates for anap:¢~eric re:i~'~.:nce, they
530
S. Garrod, T. Sanford / Thematic subjecthood and cognitive constraints
are immune to psychological shifts in scenario and act as natural vehicles through which to interpret apparently neutral psychological atmosphere statements. At the same time, the apparent universality of the distinction would suggest that having to pick out one particular character in narrative as the TS of the story fragment is not just a processing option but is forced upon the reader through some fundamental cognitive constraint and it is to this question which we now turn. One way of treating these experimental phenomena is within the framework developed by Sanford and Garrod (1981) for describing the attentional state of readers in terms of a focus theory of memory access. This framework was put for,yard as an attempt to fo,,"malise intuitive notions about how attention affects a reader's ability to interpret discourse. Sanford and Garrod argued that text interpretation occurs against a constantly changing background model of what the text is about, which is held in what we called 'focused' partitions of working memory. Two important featurcs were ascribed to, this dynamic model of the text. First that it consists not only of the er~tities mentioned in the prior text and the relations between them, but also of relevant background knowledge in the form of a referent situation or scenario elicited by the text. This second type of knowledge is important in explaining why readers seem to have no problems in interpreting references to entities or characters which one would expect to find in those settings. Hence a reference to 'the waiter' in a restaurant setting or 'the judge' in court presents readers with no apparent problem in immediate interpretation even in the case where no such character has been previously introduced (see Garrod and Sanford (1981, 1983) for details). The second feature ascribed to the dynamic model is that the entities represented in the first, extdieit foe~s, partition are mapped into the scenario representation in the second, implicit focus, representation by means of role pointers° This means that on encountering an initial sentence such as 'John was on his way to school' most readers will assume that John is being cast in the role of school-boy (even though this is not explicitly stated). In fact, Sanford and Garrod (1985) have demonstrated that readers experience processing problems if they encounter subsequent sentences in a text which violate previously ascribed role casting, as for instance'in the following example: o
John was on his way to school. He was worried about the moths lesson. He had trouble
controlling the class.
Here the final sentence seems to suggest that John is not a school-boy but a master at the school. The concept of a current model is therefore one of a ~epresentation which consists of two partitions: an explicit focus partition containing a record of the explicitly mentioned characters and an implicit focus partition containiag
S. Garrod, T. ,Ymford / Thematic subjecthood and cognitive constraints
531
background information or scenarios. There is also a mapping between the two in terms of the roles which characters play in the current scenario. So how can this model support direct interpretation of referential expressions? Sanford and Garrod proposed a scheme for characterising referential interpretation for various different types of expression in terms of the search formulation associated with each type. According to this account pronouns differ from other forms of reference in that they can only instigate searches through the explicit focus partition and hence only recover discourse antecedents currently represented in this partition. Full definite noun-phrases on the other hand can instigate searches through either partition and so enable an antecedent to be recovered either indirectly through the role pointer or directly through the specification in explicit focus. This proposed distinction between the two types of access accounts for the difl}rence in behaviour between the two types of expression and explains why pronouns and-other definite noun-phrases seem to afford differential access to t h a .~arna antar'~.dr.nt u n d e r r . ~ r t n l n e n r ! d l t l n n % something w h i r h h n ~ already been mentioned in the context of anaphoric reference to antecedents which are Thematic Subjects. This leads us to the crucial issue of how the thematic status of a discourse referent affects its representation within focus. Clearly from the evidence cited earlier Thematic Subjects seem to have a privileged place in the reader's focus of attention and this should be reflected within the theory. Furthermore, we have seen specific evidence that Thematic Subjects seem to be preferred antecedents for pronouns and in fact the only ones which enable immediate interpretation. For these reasons it is clear that Thematic Subjects must be seen as the key entities represented within the explicit focus partition, but what does this mean in terms of the theory? Let us first propose a distinction between two types of entity representation which may be found in what we have called explicit focus, key entity representations and dependent entity representations. The difference between these two will have consequences both in terms of the richness of any individuating information which may be associated with that representation, and, more importantly, in terms of the way in which what is represented in explicit focus controls the form of the scenario represented in implicit focus. First, key entities may have associated with them all sorts of individuating information, a name, known personal characteristics and so on, deriving both from what is explicitly stated in the text and what may reasonably be inferred, as in the case of the attributional inferences described earlier. In fact, we may suppose that it is one of the reader's main goals to discover and record such information for these key entities. For dependent entities, on the other hand, indiv~dua',ion only occurs as a result of the role mappings which bind the explicit focus representation to the current ~cenario. Hence, to give a slight oversimplification, if the reader has encountered a sentence such as 'Mary
532
S. Garrod, T. Sanford / Thematic subjecth6od and cognitive constraints
answered the lawyer' in a text about a court ~:ase, then explicit focus would contain a key entity representation for 'Mary' with associated individuating information about her (e.g. her name), but only a dependent entity representation for the lawyer, individuated solely in terms of his or her role in that scenario (e.g. as the lawyer who questioned Mary). The second major difference between the two types of representation relates to the effect they may have on the form of knowledge brought into implicit focus. We suggest that key entities influence the scenario in terms of the perspective which is taken in the representation. T~ understand what we mean by this, let us take a simple example of how knowledge of a situation or scenario might be represented with respect to the various entities. Consider our knowledge of buying or selling goods. This knowledge can be represented from two very different points of view, depending on the perspective we bappen to take. If we look at it from the buyer's perspective, we may note that the buyer receives the goods from the seller and hence obtains ownership u^" | them, while it is incumbent on the buyer to give cash of an equivalent value ~,,,,, m to the ~'~"~" " ,~^,~,,~u~,~ ~.,,.k. . . . for the goods. Furthermore, as a consequence of the interaction we know that the buyer is that much poorer in cash but that much better off in terms of the goods, and so on. We may also note that there are certain conditions when it is clearly advantageous for the buyer to purchase the goods (e.g. during a sale, or when they are discovered to be cheaper at one store than another or when they happen to be in glut for some reason or another, and so on). Now this same situation can also be seen somewhat differently from the seller's point of view. In this case, we way note that the seller gives certain goods to the buyer and in exchange receives cash of an equivalent amount, whereby the seller is better off. Fm'thermore, we may note that there are certain d i f f e r e n t conditions when it is advantageous to sell, as when the goods are scarce and so have greater cash value or when the seller can no longer store them cheaply and so on. Even though this knowledge is in some sense about the same situation it is clearly not the same knowledge. In particular, the significance of the various events in the situation may be quite different from two perspectives. What is 'good' in bu:dr~ may ,,,,""~,be ..t..~, ~,.~ is 'good' in selling. In fact, in many respects the two points of view conflict. Although it may be tempting to think of the two forms of knowledge as equivalent since the various roles played by the characters are given the same name in both cases - the seller is the seller and the buyer the buyer whichever way you look at i t - this does not mean that the way they map into the representation is necessarily equivalent. In the representation which reflects the buyer's perspective the buyer is the subject or topic of that representation, in the sense of being the character which the representation is about. In the seller's perspective representation it is now the seller who is the subject. Of course, the buying/selling scenario is no special case. When one considers the wide range of activities involving a number of participants such as travelling,
S. Garrod, T. Sanford / Thematic subjecthood and cognitive constraints
533
eath~g ouL g6~ng to the cinema or theatre, conversing and so on, each of these activities can be seen as involving both co-ordination between the participants and a certain degree of conflict, and to this extent underlying our knowledge of each will be many different representations reflecting different perspectives on the same basic situation. What we are therefore proposing in making the distinction between key entity representations versus dependent ones is that only key entitie~ control the exact form of background knowledge which is brought into focus, and it is in terms of this implicit knowledge that the dependent entities are individuated. Hence, given an implicit representation of buying/selling from the buyer's point of view, the seller is simply represented as playing a role which reflects the buyer's point of view. Furthermore, the individuation of the dependent entity should only persist as long as that particular scenario is relevant (see Anderson, Garrod and Sanford (1983)). It should now be clear that this distinction is being set up as a kind of cognitive counterpart to the textual distinction between Thematic Subjects and ~""1" : e r character3 in a narrative. However, we would argue that the distinction reflects a basic cognitive constraint, whereby it is only feasible within a system of limited capacity to hold a single model in mind, whether we are engaged in problem solving, understanding text, or whatever, and that it is a feature of this singularity that the model can only reflect one point of view at any time. Establishing the Thematic Subject of a narrative fragment is therefore necessary in order to construct an effective mental representation of the text world, which will in turn be instrumental in understanding the subsequent sentences in the text. The TS, once identified, will be treated as a key entity in the explicit focus representation, and so be readily available as a default reference for any referential pronoun. It will also be retained as a key entity irrespective of shifts in the discourse setting. Finally, it will be the individual with respect to which the text is interpreted, the person the text is judged to be about. Hence, it is re.asonable that the TS should attract the kind of attfibutional inferences discussed in relation to the experiment reported above. Perceived states of the world of the kind described by 'psychological atmosphere statements' will be constnJed by th, reader as relevant to the goals and problems facing the TS, and so are interpreted through the 'eyes' of the TS rather than through those of any other character. Of course, much remains to be established about aspects of information processing associated with thematic subjecthood in a broader context. For e×ample, what happens in the case where there are two 'important' characters, perhaps both named, interacting for a while in a longer narrative? How do dynamic cues influence the kind of processes outlined above? Although these and other questions still remain unanswered, we believe that the experimental work cited above demonstrates an interesting correspondence between the
534
S. Garrod, T. Sanford / Thematic subjecthood and cognitive constraints
linguistic concept of thematic subject and the information processing concept of a key entity in a cognitive representation. References Anderson, A., S. Garrod, and A.J. Sanford, 1983. The accessibility of pronominal antecedents as a function of episode shifts in narrative text. Quarterly Journal of Experimental Psychology, 35: 427--440. Clark, H.H., 1975. 'Bridging'. In R. Schank and B. Nash-Webber, eds., Theoretical issues in natural language understanding. Cambridge, MA: Bolt, Beranek and Newman. Cole, R.A. and J. Jakimik, 1980. 'A model of speech perception'. In: R.A. Cole, ed., Perception and production of fluent speech. Hilisdale, N J: Erlbaum. Haviland, S. and H.H. Clark, 1974. What's new? Acquiring new information as a process in comprehension. Journal of Verbal Learning and Verbal Behaviour 13: 512-521. Hirst, G., 1979. Anaphora in natural language understanding: A survey. Technical Report 79-2, Department of Computer Science, The University of British Columbia. Garrod, $. and A.J Sanford, 1982. The mental representation of discourse in a f~,~,ussedmemory system: !mplicatiens for the interpretation of anaphoric noun-phrases. Journal of Semantics 1: 21-41. Garrod, S. and A.J. Sanford, 1983. 'Topic dependent effects in language processing'. In: G.B. Flores d'Arcais and R.J. Jarvella eds., The process of language understanding. Chichester: Wiley. Garrod, S. and A.J. Sanford, 1985. On the real-time character of interpretation during reading. Language and Cognitive Processes 1: 43-59. Grimes, J.E., 1975. The thread of discourse. The Hague: Mouton. Karmiloff-Smith, A., 1980. 'Psychological processes underlying pronominalisat~on and non-pronominalisation in children's connected discourse'. In: J. Kreiman and A.E.Ojeda, eds., Papers from the parasession on pronouns and anaphora. Chicago: Chicago Linguistic Society. Keenan~ 3.9 1978. Infer~'ing causal connections ill prose comprehension. Paper presented at the American Psychological Convention. Kripke, S. A., 1972. 'Nam;a~ and necessity'. In: D. Davidson and G. Harman, eds., Semantics of natural language. Dordrecht: Reidel. Marslen-Wilson, W.D., E. Levy, and L.K. Tyler, 1982. 'Producing interpretable discourse: The establishment and maintenance of reference'. In: R, J. Jarvella and W. Klein, eds., Speech. ~,lace and action. Chichester: Wiley. Reii~hart, T., 1981. Pragmatics and linguistics: An analysis of sentence topics. Philosophica 27: 53--94. Rumelhart, D.E., 1975. 'Notes on a schema for stories'. In: D.G. Bobrow and A. Collins, eds., Representing and understanding: Studies in cognitive science. New York: Academic Press. Sanford, A.J. and S. Garrod, 1981. Understanding written language. Chichester: Wiley. Sanford, A.J. and S. Garrod, 1985. 'The role of background knowledge in Psychological accounts of text comprehension'. In: J. Allwood and E. Hjelmquist, eds., Foregrounding background. Lund: Doxa AB. Thorndyke, P.W.. 1976. The role of inferences in discourse. Journal of Verbal Learning and Verbal Behaviour 15: 437--446.