Possessives in Naturally Occurring Discourse: A Centering Approach
Centering is a theory that relates focus of attention, choice of referring expression, and perceived local coherence of utterances. The basic claims of this theory are that certain entities mentioned in an utterance are more central than others and that this property imposes constrains on a speaker’s use of different types of referring expressions. The more central an entity is, the more liable to be represented as a pronominal it is.
Possessives as a type of pronominal have peculiar characteristics. An NP that contains possessive adjective in fact refers to two different entities, the possessor (Por) and the possessed (Ped). The problem is how the two entities affect Backward-looking Center (Cb) determination and Forward-looking Center (Cf) ranking.
There are two approaches to this issue. One of them comes from Walker and Prince (1996). They assume that the Cf ranking is from left to right within the higher NP when an NP evokes multiple discourse entities. The other is from Di Eugenio (1998), which assumes that the Cf ranking between Por and Ped is affected by the animacy of Ped. If Ped is animate, she ranks it as preceding Por, and if Ped is inanimate, as following Por. To compare these two approaches, consider the following (contrived) discourse.
b. Shei was worried
c. i) Heri husbandj was in the hospital.
ii) Heri cark wasn’t working. (Di Eugenio 1998: 125)
As for the Cf ranking of (1ci), the former approach assumes Por (i) precedes Ped (j), but the latter one assumes Por (i) follows Ped (j). In other words, the former predicts that the following utterance will be about her (i) rather than (her) husband (j), but the latter predicts that it will be about (her) husband (j) rather than her (i).
To settle this problem, a corpus-based study is needed, because what it matters is how speakers choose different types of referring expressions in naturally occurring discourse. We collected and analyzed 7836 whole paragraphs that contain utterances with sentence-initial "His" from ACLDCI (Brown, Wall Street Journal 1987, 88, 89) through LDC online. For each paragraph, we examined the Cb of the utterance that contains sentence-initial "His" and that of the following utterance to see how possessives affect Cb determination and Cf ranking.
Our first 500-token set shows the following result. The actually analyzed 427 tokens are categorized into three sets according to the animacy of Ped. The "?Animacy" category can cope with such lexical items as government, company, country, etc., which denote entities which are intrinsically not animate but can be interpreted to denote the animate members of the entity.
Por |
Ped |
Others |
DNA (Initial) |
Total |
|
+Animacy |
80 (100%) |
0 |
0 |
5 |
85 |
-Animacy |
269 (96%) |
6 (2%) |
4 (1%) |
47 |
326 |
?Animacy |
15 (94%) |
0 |
1 (6%) |
0 |
16 |
Total |
364 (97%) |
6 (2%) |
5 (1%) |
52 |
427 |
Table 1. Correspondence with the Preceding Utterance ( % : relative to Total – DNA(Initial) )
Por |
Ped |
Others |
DNA (Final) |
Total |
|
+Animacy |
22 (37%) |
25 (42%) |
13 (22%) |
25 |
85 |
-Animacy |
121 (55%) |
51 (23%) |
47 (21%) |
107 |
326 |
?Animacy |
4 (36%) |
2 (18%) |
5 (46%) |
5 |
16 |
Total |
147 (51%) |
78 (27%) |
65 (22%) |
137 |
427 |
Table 2. Correspondence with the Following Utterance ( % : relative to Total – DNA(Final) )
Por has a dominant role for the link with the previous utterance, regardless of the animacy of Ped. As for the prediction to the following utterance, Por affects Cf ranking most as a whole. However, animate Ped seems to be more central in the following utterance than inanimate one. Now that our concern is the animate Ped case, we analyzed every token of animate Ped case from the corpus. The correspondence seems to be affected by what we assume the center update is. The result is as follows:
Por |
Ped |
Neither |
Both |
Total |
|
Clause Based |
337(36.55%) |
371(40.24%) |
189(20.50%) |
25(2.71%) |
922 |
Sentence Based |
358(44.86%) |
260(32.58%) |
156(19.55%) |
24(3.01%) |
798 |
Table 3. Result from the Whole Animate Ped Cases
When we assume tensed clauses as the center update units, animate Ped is more liable to be the Preferred Center (Cp) in the following utterance than Por. On the contrary, if we assume sentences as the center update units, the result is reverse. So, it seems to be hard to conclude that animate Ped should be higher than Por in the Cf ranking. However, we find that when the possessives are discourse-segment initial, the tendency that Ped is the Cp of the following utterance becomes more dominant, whatever the center update unit is. The following table shows this fact.
Por |
Ped |
Neither |
Both |
Total |
|
Clause Based |
48(29.81%) |
83(51.55%) |
27(16.77%) |
3(1.86%) |
161 |
Sentence Based |
53(33.76%) |
76(48.41%) |
25(15.92%) |
3(1.91%) |
157 |
Table 4. Result from the Discourse-Segment Initial Animate Ped Cases
To conclude, the following utterance of the utterance that contains possessives doesn’t seem to be about Ped rather than Por, even though Ped is animate. However, when a discourse segment starts with a possessive, the following utterance is more liable to be about Ped rather than Por.
References
Di Eugenio, Barbara (1998) "Centering in Italian," in Marilyn A. Walker, Aravind K. Joshi and
Ellen F. Prince, eds., Centering Theory in Discourse, Oxford University Press, New York,
115-38.
Walker, Marilyn A. and Ellen F. Prince (1996) "A Bilateral Approach to Givenness: A Hearer-
Status Algorithm and a Centering Algorithm," in Reference and Referent Accessibility, John
Benjamins, Amsterdam/Philadelphia, 291-306.