Episode models Referential Choice and the Environment

some measure like referential distance for the different referential forms. According to the recency model topic continuity measures should not show any discernible difference for different referential forms. The third prediction is that discourse boundary phenomena would not correlate with different referential forms. The fourth prediction is that all new referents should be introduced by the same de- vices that are used for distant referents. The final prediction is that even if there is any ranking of refer- ents it should not have any influence on the choice of referential form. In the Olo data only the prediction regarding referential distance is sustained, but even there referential distance alone does not distinguish all the referential forms.

3.3.2 Episode models

In episode models the choice of referents is based on the structure of the text andor the structures in the language producer’s mind. Early references to episodes are found in the work of Grimes 1975, 1978 and Longacre 1968, 1983. Longacre 1983 uses the term episode as something that refers to plot structure, rather than to the structure of the discourse. What others might call episodes, he consid- ers embedded discourses. In more recent times a number of researchers have wanted to make use of the ideas embodied in the term episode. They have looked at referential form based on either discourse structure or story struc- ture. In doing so they appeal to a more complicated view of memory organization than can be provided by the recency approach. All the implementations revolve around having substructures in a discourse, and that moving from one structure to another forces a use of a nominal. Some of the structures are simpler than others. In the simplest form, a discourse has episode units, and within an episode unit a recency model is used. This is characteristic of the work of Tomlin 1987 and Pu 1991. A more complicated account is that of Kintsch and van Dijk 1978, van Dijk and Kintsch 1983, and van Dijk 1982. In their work they relate episodes to macropropositions, and any referent who has been introduced in an episode, as an argument in a macroproposition that is currently in active memory, can be accessed by a pronoun. Another variant of the episode model is that of Marlsen-Wilson, Levy, and Tyler 1982. In their ac- count they find three levels: the story, the episode, and event. They used a comic book as a device to generate narrative production and claim the three levels correspond to the levels in the comic book footnote 4. Anderson, Garrod, and Sanford 1983 have shown effects of episode boundaries on subsidiary char- acters. In their work, a highly topical character could be sustained across an episode boundary with a pronoun, but a “scenario dependent” character could not be. According to their account an episode boundary does not entail the use of a noun for all participants. From their data we can conclude that a pure episode system will not account for all the choices of referential form. Since some characters, those not dependent on an individual scenario, can cross episode boundaries and still be referred to with a pronoun, we must conclude that a strict episode account is not viable. This is not to say that epi- sode boundaries will not display some effect on referential management, but that the effects are more complex than the simple “episode boundary invokes a nominal” equation suggests. The data in the Olo texts supports the idea that episodes are important to the management of reference, but that they are only one feature in the complicated task. Anyone who wants to use an episode account is faced with the problem of determining episode boundaries. When performing a production oriented experiment, 5 episode boundaries can be drawn based on outside cues such as breaks in the presentation of still pictures Tomlin 1987. This approach is not feasible for naturally occurring narrative, because it is the speaker who is manipulating the world of discourse, not the experimenter. We therefore have to rely on less discrete indicators of epi- sode boundaries: changes in time, space, and event sequences. While it is difficult to objectively determine episode boundaries when looking at any given text, it is not impossible to draw some general conclusions. Grimes 1975 proposed shifts in time or location as things that precipitate episode boundaries. Anderson, Garrod, and Sanford 1983 used these notions to conduct experiments involving distinctions between participants that are dependent on a given time or place and participants that are independent of a particular time or place. They found that these 3.3 Referential Choice and the Environment 53 5 Examples of this are work by Chafe and his colleagues 1980b, Tomlin 1987 and Pu 1991. factors do affect the accessibility of dependent referents. If a scene involved a given activity that nor- mally takes a maximum of two hours, a shift of time longer than two hours would make it difficult to retrieve a scenario dependent participant. For example, in a scene in a movie house the watching of a film will take a couple of hours. A reference to the projectionist is difficult to retrieve if a time phrase like “seven hours later” is used in the next sentence. Anderson, Sanford, and Garrod also normalized spatial distance constraints for different activities and found they worked the same way. Givón has proposed that there are action or thematic continuity breaks that are often associated with the same referential devices that are used for long distance anaphora 1983a, 1983c. In particular he identi- fies three changes besides the change of major participant that may be important: location, time, and se- quential action. Givón distinguishes sequential action that breaks off from a given action sequence and starts a new sequence, from sequential action that is temporarily broken off for some type of interjection and then resumes the same sequence Givón 1983a:158–159. Givón found a correlation between long dis- tance markers and thematic disjunction. In his study he made impressionistic judgments of where the disjunctions were and at what level, based on changes in time space and actions. While impressionistic judgments can be useful, it would be better to code the amounts of disjunction if possible, or the size of spatial or temporal shift. This could be done on an absolute scale, of some- thing like hours and minutes for time and some linear measure for space. However, this is neither prac- tical nor theoretically desirable. It is not practical because stories rarely contain closely quantified temporal or spatial measures. Rather, they will contain things like that night, tomorrow, later, now, or at the store, on the hill, in his garden. There is no practical way to turn these into discrete time or distance measures. The reason that the strict scalar measures are not theoretically desirable is that using them would assume that all episodes have the same size boundaries. Obviously spatial boundaries measured in a linear measure for in the park are much larger than in his room. The point of Anderson, Garrod, and Sanford’s 1983 work is that different scenarios have different normal space and time boundaries as- sociated with them. What they found is that going well beyond the normal boundaries invoked the change. This means that the changes must be relative to what is expected. When working on texts that have independently established scenarios with normalized temporal and spatial boundaries, we can use those boundaries to determine when a character has moved out of a scenario. But it is difficult to apply this approach to narratives that are not dependent on clear scenar- ios, either because the normalized temporal and spatial boundaries have not been determined, or the story is just not organized into discrete scenarios. When we are unable to establish the boundaries in- dependently, we either rely on impressionistic judgment or group all the boundary types together. This second choice is what I do in this work. The exact mechanisms will be discussed in chapter 4.

3.3.3 Memorial activation models