The Textual Data Referential Forms

Because this study is text based, the normally used measurements of activation 1 are not possible. From psycholinguistic studies we can estimate the impact on activation of the different referential de- vices and we can quantify other parameters of the text. Measurements are made of referential distance RD and topic persistence TP. The values for RD and TP can be compared to a variety of previous works 2 as part of the crosslinguistic typology, although there have been changes in the exact imple- mentation of the referential distance methodology. In addition, the referential forms are examined in the light of changes in time, space, and event consistency.

4.2 The Textual Data

The texts used in this work are a collection of narratives gathered between 1981 and 1991 from na- tive Olo speakers. All but one of the texts are written narratives. Most are first-person narratives, with a varying level of speaker prominence in the discourse. Some analysts e.g., Payne 1993 eschew first-person narratives and instead examine traditional stories. Traditional stories are problematic in Olo as they are normally “known” narratives and as such there is no guarantee that the speaker does not adjust his style because the listener is expected to know the story. Stories told in the first-person have often been excluded on the assumption that the speaker is involved in the speech act and is al- ways mentally available. This assumption, while not false in itself, implies that the speaker would only ever use a single device in referring to himself, or to the group he is a part of. It is possible to examine only the third-person referents in a text that is a first-person narrative, which is what is done in this study.

4.3 Dividing the Text

The texts are interlinearized and supplied in appendix A. The texts are divided into their clauses. The clauses center around the verb, its core arguments, and any marginal elements like time or place. The serial clause construction is divided into its separate clauses. Clearly subordinate clauses, those with morphosyntactic marking and quotations, are included with the clauses in which they occur. The texts are also divided according to temporal criteria, location, and punctuation. Only two punctuation marks were used in the text: the comma and period. A semantic cohesion index is also calculated for each clause. A short description of the workings of the database is provided in appendix B.

4.3.1 Temporal locations

One of the consistently cited correlates with episode boundaries is a shift in temporal location. When a shift in time occurs, a comprehender has no way of knowing who has been brought onto the scene or who has left. This means that for the comprehender to know who the participants are, they need to be identifiable. This makes a shift in temporal location a logical place to reidentify partici- pants. By examining the changes in temporal location, we can mark off the different places in the text an episode shift is likely to occur. The temporal criterion used in this work is operationally explicit. Anywhere one of the temporal words occurs in the initial position, the boundary will be marked on the clause in which it occurs. When a temporal change occurs in clause final position marking a change in time, this will be marked in the following clause. This division is done so that the boundaries are con- sistently calculated as occurring between two clauses. In Olo, clause final temporal expressions give the time at the end of the event, not the beginning of the event. So the temporal boundary is between the clause with the clause final temporal expression and the next clause. Any other temporal expres- sion is marked in the clause in which it occurs. Nonexplicit time boundaries, like sleeping at night and then getting up in the morning, are marked as temporal changes. The time markers for the Somoro 64 Methods 1 The normal measurements are based on reaction times to word probes and reading times. 2 Examples of this are to be found in the topic continuity volume edited by Givón 1983b, and his more recent volume Voice and Inversion 1994b as well as Payne 1993. dialect of Olo are given in table 4.1. The first six, sungoi, sungoi sungoi liye, nempis, nempis liye, and nomul are not normally used in a narrative story after the first setting of the time for the story. The word fei ‘todaynow’ is used frequently, as well as nolowi ‘tomorrow’. The day and night are divided up by some inexact temporal divisions. These are given following fei in table 4.1. Table 4.1. Olo temporal divisions sungoi ‘a long time ago’ The time before the speaker was born is automatically in this time. sungoi sungoi liye ‘a time before sungoi’ roughly equates with an ancestral time. nempis ‘the day before yesterday and before’ nempis liye ‘a time before nempis’ has a sense of psychological distance nomul ‘yesterday’ fei ‘today’ if ili ‘dawn’ mul ‘morning’ epli nimin ‘noon’ ningli ‘afternoon 4–6 PM’ mulpou ‘night’ wem tangu ‘middle of the night’ nolowi ‘tomorrow, the next day’ nolowi nenpe ‘the day after tomorrow’ Because of the problems, both theoretical and practical, in dealing with temporal boundaries, I have adopted a simple approach to determining if a boundary occurs. If one of the time words that are given in table 4.1 occurs, then a boundary is said to have occurred. If an event, like sleeping, or the position of the sun i.e., the sun goes down, occurs in a text, a boundary is also considered to have occurred.

4.3.2 Spatial locations

The second major change that is an indicator of episode boundaries is shift in spatial location. This change logically lends itself to a need to identify the participants at the new location, since there is no assurance of having identical participants in a new location. Spatial locations in Olo are a little more complicated than temporal expressions. In Olo, people can be somewhere, leave somewhere, travel, arrive, and again be somewhere. This makes for a fairly wide boundary between two locations. It is theoretically unclear as to what points belong to which scene. Should “leaving” go with “being” or “travel”? The same question applies to “arrival” vis-á-vis “travel” or “being”. For this reason each of these different stages in the journey are treated as separate for coding purposes. There are verbs that mark the stages of a journey. They are not all obligatory, but each often occur. In table 4.2 the different verbs marking these stages are given. Table 4.2. Olo primary spatial verbs ratei ‘be, live, stay’ ato ‘stay temporarily’ usa ‘leave’ e ‘go’ au ‘come’ fale ‘arrive’ 4.3 Dividing the Text 65 There are a few verbs that commonly occur in travel situations besides the words for go and come. They are given in table 4.3. Table 4.3. Olo secondary spatial verbs ulsi ‘follow someone’ ingi ‘follow a trail, river, or natural feature’ itipi ‘go down’ unwei ‘go up’ The likelihood of a change being related to an episode boundary is the amount of disjunction be- tween locations. What is common in Olo narratives is the gradual shift from one location to another. This shift in location can be viewed as a boundary phenomenon. In this analysis, the degree of spatial change is assigned according to a specific formula. The higher the number, the more likely a change of location has occurred. When a precise location is specified as a location for a significant temporal du- ration, one of the existential verbs like ratei ‘be, live’ or ato ‘stay temporarily’ are used. This is consis- tent with the setting of a time for a discourse. Clauses with such verbs are given a value of 4 for spatial location. The next most likely setting of a new location involves arrival at a new place. Clauses that in- volve arrival at a specific place will have one of two verbs, either fale ‘arrive’ or naro ‘come out’. This clause will be assigned a 3 for spatial location. Motion verbs encode the intermediary step from one place to another; they are part of the boundary between two settings. The use of a pure motion verb e ‘go’ or au ‘come’ or a motion verb involving direction, like itipi ‘go down’ or unwei ‘go up’, are assigned a value of 2. The start of a border involves the leaving of a place. Once a participant has left a place he is no longer in the same spatial location, but his location has not been specified yet, so inception of travel, if marked, will be given a value of 1. Inception is marked by verbs such usa ‘leave’. In this way I have attempted to quantify the degree of change from one location to another and look at how firm the boundaries are between the two settings.

4.3.3 Punctuation

Another gauge of event coherence is punctuation. In written text, punctuation serves much the same purpose as pauses. It is used to organize the text into various levels and subgroups. In this study, the text is also marked according to the different punctuation provided by the authors in the case of the written work, and by a native Olo speaker working on a transcription of the oral story. Events are grouped together by the punctuation bracketing the clauses. The clause following a punctuation mark is given a numerical value according to the punctuation used. Clauses following “sentence” punctua- tion “.?” are given a value of 4. Any clause after a colon “:” is assigned a value of 3. A clause after a semicolon “;” is assigned a value of 2. A clause following a comma “,” is assigned a value of 1. This pro- vides a second means of determining how events are grouped together.

4.3.4 Discourse markers

Olo has a variety of lexical morphemes that link one clause chain to another. These discourse mark- ers, discussed in chapter 2, are examined to see if they interact with the choice of reference. They are used to show how tightly the units are linked or coupled within the discourse. Table 4.4 gives a list of the different links. They are broken into two different categories: sequencers and results. The sequen- cers are used to move the discourse forward along the event line. They may also mark out some type of episodesubepisode boundary. Gernsbacher 1990 proposes that starting an English sentence with certain adverbials like next or then causes a shift in episode structure. If referential form is tied to boundaries, then we can expect to see an influence of the boundary on the referential form. In this study the different discourse markers are not considered independently but are treated as a single class. We will look to see if the group as a class has any effect on referential management. 66 Methods Table 4.4. Olo discourse linkers sequencers le sequence so close link sequence lo loose sequence wo ordered sequence leye lo major sequence eventresult yo non causal result leso casual effect eite reason result eis “so that”

4.3.5 Event coherence

The third commonly cited thread of an episode is event coherence. Event coherence is associated with episode structure because some types of events go together. ‘Chopping, carrying, and stacking, firewood’ go together as a group of activities, just as ‘ordering a meal, eating it, and paying for it’ do. The events cohere as components of an overall scenario. Events can also cohere by having partial over- lap of activity or being logically related. Events that are somehow related should be easier to process and make less demands on the comprehender, which could allow more resources to be allocated to tracking participants. Also events that go together in some fashion can be expected to have the same participants, whereas radically different events do not assume the continuation of character continu- ity. This makes breaks in event coherence a likely point for reidentification of participants. Event coherence is important as a possible point that would cause the reidentification of referents. However, determining event coherence is problematical. It is very difficult to divide the story in a noncircular way. Impressionistic marking of event coherence is not without risk, as the referential forms can influence the choice of degree of cohesion, which could create a degree of circularity in the analysis. In this work, the problem is dealt with by having a rigorous operational definition that recog- nizes the scalar nature of event coherence. As long as this is applied in the fashion intended, it should not be an impediment to a proper analysis. The second problem is one of deciding the level of diver- gence of two events and why. The theoretical approach I have taken involves the amount of semantic similarity and logical relatedness. Events that are completely identical have a complete coherence. At the other end of the spectrum are events that have nothing in common. The middle is more difficult to define. Given that identical events are maximally coherent, events that are not identical, but either in a generic-specific or set-subset relationship are the next level in coherence. If two events can be divided into some set of components a, b, c, d, e and a, b, c, d so that the only difference is the presence or absence of component e, it is obvious that the only difference between these two different events and two identical events is the presence of a single component. The third step is one of logical sequence or natural consequence. In this case, the second event is highly predictable from the preceding event. It can be predictable on the basis of text frequency, because of application of natural laws or logical en- tailment. So when a participant puts something, it has to be put somewhere. While in Olo it is not an absolute requirement that a placed object be specified in its spatial relation to the world, it is commonly done; so the resultant predictability is based not only on frequency, but also on logical entailment. Operationally I am assigning scalar values to the amount of divergence one event has from the fol- lowing event. An “event” is considered a single verb for Olo. A numerical value is assigned to the sec- ond clause in the sequence as an indication of how different it is from the preceding clause. Based on the preceding discussion, two clauses with identical verbs are considered a unity sequence and as- signed a value of 1. Examples of this are au au ‘come come’, elele elele ‘fell fell a tree’, or esi esi ‘hold hold’. Also considered a unity sequence is the use of e ‘go’ to indicate continued action. An example of this is given in 142, from the text Amerika. 4.3 Dividing the Text 67 142 p-osi nemple p-e p-e p-e p-e wo 3p-smite other.3p 3p-go 3p-go 3p-go 3p-go TS they smote and smote each other until… The second clause in an event sequence that is a case of generic specific action is assigned a value of 2, for instance unwei e ‘go up go’. Also assigned a 2 are second clauses in a sequence where both clauses use one of the different terms for cookingfood preparation, e.g., apli ‘cook’, eptawi ‘roast’, weisi ‘turn’ as in 143. 143 leyelo mulpou l-epe nimou-re pel-pe pe p-apli oweli, CB night m-this woman- PL p-3p they 3p-cook.3p food And then, this night, their women, they cooked food, p-aptei weli re p-apli lufe 3p-put sago.jellies and 3p-cook.3p sago.in.bamboos they put sago jellies and cooked sago in bamboos. The second clause in an event sequence which is considered “highly related” is assigned a value of 3. These would involve all the “put” verbs and their existential counterparts, as well as sequences of “mo- tion-arrival” and “arrival-existential” sequences. 144 ki k-au fale uf I 1s-come arrive village I came and arrived in the village. It would also include schema sequences of things like ‘get carry come’. These are often realized in another language as a single verb i.e., bring in English. 145 ku kapi m-afo m-apli m-aplei we get.3p 1p-put.3p 1p-cook.3p 1p-eat.3p We got them, put them into a bamboo, cooked them and ate them. In 145 all but the first in the sequence would be assigned a value of 3. The sequence of the first verb and the second would be assigned a 4. It is a logically related event sequence, but does not seem to be as “highly related” as the rest of the sequences. Finally, a 5 is assigned to anything not assigned another value.

4.4 Referential Forms

All participants that occur in a story are given a unique internal code to keep them distinct. A partic- ipant is either an individual or a group made up of individuals who can be given a unified semantic role. A particular individual can therefore be a participant on his own and as a member of one or more groups. In 146a, Tom is an individual participant realized by a proper name. In 146b, Tom is an in- dividual participant, as is Mary. Clause 146c does not have Tom or Mary as a participant as I have de- fined them, rather the group composed of Tom and Mary is a participant. In clause 146d, the group realized by people is a participant. 146 a. Tom walked downtown to hang out. b. He met Mary at McDonalds. c. They went over to the mall. d. It was filled with people. 68 Methods General information is tracked about each participant. Notations are made for gender, person, num- ber, and animacy. The animacy values used are human, animate, and inanimate. Of the information tracked for each referent, person is of importance to this study because first and second persons are naturally in the world of discourse, while participants not involved in the speech act have to be estab- lished making them clearly distinct. The actual referential form for each particular realization of a ref- erent is tracked according to the clause in which it occurs. This allows the comparisons of specific forms with the possible conditioning environments.

4.5 Analysis