Reference, Centres and Transitions in spoken Spanish

Reference, Centres and Transitions in spoken Spanish
Reference, Centers and Transitions in Spoken Spanish *
Maite Taboada
Simon Fraser University
[email protected]
Abstract
The goal of this paper is to examine the relationship between Centering transitions
(Grosz et al., 1995) and choice of referring expression. For that purpose, Centering
analyses were carried out in two different corpora of spoken Spanish. The corpus
analysis confirms reports in previous literature about what is the typical choice of
referring expression. In some cases, however, the referring expression chosen violates
expectation, or does not follow what other researchers have found (e.g., a proper name
is used when a pronoun is expected). In those cases, the most likely explanation is that
other constraints related to spoken language are at play (turn-taking and grounding).
1. Introduction
The question that much of the research on anaphora attempts to answer is: how does a speaker
choose which referring expression to use? One assumption is that the speaker uses the referring
expression that conveys the exact amount of information that the hearer will need in order to
interpret the current utterance correctly. Given a possible choice between he, this man, the man,
and John, it is plausible that a speaker will choose one that will help the hearer link to the
intended referent with the minimum amount of effort. If the conversation has been about John
throughout, with no other male referent intervening, he is probably the most common choice. If
the speaker uses John instead, she might indicate that the hearer is to pay attention to the referent,
or that a new John has been introduced in the conversation. Any explanation needs to not only
account for the most typical realization (i.e., the expected realization), but also explain what
factors are involved when the choice is contrary to expectation. Bolinger formulates the question
in the following terms:
“At X location, what reason might the speaker have for using a word that is leaner in
semantic content rather than one that is fuller, or vice versa?” Usually this means
“Why use a pronoun?” or “Why repeat the noun?” (Bolinger, 1979: 290)
*
A previous version of this paper was presented at the Vancouver Studies in Cognitive Science Workshop in
February 2003. I would like to thank Loreley Hadic Zabala for helping in the coding of the data, and Fernando
Chicharro, Laurie Fais, Jeanette Gundel, and Nancy Hedberg for comments on a draft of the paper. This work was
supported by the Ministry of Science and Technology of Spain, and the Xunta de Galicia, under project MCYTFEDER BFF2002-02441/XUGA PGIDIT03PXIC20403PN, and by Simon Fraser University, under a SSHRC grant
and a Discovery Parks grant.
In J. Gundel and N. Hedberg (eds.) 2008. Reference: Interdisciplinary Perspectives. Oxford:
Oxford University Press, pp. 176-215.
Different explanations have been proposed to account for how the choices are made, and for
the effects of such choices, such as Gundel et al.’s (1993) Givenness Hierarchy or Ariel’s (1996)
accessibility marking scale. In these, the form of the referring expression is linked to the salience
of the referent. Other explanations emphasize the importance of first mention (Carreiras et al.,
1995; Gernsbacher and Hargreaves, 1988), or syntactic organization (Gordon et al., 1999).
In this paper, I explore a different way to explain the form of a referring expression, by
applying Centering Theory (Grosz et al., 1995). Centering Theory is a theory of local focus in
discourse that proposes different transition types between any pair of utterances. Those
transitions are based on salience, but also on the expectations that the hearer might have about the
focus of the next utterance. Researchers within Centering Theory have already proposed that
there is a relation between the form of a referring expression in a given utterance and the
transition linking that utterance to the previous one, or that Centering structures guide the
interpretation of pronouns in discourse (Brennan, 1995; Di Eugenio, 1998; Gordon et al., 1993;
Hudson-D'Zmura and Tanenhaus, 1998; Roberts, 1998; Walker, 1998). I extend that research by
applying Centering to Spanish spoken discourse.
It should be obvious that transition type is not the only factor involved: Centering proposes
four transition types; most languages number more than four choices in their repertoire of
referring expressions, meaning that more than four referring forms are possible for a given entity.
For example, Gundel et al. (1993) propose six cognitive statuses and at least seven different
referring expressions in English that denote them. That means that other factors must be at play
in the choice. The paper also explores some of those factors.
The study was carried out on two corpora of spoken Spanish. The first one, the Interactive
Systems Lab corpus, is a collection of task-oriented conversations between two speakers. The
second one is the CallHome corpus, a set of telephone conversations between relatives or friends.
A total of fourteen conversations from the two corpora were annotated according to Centering
theory.
The paper is structured as follows: Section 2 will briefly introduce Centering Theory; Section
3 describes its application to spoken discourse, in particular as regards to segmentation. Section 4
explains the process of constructing the list of entities, the Cf list. The results of the corpus
analysis are presented and discussed in Section 5, with Section 6 providing conclusions.
2. Centering Theory
Centering (Grosz et al., 1995; Walker et al., 1998) was developed within a theory of discourse
structure (Grosz and Sidner, 1986) that considers the interaction between (i) the intentions, or
purposes, of the discourse and the discourse participants, (ii) the attention of the participants and
(iii) the structure of the discourse. Centering is concerned with the participants’ attention and how
the global and local structures of the discourse affect the referring expressions and the overall
coherence of the discourse. It models the structure of local foci in discourse, i.e., foci within a
discourse segment.
Centers are semantic entities that are part of the discourse model of each utterance in the
segment. For each utterance, Centering establishes a ranked list of entities mentioned or evoked,
the forward-looking center list (Cf). The list is ranked according to salience, defined most often
in terms of grammatical relations (see Section 4). The first member in the Cf list is the preferred
2
center (Cp). Additionally, one of the members of the Cf list is a backward-looking center (Cb),
the highest-ranked entity from the previous utterance that is realized in the current utterance.
Example (1) illustrates these concepts 1 . Let us assume that the utterances in the example
constitute a discourse segment. In the first utterance, (1a), there are two centers: Harry and snort.
(1a) does not have a backward-looking center (the center is empty), because this is the first
utterance in the segment. In (1b), two new centers appear: the Dursleys and their son, Dudley.
The lists include centers ranked according to two main criteria: grammatical function and linear
order. (Ranking will be further discussed in Section 4.) The Cf list for (1b) is: DURSLEYS,
DUDLEY 2 . The preferred center in that utterance is the highest-ranked member of the Cf list, i.e.,
DURSLEYS. The Cb of (1b) is empty, since there are no common entities between (1a) and (1b). In
(1c), a few more entities are presented, and they could be ranked in a number of ways. To shorten
the discussion at this point, I will rank them in linear order, left-to-right. In any event, the most
important entities seem to be the Subject, which is the same as in (1b), DURSLEYS; and DUDLEY,
realized by in the possessive adjective his (twice). The Cp is DURSLEYS, since it is the highestranked member of the Cf list, and the Cb is also DURSLEYS, because it is the highest-ranked
member of (1b) repeated in (1c). The new utterance, (1d), reintroduces Harry to the discourse,
and links to (1c) through DUDLEY, which is the Cb in (1d).
(1) a. Harry suppressed a snort with difficulty.
b. The Dursleys really were astonishingly stupid about their son, Dudley.
c. They had swallowed all his dim-witted lies about having tea with a different member of his gang
every night of the summer holidays.
d. Harry knew perfectly well that Dudley had not been to tea anywhere;
e. he and his gang spent every evening vandalising the play park, [...]
In (2) we see the Cf, Cp and Cb for each of the utterances in the segment:
(2) a. Cf: HARRY, SNORT
Cp: HARRY – Cb: Ø
b. Cf: DURSLEYS, DUDLEY
Cp: DURSLEYS – Cb: Ø
c. Cf: DURSLEYS, DUDLEY, LIES, TEA, MEMBER, GANG, NIGHT, HOLIDAYS
Cp: DURSLEYS – Cb: DURSLEYS
d. Cf: HARRY, DUDLEY, TEA
Cp: HARRY – Cb: DUDLEY
e. Cf: DUDLEY, GANG, EVENING, PARK
Cp: DUDLEY – Cb: DUDLEY
In addition to the different types of centers, Centering proposes transition types, based on the
relationship between the backward-looking centers of any given pair of utterances, and the
1
From J.K. Rowling (2003) Harry Potter and the Order of the Phoenix. Vancouver: Raincoast Books (p. 8).
Small capitals indicate that the list contains entities, not their linguistic realization. The reference to Dudley is
conveyed by two different referring expressions: their son and Dudley.
2
3
relationship of the Cb and Cp of each utterance in the pair. Transitions, shown in Table 1, capture
the introduction and continuation of new topics. Cbi and Cpi refer to the centers in the current
utterance. Cbi-1 refers to the backward-looking center of the previous utterance. Thus, a
CONTINUE occurs when the Cb and Cp of the current utterance are the same and, in addition, the
Cb of the current utterance is the same as the Cb of the previous utterance. Transitions capture
the different types of ways in which a conversation can progress: from how an utterance refers to
a previous topic, the Cbi-1, and it is still concerned with that topic, the Cpi, in a CONTINUE, to how
it can be not linked at all to the previous topic, in a ROUGH SHIFT. Transitions are one explanation 3
for how coherence is achieved: a text that maintains the same centers is perceived as more
coherent.
In Example (1), the first utterance has no Cb, because it is segment-initial, and therefore no
transition (or a zero-Cb transition). The transition between (1a) and (1b) is also zero. Between
(1b) and (1c) there is a CONTINUE transition, because the Cb of (1b) is empty, and the Cp and Cb
of (1c) are the same, DURSLEYS 4 . Utterance (1d) has a different Cb from (1c), and it also shows
different Cb and Cp, producing then a ROUGH SHIFT in the transition between (1c) and (1d).
Finally, (1e) and (1d) are linked by a RETAIN transition.
Cbi=Cbi-1
or Cbi-1= Ø
Cbi≠Cbi-1
Cbi=Cpi
CONTINUE
SMOOTH SHIFT
Cbi≠Cpi
RETAIN
ROUGH SHIFT
Table 1. Transition types.
Because transitions capture topic shifts in the conversation, they are ranked according to the
demands they pose on the reader. The ranking is: CONTINUE > RETAIN > SMOOTH SHIFT > ROUGH
SHIFT. This transition ranking is often referred to as Rule 2 in the Centering paradigm. Centering
predicts that CONTINUE will be preferred to RETAIN, and RETAIN to SHIFTS, all other things being
equal. The preference applies both to single transitions and to sequences of transitions.
Rule 1 captures the preference for pronouns when the same topic of discourse is continued.
The formulation of Rule 1 is as follows:
For each Ui in a discourse segment D consisting of utterances U1, ..., Um, if some element of
Cf(Ui-1, D) is realized as a pronoun in Ui, then so is Cb(Ui, D).
Rule 1 is sometimes referred to as the Pronoun Rule. It captures the fact that a topic that is
continued from a previous utterance does not need to be signalled by more explicit means than a
pronoun (or a zero pronoun, in languages that allow those). Other pronouns are of course allowed
in the same utterance, but the most salient entity must be realized by the least marked referring
3
Centering transitions are just one explanation for coherence. A text can be coherent without repeating or referring
to the same entities (Brown and Yule, 1983: 195-199; Poesio et al., 2000).
4 Other proposals suggest that transitions for utterances after an empty Cb should be different: if Cb is not empty,
i
but Cbi-1 is, the transition is a CENTER ESTABLISHMENT; if Cbi is empty and it follows an also empty Cbi-1, the
transition is NULL. It is only when Cbi is empty, and Cbi-1 is not that we have a ZERO transition (Kameyama, 1986;
Poesio et al., 2004).
4
expression. In (1c), the backward-looking center, DURSLEYS, is realized as a pronoun, following
Rule 1, since other pronouns are also present in the utterance (his to refer to DUDLEY).
Relationships have been established between the transition type between a pair of utterances,
and the type of referring expression chosen to realize entities in the second utterance in the pair.
Di Eugenio (1998) found that CONTINUE transitions, because they keep the same center, often
encode the subject as a zero pronoun in Italian. Shifts (smooth or rough) result in less
pronominalization. We will see that these relationships are quite complex, and different factors
come into play in the choice of referring expression.
3. Centering and spoken language
The Centering framework has been applied to both constructed examples and naturally occurring
discourse, but not widely to spontaneous conversation. There are a number of issues involved in
such application, namely the segmentation into Centering units (utterances), the presence of false
starts and backchannels, linearity and overlap, and the presence of first and second person
pronouns. I discuss each one of those in this section.
The approach taken here to apply Centering to spoken dialogue owes much to the work done
by Byron and Stent (1998). They report experiments on different variations of segmentation,
false starts, inclusion of first and second person pronouns, and linearity. The model for dialogue
adopted here is Byron and Stent’s Model 1, that is, a model where both first and second person
pronouns are included in the Cf list. In addition, utterances are consecutive: in the search for Cbn,
only Cfn-1 is searched, whether it was produced by the same speaker or not. Byron and Stent
(1998) found that this model performed better than models that discarded first and second person
pronouns, and models that considered previous or current speaker’s previous utterance 5 .
3.1 Utterance segmentation
The first step in a Centering analysis involves deciding on the minimal units of analysis,
commonly referred to as ‘utterances’. The notions of discourse segment and utterance are very
important: Centering predicts the behaviour of entities within a discourse segment; centers are
established with respect to the utterance. In this paper, I use the term ‘utterance’ or ‘segment’ to
refer to the units of analysis in Centering Theory. In other applications, ‘segment’ or ‘discourse
segment’ refers to the broad parts into which a discourse can be divided (e.g., introduction, thesis
statement), or to discourse segments that achieve a purpose each (Grosz and Sidner, 1986). I am
not concerned with those higher-level discourse segments here, but only with minimal units of
analysis, typically interpreted to be either entire sentences or finite clauses. These concerns are
general to Centering applications, but even more pressing when dealing with spoken language,
where the notion of sentence is more difficult to instantiate. That is why, in spoken language,
traditional notions of clause and sentence are abandoned in favour of the idea of an utterance
(Schiffrin, 1994).
5
Their performance measures were based on (i) number of zero Cbs, (ii) whether the Cb that Centering found
corresponded with a loose notion of sentence topic, and (iii) number of cheap vs. expensive transitions. The
cheap/expensive distinction refers to inference load on the hearer (Strube and Hahn, 1999), according to whether
Cpn-1 , expected to be Cbn, is actually realized as such.
5
In general, an utterance is an intonation unit. In the corpora studied, utterances are already
marked in the transcripts. For the ISL corpus, an utterance is defined as an intonation unit marked
by either a period or a question mark. Note that a comma does not always define an utterance. In
Example (3), the period after Miriam indicates falling intonation, as in the end of a sentence.
There are, therefore, two Centering units in (3) 6 .
(3) a. Miriam.
‘Miriam.’
b. yo creo que /uh/ no nos va a alcanzar el tiempo.
‘I believe that, uh, we won’t have enough time.’
In the CallHome corpus, utterances, at the first level of granularity, are equivalent to dialogue
acts, which were assigned to the Spanish CallHome corpus (Levin et al., 1999). In this corpus,
the speech act was more important than intonation when it came to segmenting speech into
utterances. The following example was segmented into two dialogue acts, which also correspond
to two tensed clauses.
(4) a. Se supone que hay mucho ganado,
‘Supposedly there are a lot of animals,’
b. pero yo no vi nada.
‘but I didn’t see any.’
Pauses also indicate a new segment, whether a segment was introduced already in the
transcripts or not. Example (5) was one unit, but since a pause exists after de él, the second part
was considered to be a new Centering unit.
(5) a. claro, pero, o sea, él, según él, soy el socio de él [pause]
‘right, but, I mean, he, according to him, (I) am his partner’
b. según él, ¿no es cierto?
‘according to him, right?’
Segmentation into utterances has been a topic of study in the Centering literature. In the
analysis, I have followed Kameyama’s (1998) proposals for intra-sentential Centering. They
consist of separating any tensed coordinate or subordinate clauses from their matrix, and of
including report complements and reported speech together with the reporting units. Tenseless
subordinate clauses are part of the matrix clause 7 . In addition to the segmentation already in the
corpora (utterances and dialogue acts), complex clauses are broken up according to Kameyama’s
rules. Tensed adjuncts are separated from the main clause, as in Example (6).
(6) a. No compro nada, no nada, nada
‘(I) don’t buy anything, nothing, nothing’
6 Spanish examples are glossed word-by-word only when the gloss provides information considered relevant. In all
other cases, they are translated as close to the original as possible, which may sometimes make them sound awkward.
Parentheses around a pronoun in the translation indicate that it is null in Spanish. Slashes (/eh/) indicate filled pauses
or backchannels. Angle brackets (<de>) indicate false starts.
7 For a more detailed explanation of the segmentation, see Hadic Zabala and Taboada (2004) and Taboada and Hadic
Zabala (2005).
6
b. porque quiero irme a ver a mi hermana.
‘because (I) want to go see my sister.’
Kameyama (1998) considers reported speech a hierarchical unit, embedded with the reporting
unit, and I followed that approach. That is, in cases where reported speech appears, the reported
unit is processed, and Centering structures are created within it. But once it has been processed,
the next unit looks back to the reporting unit for antecedents, and for Cb comparison purposes. I
also included relative clauses together with their antecedent NP, i.e., relative clauses were treated
as embedded. Poesio et al. (Poesio et al., 2000; 2004) report that this produces fewer violations of
Centering constraints (specifically, of Constraint 1, that all utterances of a segment, except the
first one, have one Cb).
The final issue in segmentation was the speech addressed to a third party. In CallHome
conversations, which are on the telephone, one of the interlocutors sometimes directs speech to
another person on his or her side of the line. This was recorded, and quite likely audible to the
other interlocutor. I considered speech directed to a third party as a separate Centering unit, and
included it in the Centering analysis, because entities mentioned in the speech to the third party
often appear in the conversation between the main interlocutors. We can see an illustration in (7).
The speakers, A and B, are debating how long they have been on the phone (7a and 7b). Speaker
B then asks somebody else (mamá), and reports back the answer. The vocative mamá is included
in the Cf list of (7c) 8 . A Centering analysis including (7c) shows that speech directed to a third
party must be included in the analysis since it contains the antecedent for the null pronoun in
(7d), which is speech directed at A, and as a consequence part of the main conversation. Without
(7c), the transition between (7b) and (7d) is a zero transition (no Cb).
(7) A: a. ¿Te late que como quince?
‘Does fifteen (minutes) sound about right?
B: b. Pues no sé yo.
‘Well, I don’t know.’
c. llevamos como quince minutos, mamá?
‘Have (we) been (talking) for about fifteen minutes, Mom?’
d. dice que más o menos.
‘(She) says that more or less.’
The segmentation was performed by two annotators separately. We first segmented one
CallHome and four ISL conversations as training, compared the results and refined the coding
manual (Hadic Zabala and Taboada, 2004). Then an evaluation was performed, segmenting four
additional CallHome conversations, which amounted to 895 segments in the final agreement. The
disagreement in those 895 segments was 18.7% of the total. This included any instance of
disagreement (two instead of one segments, or vice versa, or disagreements in the inclusion of
segments for the analysis). The high disagreement rate is due to problems in interpreting spoken
data (boundaries are not clear), deciding on whether to include inferables (if an utterance contains
8
I believe vocatives should be part of the Cf list (see Lambrecht, 1994 about vocatives being topics, and therefore
referential), but I am not sure where they belong in the Cf ranking. The current coding includes them in the highest
position, following Lambrecht’s (1994) suggestion that they are topics.
7
no entities, it is not considered a unit for the analysis), and, to a lesser extent, also due to human
error. Current efforts are directed toward making the coding manual more transparent, and
devising a training process, which might include segmenting on-line, without looking ahead, as
Brennan (1995) suggests.
3.2 False starts and backchannels
An utterance may not be complete syntactically, but still include referential information that
affects the rest of the discourse. Some of these incomplete utterances are referred to as false
starts. In the analysis, I considered false starts that included some referential information, whether
the utterance was complete or not, which was also the approach followed by Eckert and Strube
(1999). Most of those false starts were not utterances in themselves. For instance, in (8), the
speaker introduces te (‘you’), but then changes her mind, and produces a different sentence. The
entity you, however has already been introduced, and therefore it has to be considered as part of
the Cf list.
(8) bueno. <te> /mm/ entonces quedamos así.
‘Good. you mm then (we) agree on that.’
Following Byron and Stent (1998), “empty utterances”, that is, utterances that contain no
discourse entities, are attached to their preceding or following utterance, according to context.
This applies to empty utterances across turns as well, so that backchannels (Yngve, 1970) are
ignored for Centering purposes. (9b) is a backchannel signal, making (9a) and (9c) the adjacent
utterances for Centering.
(9) A: a. Me levanto a las siete
‘(I) get up at seven’
B: b. Sí.
‘Yes’
A: c. empiezo las clases de ocho a nueve cuarenta
‘(I) start class from eight till nine forty’
3.3 Linearity and overlapping
A conversation is the combined effort of two or more participants. Reference passes back and
forth between speakers, producing a sense of coherent whole for the entire conversation. As a
consequence, I considered that Centering transitions applied from one utterance to the next,
regardless of whether the two utterances were produced by the same speaker or by different
speakers, in line with Byron and Stent’s (1998) proposal. This applies when the turns are actually
floor-holding (Edelsky, 1981), rather than backchannel signals, as in Example (9) above.
Example (10) shows two turns. The centers in B’s turn include an entity in A’s turn, a reference
to B herself.
(10) A: a. qué tal te viene?
‘how is (that) for you?’
Cf: MEETING (null), B (te, ‘you’)
8
B: b. no. te contesté recién que /eh/ hoy viernes yo no puedo.
‘no. (I) just told you that uh today Friday I can’t.’
Cf: B (‘I’, null), A (te, ‘you’), FRIDAY
Cb: B
3.4 First and second person pronouns
Spoken language usually contains a high number of first and second person pronouns. Centering
was devised explicitly with third person pronouns in mind, and most applications of Centering do
not take first and second person pronouns into account. Byron and Stent (1998) found that it was
necessary to include them in the Cf list. This is certainly the case in the data, where the
antecedent for null first and second person pronouns is to be found in previous utterances. In the
following example, I and you in (11b) are linked to we in (11a). Of course, part of that reference
is situational, but it can certainly be included in a Centering analysis.
(11) a. Mónica. /eh/ te parece que nos juntemos algún día en la mañana, toda la mañana entera? y
trabajemos?
‘Monica uh what do you think (we) get together some day in the morning, all morning? and
work?’
b. así que querría saber si vos el miércoles diecisiete podés.
‘So (I)’d like to know if you can Wednesday the 17th in the morning.’
First and second pronouns, in this data, constitute a large number of the entities for each
utterance; in fact, the only entities in many cases. Were they not included in the Cf list, we would
find many more instances of transitions with no backward-looking center.
4. What is salient? And how much?
Once the conversations have been segmented into utterances to be considered for the Centering
analysis, the next step is to assign a Centering structure to each one of them. This involves (i)
building the Cf list (the list of forward-looking centers), (ii) determining the Cb, and (iii)
establishing which transition holds between two consecutive utterances. The thorniest of those
tasks is the construction of the Cf list. In this section, I discuss the different issues involved in
populating the Cf list.
4.1 Entity realization
In Centering, the list of forward-looking centers is a partial ordering of the entities realized in the
utterance. Precisely what the definition of ‘realized’ is, and what criteria we should use for that
ordering are the two problems in ranking the Cf list, that is, in deciding which entities are salient
in the discourse, and how salient they are in relation to each other. The definition of ‘realize’
depends, according to Walker, Joshi & Prince (1998: 4), on the semantic theory one chooses. But,
in general, “realize describes pronouns, zero pronouns, explicitly realized discourse entities, and
those implicitly realized centers that are entities inferable from the discourse situation”.
9
Cornish (2005) argues, in general, that entities in focus are not only those that have been
explicitly introduced in the discourse. We need to consider, then, inferable entities. Inferable
entities are of particular importance in dialogue because it relies more than monologue on the
context outside the text proper. To populate the Cf list, indirect realization of entities was
permitted: null subjects; member-set relations (Mom-Mom and Dad) and part-whole relations
(branches-trees). A strict direct realization (where the entities have to be mentioned explicitly in
the utterance) resulted in a large number of empty Cbs. What exactly an indirectly realized entity
is may, of course, not be obvious. I used the relations identified by Halliday and Hasan (1976) as
lexical cohesion (synonymy, hyponymy, superordinate, but not collocation, which does not
necessarily involve reference to the same entity). Particularly difficult in this respect were
decisions having to do with dates and times, and how those are related to each other. I considered
mostly ‘include’ relations (Hurewitz, 1998), such that, for instance, a date was deemed to be
related to the previous utterance’s Cf list if it was part of a date range mentioned there. However,
when the date was not within the time frame established, it is plausible to think that the hearer
had to construct a new model for it. In Example (12), speaker A proposes the week of the fourth,
after having discussed the previous week. However, speaker B returns to the previous week, and
mentions Friday, October 1st, i.e. a date not in the week of the fourth. This is a new entity, and
cannot be related to the immediately preceding utterance. As it happens, this results in a empty
Cb, since there are no entities in common between the two utterances.
(12) A: … quieres tratar la semana de cuatro?
‘... do you want to try for the week of the 4th?’
B: qué te parece el viernes primero de octubre, luego de las once de la mañana?
‘what do you think of Friday October 1st, after 11am?’
Spoken language tends to leave much unsaid. That characteristic poses further problems for an
account of the ‘realize’ constraint in Centering. It has been proposed that bridging inferences
(Clark, 1977) can be used to relate entities between utterances. In Example (13a), speaker A
mentions Internet, which is continued in (13b) and (13c), in two null subjects. In (13d), speaker B
does not refer to Internet at all, but introduces computer in the conversation, with a definite
article. Usually, there would be no connection between (13c) and (13d): the Cf list for (13c)
includes only INTERNET, and the Cf list for (13d) is: B (THE SPEAKER), COMPUTER. However,
computer is an inferable (Prince, 1981), a computer being needed to access the Internet, and it
can therefore become the Cb of (13d), picking up on Internet in (13c).
(13) A: a. estoy conectado con Internet y todo
‘(I)’m connected to Internet and all.’
B: b. qué tal
‘How’s (that)?’
A: c. es bárbaro
‘(It)’s great.’
B: d. yo no me pude comprar la máquina todavía, loco
‘I haven’t been able to buy the computer yet, man.’
Example (14) shows another instance of an inferable entity. The speaker in (14a) says that he
wrote ‘a lot’ (muchísimo is an adverb). In the next utterance, he says that ‘(they) don’t arrive’.
10
The plural null pronoun can be interpreted as being a reference to the product of his writing,
probably letters. The two utterances were considered to have LETTERS in common, which is then
the Cb of (14b).
(14) a. Escribí
muchísimo,
write:1SG.PAST very.much
‘I wrote a lot,’
b. lo que
the what
pasa
es que no llegan.
happen:3SG.PRES is that not arrive:3PL.PRES
‘what happens is that (they) don’t arrive.’
Null, or zero, subjects, are common in Spanish, but always recoverable from the context and
the morphology of the verb, and are always added to the Cf list. Ambiguous cases do occur, just
as pronouns in English can be ambiguous. Those are disambiguated to the most plausible referent
when creating the Cf list. There exist other instances of implicit entities, beyond null subjects. In
Example (15), the conversation is clearly about children, those of both interlocutors. However,
children are only mentioned once, in the first turn. We have to assume that they are implicit in the
rest of the exchange, as are the subjects, so that the sentences read: Do you have children? and
We don’t have children yet. The summary in (16) represents the two lists of entities of the
exchange, depending on a literal interpretation, or one that allows inferable entities 9 . I decided to
use the one on the right, which includes all the entities inferable from the context. It is plausible
to assume that those entities are in the focus of attention throughout the exchange.
(15) B: a. ... ¿Y chicos?
‘And children?’
A: b. Sí. Todavía no
‘Yes. Not yet.’
B: c. ¿Ah?
‘Huh?’
A: d. Todavía no
‘Not yet.’
B: e. ¿Todavía no?
‘Not yet?’
A: f. ¿Ustedes?
‘You (plural)?’
B: g. Ah bueno, dos ya
‘Ah well, two already.’
9
There is one further complication in (15), a request for repetition in (15c). In (16) I have excluded that turn, since it
does not contain entities, under either view (with or without inferable entities).
11
(16) Dialogue
Cf list without
inferables
Cf list with
inferables
B: And children?
children
A, children
A: Yes, not yet
-
A, children
A: Not yet
-
A, children
B: Not yet?
-
A, children
A: And you?
B
B, children
B: Ah well, two already
2 (children)
B, 2 children
4.2 Cf ranking
The ranking of the entities in the Cf list is most often performed by following grammatical
relations. Thus, subjects are ranked higher than objects, and these higher than adverbials. In English,
this results in the following order (Walker et al., 1998):
(17) Subject > Object(s) > Other
The ranking is, however, not fixed, and considered to be language-dependent. When a new
language is considered, a Cf template (Cote, 1998) for that language needs to be developed.
Several languages have been studied using Centering, and thus different templates exist. For
instance, the template for Japanese includes topic markers (wa) and empathy markers on verbs,
resulting in the following template (Walker et al., 1994).
(18) (Grammatical or zero) Topic > Empathy > Subject > Object2 > Object > Others
Di Eugenio (1998) also ranks empathy highest in her template for Italian, following Turan’s
(1995) for Turkish. Turan and Di Eugenio take the notion of empathy from Japanese, and view it
as reflected in psychological verbs (interest, seem), perception verbs (feel, appear) and certain
expressions that refer to point of view (in her opinion). There are proposals to incorporate other
factors in the Cf template, such as Strube and Hahn’s (1999) use of discourse status, whether
hearer-old or hearer-new (Prince, 1981), to analyze German. Cote (1998) uses Jackendoff’s
(1990) Lexical Conceptual Structures. Gordon, Grosz and Gilliom (1993) discovered that both
grammatical function and surface order had a role in giving an entity prominence within the Cf.
In the next few sections I discuss some of the factors that affect Cf ranking in Spanish.
4.3 Empathy and animacy
Spanish is a pro-drop language; subjects do not need to be realized as pronouns if they are known
in context. Additionally, it has direct and indirect object clitics (unstressed pronouns).
Corresponding stressed object pronouns are possible for animate entities only. I mainly follow
grammatical relations as the basis for ordering the Cf list in Spanish. Therefore, subjects are
ranked higher than objects, whether they appear as full pronouns, or as null pronouns.
There are two other criteria that play a role in the Cf ordering in Spanish: empathy and
animacy. Following Di Eugenio (1998), I take empathy with the speaker or hearer over strict
word order as a ranking criterion. Empathy, as defined by Kuno (1987: 206), “is the speaker’s
12
identification, which may vary in degree, with a person/thing that participates in the event or state
that he describes in a sentence.”
There are no studies, to my knowledge, of how empathy and point of view are expressed in
Spanish, in general 10 . The main place where I observe empathy-related effects is in the argument
structure of psychological verbs. In those, the point of view taken is that of the experiencer,
regardless of whether it is the subject or not (e.g., ‘it seems to me’, ‘I think’, and the like). In (19)
the speaker is the highest-ranked entity, because it is the experiencer of a psychological verb
(parece). In this case, the experiencer is encoded with clitic doubling (Fernández Soriano, 1999;
Suñer, 1988): the PP a mí, plus the clitic me. In Example (20), the clitic me refers to the speaker,
for whom Thursday is a better date 11 .
(19) a mí me
parece
también, bueno de hacer una reunión,
to me CL.1SG seem:3SG.PRES too
good of do:INF a
meeting,
‘It also seems good to me to have a meeting,’
Cf: I (a mí, me), TO HAVE A MEETING, MEETING
(20) me
viene
mejor el jueves,
CL.1SG come:3SG.PRES better the Thursday
‘Thursday is better for me.’
Cf: I (me), IT (the meeting, null), THURSDAY
However, the point of view criterion need not apply to the speaker only. In (21), the point of
view is that of the interlocutor.
(21) este qué tal para ti,
del
quince al
so how for you:SG from.the fifteen to.the
diecinueve.
nineteen
‘So, how is it for you from the fifteenth to the nineteenth?’
Cf: YOU (para tí), IT (the meeting, null), FROM THE 15TH TO THE 19TH
A number of verbs in Spanish follow this pattern (“me conviene”, “me viene mejor”, “se me
hace que”, it’s good for me, it’s better for me, it seems to me). Thus, for these verbs, the thematic
role of experiencer takes precedence over the grammatical function of subject. Empathy also
includes verbs with clausal grammatical subjects, but with an animate experiencer, or person
from whose point of view the statement is to be interpreted. In (22), there is a displaced clausal
subject, ‘to meet with you that day’. The subject is included in the Cf list as a single entity. The
speaker is the most salient entity, represented in para mí ‘for me’.
(22) así que
so that
para mí sería
imposible juntar-me
con vos
/eh/ ese día
for me be:PRES.COND impossible join:INF-CL.1SG with you:SG uh that day
‘So it would be impossible for me to meet with you that day.’
Cf: I (para mí), TO MEET WITH YOU, THAT DAY
10
Although Wanner (1994) and Heap (1998) discuss how empathy affects the ordering of clitics in Spanish.
Abbreviations used in the examples: 1/2/3 – first/second/third person; CL – clitic; NOM – nominative; ACC –
accusative; DAT – dative; SG – singular; PL – plural; FEM – feminine; MASC – masculine; POSS – possessive; PRES –
present; PRET – preterite; INF – infinitive; GER – gerund; SUBJ – subjunctive; COND – conditional.
11
13
Not all experiencers, however, seem to be good candidates for higher placement. In a sentence
like Juan asusta a María, ‘John frightens Mary’, the subject Juan seems to me to be more
prominent than María, although María is an experiencer. It is possible that experiencers are
ranked higher only when they are first and second person, which also happen to be higher in most
hierarchies of animacy 12 .
Animacy is a relevant feature in the ordering of clitics and reflexive pronouns that refer to
participants in the discourse. Animacy is considered relevant in general for salience and topicality
(Givón, 1983). Stevenson et al. (1994) found that animacy has a role in deciding which entity
will be in focus, and it was also found to have an effect in pronominalization (GNOME, 2000) 13 .
Clitics and reflexive pronouns, in addition to conveying empathy (see above), are also placed
before the verb, linearly before (clitic) direct objects (whether empathy is involved or not) 14 . It is
usually the case that indirect objects are animate, whereas direct objects may not be. In summary,
three reasons speak for ordering the objects as indirect before direct: (i) indirect objects can
convey empathy; (ii) indirect object clitics are always placed before direct object clitics; (iii)
indirect objects tend to be animate. Wanner (1994) argues that clitic sequences in Spanish obey
constraints of empathy and animacy. An illustration is to be found in (23), where the indirect
clitic se ‘to her’ precedes the direct lo ‘it’, which refers to a scholarship for a program that was
given to the speaker’s sister. Notice that the null subject is arbitrary (see below), and thus ranked
last.
(23) a. Mi hermana solicitó un programa de arqueología y antropología en Grecia.
‘My sister applied to a program in archeology and anthropology in Greece.’
b. ¡Y que
and that
se
lo
CL.3SG.DAT CL.3SG.MASC.ACC
dan!
give:3PL.PRES
‘And they give (gave) it to her!’
Cf: SISTER (se, ‘to her’), PROGRAM (lo, ‘it’), THEY (null)
4.4 Cf proposal for Spanish
Subjects take precedence in the Cf list in most other cases (i.e., when they are not clausal, and
when there are no experiencers). Accordingly, the elements of the Cf list follow the order in
(24) 15 . This ranking applies first to main (matrix) clauses, and then to subordinate clauses, when
the two are within the same Centering unit (usually, because the subordinate clause is non-finite;
see Section 3.1 on segmentation).
(24) Experiencer > Subj > Animate IObj > DObj > Other > Impersonal/Arbitrary pronouns
At the end of the ranking are null arbitrary subjects (Jaeggli, 1986), as in (23) above, and
subjects in impersonal constructions with se, as in Example (25). The word se in this example
12
Thanks to Jeanette Gundel and Nancy Hedberg for bringing up this point and suggesting the example.
Zaenen et al. (2004) discuss previous literature on the importance of animacy in a number of areas, including the
choice between Saxon genitive and the of-genitive, which may affect ranking in Centering.
14 See Heap (1998) for an Optimality Theory account of how empathy is also involved in non-standard
rearrangements of clitics.
15 This Cf template is slightly different from previous proposals (Taboada, 2002a, 2002b).
13
14
indicates a non-specific subject in an impersonal middle voice construction (Mendikoetxea,
1999), meaning “one can hear that you are well”.
(25) Ya
se
already se
te
CL.2SG.ACC
oye
muy bien.
hear:3SG.PRES very well
‘You already sound very well.’
Cf: YOU (te), ONE (se)
Also included as impersonal pronouns are instances of the second person singular, which can
be used impersonally (Butt and Benjamin, 2000). It is interesting to note that this second person
form is often used as an indirect form of reference to the speaker. In Example (26), the speaker is
implying that he has to take one exam every year. The tú form might indicate simply that that’s
the norm, and he is no exception. If we were to consider that the second person form has some
reference to the speaker, its ranking in the Cf list would have to change to: I (SPEAKER), EXAMS,
EVERY YEAR, since the subject is the second person singular. The sentence, however, seems to be
more about the exams than about who has to submit them.
(26) a. Son, son los tutoriales.
‘(They) are the exams.’
b. Tienes
que presentar uno
have:2SG.PRES that submit:INF one
cada año.
every year
‘(You) have to submit one every year.’
Cf: EXAMS (uno), EVERY YEAR, ONE/YOU (null subject)
4.5 Noun phrases with more than one entity
A few other issues need to be addressed in the Cf ranking. The first is related to noun phrases that
contain more than one referent or entity, whether possessives (my brother, my letter), nouns with
a prepositional phrase (the census of the city), or conjoined NPs (Juan and María). For
possessives I follow Di Eugenio (1998): the possessor is ranked before the possessed, if the
possessed is inanimate, and the possessor after the possessed, if the possessed is animate 16 . In
(27), the ranking of mi examen (‘my exam’) is SPEAKER > EXAM. However, in (28), the ranking of
mi mamá (‘my Mom’) is MOM > SPEAKER.
(27) Una maestra este, me
tuvo
que venir
a
a teacher eh
CL.1SG have:3SG.PRET that come:INF to
hacer
mi último examen aquí.
exam
here
make:INF my last
‘A teacher uh, had to come and give me my last exam here.’
16
Gordon et al. (1999) suggest that the head of the NP (i.e., the possessed) is always the most salient. However, their
experiments were based on NPs with animate possessor and possessed. The experiments were designed to test (and
debunk) a linearity hypothesis (Gernsbacher and Hargreaves, 1988; Walker and Prince, 1996), but they were all
conducted in English. Further crosslinguistic experiments would be desirable: in Spanish, and in other languages,
possessives with two full NPs (e.g., Mary’s letters) have a different word order (las cartas de María). Tetreault
(2001) shows that an anaphora resolution algorithm performs better using Gordon and colleagues’ ranking—though
the corpus used was English as well.
15
Cf: TEACHER, I (mi), EXAM, HERE
(28) mi mamá posiblemente llegue
la
my Mom possibly
arrive:3SG.PRES.SUBJ the
otra semana
other week
‘My Mom will probably arrive next week.’
Cf: MOTHER, I (mi), NEXT WEEK
The same principle applies to noun phrases with a PP modifier usually headed by ‘of’ (de in
Spanish). In most of those constructions, the meaning is that of a genitive (las cartas de Marta =
Marta’s letters). The approach taken here is different from Walker and Prince’s (1996) Complex
NP Assumption, which ranks NPs with a possessive determiner in linear order, left-to-right.
Since I are considering animacy as a relevant feature, I preferred to follow Di Eugenio’s ranking
for possessives, and to expand it to other NPs that include more than one entity. Thus, in
Example (29), una de Marta refers to one (letter) from Marta. Since Marta is animate, it is
ranked higher than letter.
(29) Y
una
de Marta.
and one
of Marta
‘And one (letter) from Marta.’
Conjoined NPs activate as most salient entity the group denoted by the conjoint. Thus, in John
and Mary, the most salient entity is the group JOHN AND MARY. The individual entities, JOHN and
MARY, are less salient than the group (Gordon et al., 1999). In that same paper, Gordon and
colleagues suggest that the individual entities are equally salient. The mention of either JOHN or
MARY results in the same processing time in a psycholinguistic experiment. It could be argued
that this result would lead to multiple entities in the same position within the Cf list, as in (30),
where the separate entities JOHN and MARY occupy the same place in the Cf list. However, I feel
that allowing multiple entities in the same position would make ranking too complex, and would
also complicate future attempts at implementing these methods in an anaphora resolution
system 17 , and prefer to use linear order to sort the two entities (31).
(30) John and Mary went to the store.
Cf: John and Mary,
John,
store
Mary,
(31) Cf: John and Mary, John, Mary, store
4.6 Wh-pronouns
Wh-pronouns, qué (‘what’), quién (‘who’), cuándo (‘when’), are included in the list of forwardlooking centers, and are ranked according to the syntactic role they have in the clause. Although
wh-pronouns do not have a specific referent, they do serve as antecedents for other referring
expressions. According to Halliday (Halliday, 1967; Halliday and Matthiessen, 2004), wh-words
can be Themes in a clause, and I believe that they can establish cohesive ties throughout a text 18 .
17 Poesio et al. (2004) discuss the need for a second criterion when two entities may be ranked in the same place.
They use linearity.
18 Pesetsky (1987; 2000) proposes that some wh-words are D(iscourse)-linked, that is, they ask a question whose
answer is drawn from a salient set. However, he says that only which questions are D-linked. I think that all wh-
16
In (32b), qué ‘what’ is included in the Cf list, and used as an antecedent for ecología in (32c),
thus becoming the Cb of that utterance.
(32) B: a. se va a la Universidad de Gales, del Sur, donde estudió Sarucán, también.
‘She is going to the University of South Wales, where Sarucán studied as well.’
A: b. A hacer qué.
‘To do what?’
B: c. Este. A hacer ecología.
‘Eh, to do environmental science.’
4.7 Reference through more than one expression
An utterance may contain reference to the same entity through more than one referring
expression. For instance, the utterance in (33) contains reference to the subject both through the
null subject pronoun and through a clitic (nos). In Centering we are usually concerned with the
entities mentioned in the utterance, not so much with the referring expression(s) used to evoke
them. However, since my concern in this paper is the link between Centering transitions and
referring expressions, this was an important issue. The ranking of such entities is straightforward:
the most salient grammatical function (or other criterion that may apply) is used to list the entity
in the Cf list. The problem is which form should be used to categorize the form of the Cb in that
utterance (see Table 4 below). I have, for the time being, categorized such examples under the
most marked form of reference. In Example (33), the referring expression used to denote the
entity “first person plural” is listed as a clitic, not as a null pronoun (clitics are considered more
marked than null pronouns). It could be argued that the least marked form should be used to
classify the Cb, but that would not show the fact that the Cb is, in a way, reinforced by another
referring expression, by being referred to twice in the same utterance.
(33) nos
CL.1PL
vamos
go:1PL.PRES
con mi madre
with my mother
‘(We) are going with my mother.’
The verb be (ser and estar in Spanish) functions as a linking verb, so subjects and predicates
(nominal and adjectival) of the verb to be are coreferential and only need to be listed once in the
Cf list. In (34), there are two references to the person the speaker is talking about, his teacher.
The first reference is through the null subject, and the second through the predicate noun, amiga.
The case is similar to the one above, where two referring expressions are used to refer to the same
entity. As above, I classified the most marked one (NP in this case).
(34) porque aparte es
mi amiga,
because besides be:3SG.PRES my friend:FEM.SG
‘because (she)’s also my friend,’
It is possible to have only a predicate (elliptical subject and predicator) in an utterance. In
these cases, since the predicate is coreferential with the elliptical subject of the elliptical
predicator, I include the subject in the list of forward-looking centers. In Example (35), the
words establish a link between the question and its answer.
17
speaker refers to himself with ‘covered’. Although there is no predicator in the sentence,
reference to the speaker is included as if a null subject were present.
(35) Lleno
full:MASC.SG
de granitos, no, este
of zits
no eh
‘(I’m) covered in zits.’
In most cases, the subject and the nominal predicate have exactly the same reference. In some
cases, the reference may be slightly different: The dinner choice is pasta 19 . Miltsakaki and
Kukich (2004) label these predicates as specificational (and predicates such as the one in
Example (34) as predicational). They rank specificational predicates higher than their
corresponding subjects. I did not make such distinction, and treated all linking verb predicates in
the same manner, as described above: the first (subject) reference determines the location in the
Cf list; the predicate determines the type of referring expression used to refer to the Cb, if the
entity in question is the Cb of the utterance.
4.8 Right and left-dislocation
The ordering of the Cf list is affected by other factors, among them right and left-dislocation. I
have not, for the moment, dealt with those, but a closer look at the data suggests that the ranking
will be affected by dislocated elements. In Example (36), two different rankings are possible. The
first one (37) ranks modem according to its grammatical function, object. The alternative (38) is
to rank it higher than the pro subject we, because it is left-dislocated. The usual ranking produces
a RETAIN transition from (36a) to (36b), and a SMOOTH SHIFT from (36b) to (36c). The alternative
ranking, with modem higher, results in a CONTINUE followed by a RETAIN 20 .
(36) A: a. ¿módem?
‘Modem?’
B: b. módem, los tenemos
‘Modems, (we) have them’
c. pero no los instalamos todavía
‘but (we) haven’t installed them yet.’
(37) Grammatical ranking
a. Cf: MODEM
Cb: 0
b. Cf: WE, MODEMS
Cb: MODEMS – Transition: RETAIN
c. Cf: WE, MODEMS
19
Thanks to Laurie Fais for this point and for the example.
Transition preference for individual utterances is perhaps not enough of a reason to consider the alternative. Rule 2
is mostly about preference for sequences of certain transitions. Another complicating factor is that left-dislocation
may not signal salience: Givón’s (1983) topic accessibility scale ranks left-dislocated NPs as less accessible than
neutral-ordered NPs. It is not clear whether less accessible in Givón’s scale means more salient in Centering terms.
20
18
Cb: WE – Transition: SMOOTH SHIFT
(38) Alternative ranking
b. Cf: MODEMS, WE
Cb: MODEMS – Transition: CONTINUE
c. Cf: WE, MODEMS
Cb: MODEMS – Transition: RETAIN
4.9 Unresolved issues
There are a number of unresolved issues in the ranking of the Cf list. The first one is the use of
prosody in addition to the other factors that affect the ranking. A number of researchers have
pointed out that prosody and stress affect the order of elements in the list when dealing with
spoken language (Brennan, 1995; Cornish, 1999). This remains a task to be addressed in future
research. Another difficulty within Centering is the treatment of pronouns that refer to discourse
segments, or to abstract entities (Asher, 1993; Byron, 2002). I have excluded them from analysis
for the time being.
5. Which referring expression?
Anaphora resolution, and the form of the anaphoric term itself have long been linked to the
relative prominence of entities in the discourse (Gundel et al., 1993; Prince, 1981; Sidner, 1983).
Rule 1 of Centering Theory establishes that the Cb of an utterance must be a pronoun, if other
pronouns are present. That is, the Cb will be realized by the most reduced form (a pronoun) if
other pronouns are present. Centering does not suggest any other rules for what will happen in
other situations, i.e., when there are no pronouns at all. However, researchers have proposed a
relation between the transition type, i.e., the progression of local discourse topics, and either the
form of referring expressions used to realize the subject (Di Eugenio, 1998), or the Cb of an
utterance (Taboada, 2002a). The main purpose of this paper is to determine what relationship
there is between Centering transitions and referring expressions. For that purpose, I carried out a
corpus analysis of two types of spoken language corpora in Spanish. The corpora are the ISL
corpus and the CallHome corpus.
The ISL corpus is a large collection (a total of about 500 conversations) of task-oriented
conversations recorded in a lab, with externally-controlled turns. The participants, who were
native speakers of Spanish 21 , had to press the ‘Enter’ key on a keyboard to yield the turn, which
makes the conversations similar to one-way radio, although the speakers are present in the same
room. The task was to arrange for a two-hour meeting within a time period that ranged from two
to four weeks. The speakers had conflicting agendas, and usually proposed a number of dates
before an agreement was reached. Nine conversations from this corpus were analyzed, three each
of dyads of female-female, male-male and female-male speakers. The nine conversations
amounted to 262 utterances, as defined in Section 3.1, and a total of 2,798 words.
21
The speakers came from all corners of the Spanish-speaking world. For more details on the corpus, see Taboada
(2004).
19
The CallHome corpus is a collection of telephone conversations lasting up to 30 minutes
between native speakers of Spanish. One party was given a free long-distance call, free choice of
who to call, and no restriction on topics. Most participants called relatives or friends 22 . For this
study, five conversations were used, a total of 1,198 utterances and 8,694 words.
The conversations were first segmented according to the guidelines outlined in Section 3.1.
Then each utterance was coded according to Centering principles, including Cf list and type of
transition. Table 2 shows the number of non-zero transitions for both corpora.
Continue Retain Smooth Rough
shift
shift
Total
ISL
121
65%
27
16%
22
15%
11
4%
181
CallHome
515
67%
129
15%
116
12%
30
6%
790
Table 2. Centering transitions in two corpora.
The results are as predicted in Centering: CONTINUE transitions are preferred
(overwhelmingly) over other types of transitions, and RETAINS are preferred over SHIFTS. ROUGH
SHIFTS are relatively rare. It is interesting to see that the two corpora have similar percentages of
all types. Although the corpora are both spontaneous spoken conversations, they are somewhat
different, in that the ISL conversations are task-oriented, whereas the CallHome recordings are
casual. Those differences do not seem to affect the distribution of Centering transitions.
The numbers shown in Table 2 are for transitions that had a backward-looking center. A large
number of transitions had an empty Cb, and were not included in the analysis. The numbers are
presented in Table 3.
ISL
CallHome
Utterances
Cb=0
%
262
81
30.92%
1198
408
34.06%
Table 3. Utterances with empty backward-looking centers.
There exist a number of reasons for the high occurrence of utterances with an empty
backward-looking center. Some of those utterances do introduce completely new entities in the
discourse, thus beginning a new discourse segment: Centering operates at the local discourse
level; transitions between discourse segments are part of the global structure, and strictly not part
of a Centering analysis 23 . In a number of cases, however, the entities were inferable from the
22
Participants were also speakers of different dialects. Details about the transcriptions are available at:
http://www.ldc.upenn.edu/Catalog/docs/LDC96T17/ch_span.txt
23 Identifying discourse segments is not a trivial matter. My observations here about when discourse segments start
are impressionistic; rigorous analysis and annotation needs to be done to integrate Centering into the global structure
of the discourse.
20
context, but the inference seemed a bit far-fetched, and I decided not to establish it. That is the
case in (39b), where the speaker refers to the days she has mentioned in utterance (39a). The
utterance could read “check if you can meet on Tuesday the 16th after 12 noon”, but instead it is
“check if you can”. This is not just a question of a null object, but a null VP. I decided to not
include the date in the Cf list for (39b).
(39) a. así que recién podría el martes /eh/ dieciséis después de las doce del mediodía.
‘So I could on Tuesday, uh, the 16th after 12 noon.’
b. fijate si vos podés.
‘Check if you can.’
As I pointed out in Section 4.1, the issue here is what kind of inferables can be included in the
Cf list of an utterance. Hurewitz proposes to include entities that are in a functional dependency
with previously mentioned entities or that are subsets of other entities (Hurewitz, 1998), and also
discourse deictic pronouns, i.e., pronouns that refer to a part of the discourse, such as events or
clauses (Webber, 1981). In Hurewitz’s account, utterances joined by one of those relations
constitute a new type of transition, a PARTIAL SHIFT. Fais (2004) links entities in the discourse to
other previously mentioned entities using cohesive relations (Halliday and Hasan, 1976).
In some other cases, empty Cbs resulted from problems with the segmentation (Poesio et al.,
2000), or from the strict adjacency constraint in Centering: only entities in the previous utterance
can become the Cb of the current one. Some empty Cbs were as predicted by Centering, that is,
they initiated a new discourse segment; for instance, a new topic is being discussed, or a new date
is being proposed 24 , and therefore contained no link to the previous utterance. The new discourse
segments are often a completely new ‘push’ onto the focus stack (Grosz and Sidner, 1986), but
they can also be insertion or side sequences (Jefferson, 1972) or corrections (Schegloff et al.,
1977).
5.1. Referring expressions and transitions
The Cb of each utterance was coded according to whether it was one of the several possible
referring expressions, and those types of expressions were related to the transition types. The
referring expressions are illustrated in (40) to (46). The referring expression in question is in
bold.
Zero pronoun
(40) a. Conozco, en serio, un doctor que hizo su doctorado en Japón,
‘Seriously, I know a doctor who did his Ph.D. in Japan,’
b. acabó
y [pause]
finish:3SG.PRET and
‘(He) finished and’
c. no [-]
encontró
chamba,
24
See Taboada (2000; 2004, ch. 6) for a discussion of discourse segments in the ISL conversations. A new discourse
segment was always initiated when a new date is being proposed, as evidenced by a break in the chain of cohesive
links in the conversation.
21
not null find:3SG.PRET employment
‘didn’t find a job,’
Clitic
(41) a. Llega a Atenas
‘(She) will arrive in Athens’
b. y va a estar ahí tres semanas
‘and (she) is going to be there for three weeks’
c. y luego la
andan
paseando de
isla
en isla
and then CL.3SG.FEM.ACC go:3PL.PRES walk:GER from island to island
‘and then (they) are going to take her around from one island to the next.’
Pronoun
(42) a. No, no. Si de hambre no me muero.
‘No, no. (I)’m not going to starve.’ (lit., ‘die of hunger’)
b. Pero yo quiero
but I want:1SG.PRES
ser
astrofísica
be:INF astrophysicist
‘But I want to be an astrophysicist.’
Demonstrative pronoun
(43) B: a. Aquí le llaman tutorial.
‘Here they call it tutorial.’
A: b. Sí, pues ha
de ser
eso.
yes then have:3SG.PRES of be:INF that
‘Yes, then it must be that.’
Full noun phrase
(44) B: a. ¿Y tu hermana?
‘And your sister?’
A: b. Mi
My
hermana está
biesister
be:3SG.PRES we-
‘My sister is well.’
Other
(45) Wh- pronoun
A: a. Ay, pero no muchos días más.
‘Ah, but not many more days.’
B. b. Cuánto más.
‘How much more?’
22
(46) Adverbial (NP or PP) 25
A: a. no. el lunes en la mañana <no> no puedo.
‘No. Monday morning (I) can’t.’
b. tal vez el lunes en la tarde, después de las doce?
‘Maybe Monday in the afternoon, after twelve?’
B: c. bueno el lunes
tengo
una reunión
well the Monday have:1SG.PRES a meeting
de
<d> uno a cuatro
from tw- one to four
‘Well, on Monday (I) have a meeting from tw- one to four.’
Continue
Retain
Smooth shift
Rough shift
Zero pronoun
350 55.0%
44 28.2%
74 53.6%
9 21.9%
Clitic
114 17.9%
48 30.8%
24 17.4%
14 34.1%
Pronoun
53
8.3%
8
5.1%
9
6.5%
4
9.8%
Demonstr. pr.
15
2.4%
4
2.6%
4
2.9%
4
9.8%
Full NP
86 13.5%
26 16.7%
Other
18
26 16.7%
n
636
2.8%
156
22 15.9%
5
138
3.6%
9 21.9%
1
2.4%
41
Table 4. Referring expressions for the Cb of each utterance, according to transition.
Table 4 shows that, overall, the Cb tends to be expressed through a zero pronoun. This is the
least marked form available in Spanish. For that reason, it is to be expected that the Cb will be
coded as a zero pronoun when the transition is a CONTINUE. Such is the case: out of the 636
continue transitions (for both corpora together), 55% had a zero pronoun as Cb. When we move
onto RETAIN, where the Cb is continued from the previous utterance, but will likely not be
continued further, the percentage of zero pronouns decreases. However, it grows again in the
SMOOTH SHIFTS, to almost the same percentage as for CONTINUE (53.6%).
Di Eugenio (1990; 1998) found that in Italian 26 , speakers typically encode center continuation
with zero subjects, and center retention and shift with stressed pronouns. She also found that
instances of RETAIN and SHIFT with null pronoun subjects are possible if the utterance that
constitutes the change contains syntactic features that force the zero subject to refer to an entity
other than the Cb of the previous utterance. Indeed, I found many cases of null pronouns in
subject position that made the referent clear, when it was other than the Cbi-1. In Example (47c),
the number agreement on the verb links the null subject to the object in the previous utterance
(‘mountains’), not its subject and Cb, the Yosemite National Park that the speakers have been
discussing.
(47) a. Sí. Sí, es un parque nacional
25
Adverbials that are added to the Cf list are mostly those that denote times and places.
Di Eugenio analyzed excerpts from two novels, newspaper articles, short stories, and a bulletin board post. There
is a difference between Di Eugenio’s analysis and mine: she studied the realization of the subject; I examine the Cb.
26
23
‘Yes, yes, (it)’s a national park’
b. y es, tiene así montañas,
‘and (it)’s got like mountains,’
c. no, no son muy grandes,
‘(they) are not, not very big,’
Di Eugenio also found that speakers encode center retention or shift with a stressed subject
pronoun (presumably in the cases when syntactic factors do not exclude reference resolution to
the previous Cb). If we look at Table 4, we can see that pronouns are not used very often, across
all four transition types. They actually occur less often in RETAIN and SMOOTH SHIFT transitions
than in CONTINUE, and only increase within ROUGH SHIFT, to 9.8%, which are only four instances,
given the low number of ROUGH SHIFTS.
More numerous are full noun phrases (definite noun phrases or proper nouns), and for those
we can see a steady increase from CONTINUE to ROUGH SHIFT. It is possible that center change is
expressed more often in (spoken) Spanish via a full noun phrase. For instance, in (48), the
conversation has been about B’s activities, and she is then the Cb in (48a). When B takes her
turn, she shifts and talks about Cristina, previously introduced. She could have used a stressed
personal pronoun (ella), especially given that there is no competing referent, but instead chose to
repeat the proper name.
(48) A: a. Mary, tú fuiste por tu vestido rojo donde Cristina.
‘Mary, did you go get your red dress from Cristina’s?’
B: b. Mmm. Ay, sí, pero Cristina está en Bogotá
‘Mmm. Oh, yes, but Cristina is in Bogotá.’
Clitics are, after null pronouns, the preferred form of realization across transition types. They
are used in CONTINUE to refer to the speaker quite often, with psychological verbs (49), or other
verbs, as indirect objects (50).
(49) me
parece
CL.1SG seem:3SG.PRES
para la otra
for the other
lo mejor dejar-lo
the best leave:INF-CL.3SG.MASC.ACC
semana,
week
‘(It) seems better to me to leave it for next week,’
(50) para que me
lo
so that CL.1SG CL.3SG.MASC.ACC
arregle
fix:3SG.PRES.SUBJ
ahora para diciembre.
now for December
‘So that (she) can fix it for me now for December.’
Clitics do not always refer to the speaker. They can refer to the interlocutor (51) or to a third
party, as in (52), with a pronominal verb, se vino (‘came’) 27 .
27
The word se in this example is a clitic, co-referential with the subject, and different from the se in Example (25).
This se is in a paradigm with other clitics: me for first person singular subject; te for second person singular subject,
etc. These constructions are referred to as pseudo-reflexive or middle voice constructions (Mendikoetxea, 1999).
They appear to be reflexive, but are used with intransitive verbs, some of which have both an intransitive and a
pseudo-reflexive use (hence the term ‘pronominal verbs’ when used pseudo-reflexively). See also Sharp (2005) for a
24
(51) correcto Mónica, te
decía
el viernes
correct Mónica CL.2SG.ACC tell:1SG.PRET the Friday
por aquello de la muerte de Gaitán,
for that
of the death of Gaitán
‘Right Mónica, (I) was telling you Friday because of Gaitán’s death,’
(52) Sí, se
vino
para
acá estar
yes CL.3SG come:3SG.PRET towards here be:INF
conmigo.
with.me
Yes, (she) came to be here with me.’
Demonstrative pronouns are not very frequent in general: there are only four instances each
for RETAIN and both SHIFTS. They are slightly more common in CONTINUE transitions, but still
only account for 2.4% of the Cbs in those. They are used to refer to both things (53) and people
(54). In some cases, they also refer to abstract entities, as in (55). Gundel et al. (1993) found that
demonstrative pronouns are rarely used for referents that are familiar or in focus, which would be
the case with most Cbs in this study, and Ariel (1988) also found a very low level of
demonstratives in her corpus analysis.
(53) B: a. lo que sí son buenos, y no los sé usar, son los los enlaces para estar así platicando
‘What’s good, and (I) don’t know how to use them, are the the links to be like chatting.’
A: b. Ahá
‘Uh-huh.’
B: c. esos son buenos
‘Those are good.’
(54) B: a. No, eh, ay, mami, el viernes se viene Alicia, la que tú tenías, para acá.
‘No, uh, uh, mami, on Friday comes Alicia, the one you used to have, here.’
B: b. Vamos a ver qué tal me resulta.
‘(We)’ll see how she turns out.’
A: c. Esa es una fiera.
‘That (one) is amazing.’ (lit. ‘She’s an animal.’)
(55) A: a. Aquí aprovechando las llamaditas estas que nos dan, ah
‘Here, taking advantage of these calls (they) give us, uh’
B: b. Ay, sí, claro, pero eso cómo es, Chipi,
‘Ah, yes, right, but how’s that, Chipi,’
c. eso cómo funciona.
‘How does that work?’
Finally, the category “Other” includes a number of other realizations: wh-pronouns, adverbial
NPs, and possessive determiners and pronouns. Some of these appear frequently in RETAIN
transitions, in preparation for a change of topic. In (56), for instance, speaker A is expressing
unified account of all instances of se in Spanish.
25
despair, and ends his turn with a rhetorical question that includes reference to himself in a null
pronoun. Speaker B continues talking about speaker A, but uses a possessive determiner (tu
‘your’), trying to steer the conversation towards exactly what is the problem (resentimiento
‘resentment’).
(56) A: … qué voy a hacer?
‘What am (I) going to do?’
B: ya, hay mucho resentimiento en tu voz, no?
‘I see, there’s a lot of resentment in your voice, isn’t there?’
In summary, a CONTINUE transition generally realizes the Cb as a zero pronoun, followed by a
clitic. RETAIN transitions are also realized through zero or a clitic, although other possibilities
exist. These realizations are as expected, and reflect the types of situations that the different
transitions were meant to encode. The next section deals with realizations that appear to be
contrary to expectation.
5.2. Realization against expectation
The descriptions in the previous section are all of the type ‘x transition tends to encode the Cb in
y form’. We have seen there are some clear tendencies. My concern here is the realizations that
do not follow those tendencies.
The most clearly stated tendency, in this paper and in the literature, is that the Cb of a
CONTINUE transition is realized via a reduced expression: zero pronoun, clitic, unaccented
pronoun, etc. Other realizations are said to make processing more difficult. For instance, Gordon
et al. (1993: 341) establish that there exists a ‘repeated name penalty’, where repeating a name
that continues to be the Cb in the discourse deprives the reader 28 of an important cue that the
current utterance is coherent with the previous one. And yet, 13.5% of the Cbs in CONTINUE
transitions are realized as full noun phrases, many of them proper names 29 .
The explanations for repeated noun phrases all have to do with spoken language phenomena.
For instance, in the CallHome corpus, speakers frequently ask about other friends or relatives.
These exchanges typically involve one speaker mentioning the name of the person, and the other
repeating the name, as in (57). (Proper names are included in the full NP category).
(57) A: Qué han sabido de Eddie.
‘What have you heard from Eddie?’
B: De Eddie nada,
‘From Eddie, nothing,’
This is quite frequent when the turn changes, but it also happens within a speaker’s turn. In
(58), the speaker repeats Mónica, although the referent should be clear, and the clitic la would
have sufficed.
28
Gordon et al.’s (1993) experiments were written. My explanations for the lack of ‘repeated name penalty’ are all
related to the fact that the data analyzed here is spoken.
29 Di Eugenio (1998) found a few instances of strong pronouns in subject position with CONTINUE transitions. She
relates it to the transition type preceding the CONTINUE, a possibility I have not yet explored in my data.
26
(58) a. y Mónica sin embargo ha crecido un montón.
‘And Mónica, however, has grown a lot.’
b. Tu papá se
your Dad CL.3SG
asombra
de ver-la
surprise:3SG.PRES of see:INF-CL.3SG.FEM.ACC
a Mónica,
to Mónica
‘Your Dad is surprised to see her, Mónica,’
Brennan (1995) found that referents introduced in object position were then re-introduced in
subject position with a full noun phrase. Only after that were they referred with a pronoun.
Brennan believes that the referent needs to be in subject position so that it can become a
backward-looking center, and thus candidate for pronominalization. This is the case in some of
the examples, as in Example (57), where the repeated NP/proper name becomes the backwardlooking center of the utterance. I also found in the corpus instances of entities in subject position,
but left-dislocated (Y Juan, ¿cómo está? ‘And Juan, how’s he?’). The proper name is repeated in
subject position before it is pronominalized. It is possible that a neutral subject position is
necessary before pronominalization takes place.
In general, proper name repetition might be a device to establish common ground between the
interlocutors. Downing (1996) points out that proper names are used very often in conversation:
to introduce individuals in the conversation, as the most easily identifiable form of reference; and
to refer again to those individuals, as a marker of true familiarity with the referent denoted by the
proper noun.
In the ISL corpus, repeated referents across turns are either the participants or the dates being
discussed. In (59), speaker B refers to herself with a full pronoun at the beginning of her turn.
Amaral and Schwenter (2005) discuss cases like (59), and propose that the pronoun is obligatory,
because it establishes a contrast 30 .
(59) A: puedes reunirte conmigo en mayo?
‘can you meet with me in May?
B: a ver yo estoy de viaje del
treinta y uno hasta...
to see I
am of travel from.the thirty and one until…
‘let’s see, I am away from the 31st until...’
In (60), speaker B uses a full NP, el jueves to refer to the date being discussed, present in the
immediately preceding utterance as a null pronoun. Note that in this case, contrast does not play a
role.
(60) A: a. creo que el jueves veintisiete, que lo tengo totalmente libre podría ser.
‘I think Thursday the 27th, which (I) have completely free, it could be.’
b. qué te parece?
‘What do you think (of that date)?’
30 Dimitriadis (1996) proposes that a pronoun is chosen when the antecedent is not the Cp of the previous sentence
(i.e., it is not the most salient entity in the previous sentence). It is possible that that is the case in many situations,
but not in Example (59), where tú (‘you’), the null pronoun from the first utterance is realized as a strong pronoun
(yo) in the second utterance, of course with the change in person due to the change of speaker. Contrast and the
change of turn seem to be the decisive factors here.
27
B: c. bueno. el jueves realmente es un día ocupado para mí.
‘Well, Thursday is actually a busy day for me.’
The presence or absence of the personal pronoun subject in Spanish has received a great deal
of attention (e.g., Alonso-Ovalle et al., 2002; Cameron, 1992; Davidson, 1996; Enríquez, 1984).
Stewart (1999) proposes that the use of the first person singular pronoun is a politeness resource,
which helps contrast the speaker with other individuals or groups. Luján (1999) also points out
the contrastive character of first and second person pronouns. This seems to be the case in the
ISL corpus, where the speaker’s agenda is contrasted with the interlocutor’s. Davidson (1996)
finds that the personal pronoun is used for emphasis and to negotiate conversational turns (to
claim the floor for an extended period of time). He also found that the first person pronoun was
used more frequently than second or third person pronouns. Those three factors might account for
the presence of yo in examples such as (59).
Conversely, using a zero pronoun when something else is expected could result in more
difficulty in processing. In Example (61), speaker A has been talking about visiting his sister in
Greece. The Cb at the end of A’s turn is sister. B then replies with a question, ‘isn’t (that) very
expensive?’. There is no repeated entity across the turns, but B uses a zero for the third person
singular subject of (61e). One possible referent is the idea of sightseeing, which A used at the end
of his turn. It is possible that B realizes this possible mistaken interpretation, and reformulates in
(61f), to specify that she is referring to the cost of the flight, not of doing tourism 31 .
(61) A: a. Sí. Sí pues, es que se va a ir a Grecia
‘Yes. Yes, so she’s going to Greece’
b. y luego se queda las tres últimas semanas
‘and then she’s staying the last three weeks’
c. tres, tres semanas más, se queda
‘three, three more weeks, she’s staying’
d. y no más se
pasea
and not more CL.3SG walk:3SG.PRES
‘and she’s just going to do tourism.’ (lit. ‘she’s just going to walk around’)
B: e. Pues, ¿no te
but not CL.2SG
sale
carísimo?
come.out:3SG.PRES very.expensive
‘Well, isn’t (that) very expensive for you?’
f. o sea el avión, yo digo.
‘I mean the plane, I mean.’
31
Geluykens (1994) attributes this type of repair to a conflict between principles of Clarity and Economy, derived
from Grice’s (1975) maxims.
28
6. Conclusions
I have presented an application of Centering theory to two corpora of spoken Spanish. The study
contributes to an understanding of the relationship between Centering transitions and choice of
referring expression. The analysis shows that, when the topic stays constant, i.e., when a
CONTINUE transition is present, the most common realization of the backward-looking center is in
a null pronoun. Null pronouns are also used in the other three transition types, likely because they
are clearly identifiable from context, through person or number marking.
Full noun phrases and pronouns are used quite often to encode the backward-looking center.
This is contrary to the expectation that the topic of the utterance is encoded with the minimum
amount of information. According to Gordon and colleagues (Gordon et al., 1993), there is a
‘repeated name penalty’ when using a more informative referring expression than necessary. It
was found that speakers tend to repeat pronouns referring to themselves, and proper names
referring to third persons. This occurs most often when there is a change of turn, but also within
the turn. In spontaneous conversation, Downing (1996) found that proper names are often used,
even when pronouns would ensure correct identification of referent, to establish the referent in
the discourse as common to both speakers.
It is possible, then, that we may be forced to revise Centering predictions as to center
realization, to take into account spoken language phenomena. Another source of evidence in
support of this view is that a number of empty backward-looking centers were attributed to the
presence of side or insertion sequences, which are characteristic of conversation. This ties in with
the relationship of Centering and the global structure of the discourse. Centering was designed as
a model for the local focus of attention. It is not clear how Centering can relate the two levels.
For instance, in (62), there is a global story about how speaker B’s boss was quite proud of his
work in a particular situation, because speaker B and his boss had issued 2,700 notices for back
taxes on cars. Speaker B then starts a small story about how his boss came to know that he had
done well in comparison to others. The story covers utterances (62c) to (62j). In utterance (62k),
speaker B refers again to his boss, which is part of the global focus of discourse, as part of the
‘we’ in (62a). However, in (62k) the subject el tipo ‘the guy’ cannot be linked to the immediately
preceding utterance, and thus results in an empty backward-looking center.
(62) B: a. en en en cinco días - hicimos dos mil setecientas citaciones
‘in in five days, (we) did two thousand seven hundred notices’
A: b. ahá
‘uh huh’
B: c. cuando él fue a la reunión de los abogados
‘when he went to the meeting with the lawyers’
d. todos habían hecho cien, ciento veinte
‘(they) had all done a hundred, a hundred and twenty’
e. no lo podían creer, viste
‘(they) couldn’t believe it, you know’
A: f. mirá, vos
‘really?’
29
B: g. y, pero así también fue la gente que empezó a caer
‘and, also that’s how people realized’
h. imagínate, la mitad de la gente, toda caliente
‘imagine, half the people, all mad’
i. porque le pedían impuestos que ya se, autos de hace treinta años que se transfirieron
‘because (they) were being asked for taxes for cars that already, cars that had been transferred
thirty years ago’
j. [pause] que no existen más, viste, ¡una goma! [A: { laugh } ] - tremenda,
‘that don’t exist any more, you see, what a situation!’
k. entonces viste, el tipo vino calentón, así
‘then, you see, the guy came back all excited, you know.’
Future research will be focused on the relationship between Centering and the discourse
structure of the conversations, paying attention to conversational phenomena such as side
sequences and turn-taking. I will also study the relationship between the local focus of attention
(which Centering was devised to model) and the global structure of the conversations.
References
Alonso-Ovalle, Luis, Susana Fernández-Solera, Lyn Frazier and Charles Clifton. (2002). Null vs. overt
pronouns and the topic-focus articulation in Spanish. Rivista di Linguistica (Italian Journal of
Linguistics), 14 (2), 151-170.
Amaral, Patrícia Matos and Scott A. Schwenter. (2005). Contrast and the (non-) occurrence of subject
pronouns. In D. Eddington (Ed.), Selected Proceedings of the 7th Hispanic Linguistics Symposium (pp.
116-127). Somerville, MA: Cascadilla Press.
Ariel, Mira. (1988). Referring and accessibility. Journal of Linguistics, 24, 65-87.
Ariel, Mira. (1996). Referring expressions and the +/- coreference distinction. In T. Fretheim and J. K.
Gundel (Eds.), Reference and Referent Accessibility (pp. 13-25). Amsterdam and Philadelphia: John
Benjamins.
Asher, Nicholas. (1993). Reference to Abstract Objects in Discourse. Dordrecht: Kluwer.
Bolinger, Dwight L. (1979). Pronouns in discourse. In T. Givón (Ed.), Syntax and Semantics, Vol. 12:
Discourse and Syntax (pp. 289-309). New York: Academic Press.
Brennan, Susan E. (1995). Centering attention in discourse. Language and Cognitive Processes, 10 (2),
137-167.
Brown, Gillian and George Yule. (1983). Discourse Analysis. Cambridge: Cambridge University Press.
Butt, John and Carmen Benjamin. (2000). A New Reference Grammar of Modern Spanish. Chicago:
McGraw-Hill. (3rd edition).
Byron, Donna K. (2002). Resolving pronominal reference to abstract entities, Proceedings of the Annual
Meeting of the Association for Computational Linguistics (ACL-02) (pp. 80-87). Philadelphia, PA.
30
Byron, Donna K. and Amanda Stent. (1998). A preliminary model of Centering in dialog, Proceedings of
the 36th Annual Meeting of the Association for Computational Linguistics (ACL-98) (pp. 1475-1477).
Montréal, Canada.
Cameron, Richard. (1992). Pronominal and Null Subject Variation in Spanish: Constraints, Dialects, and
Functional Compensation. Unpublished Ph.D. dissertation, University of Pennsylvania.
Carreiras, Manuel, Morton Ann Gernsbacher and Victor Villa. (1995). The advantage of first mention in
Spanish. Psychonomic Bulletin and Review, 2 (1), 124-129.
Clark, Herbert H. (1977). Bridging. In P. N. Johnson-Laird and P. C. Wason (Eds.), Thinking: Readings in
Cognitive Science (pp. 411-420). Cambridge: Cambridge University Press.
Cornish, Francis. (1999). Anaphora, Discourse, and Understanding: Evidence from English and French.
Oxford: Clarendon.
Cornish, Francis. (2005). Degrees of indirectness: Two types of implicit referents and their retrieval via
unaccented pronouns. In A. Branco, T. McEnery and R. Mitkov (Eds.), Anaphora Processing:
Linguistic, Cognitive and Computational Modelling (pp. 199-220). Amsterdam and Philadelphia: John
Benjamins.
Cote, Sharon. (1998). Ranking forward-looking centers. In M. A. Walker, A. K. Joshi and E. F. Prince
(Eds.), Centering Theory in Discourse (pp. 55-69). Oxford: Clarendon.
Davidson, Brad. (1996). 'Pragmatic weight' and Spanish subject pronouns: The pragmatic and discourse
uses of 'tú' and 'yo' in spoken Madrid Spanish. Journal of Pragmatics, 26 (4), 543-565.
Di Eugenio, Barbara. (1990). Centering theory and the Italian pronominal system, Proceedings of the 13th
International Conference on Computational Linguistics (COLING-90) (pp. 270-275). Helsinki,
Finland.
Di Eugenio, Barbara. (1998). Centering in Italian. In M. A. Walker, A. K. Joshi and E. F. Prince (Eds.),
Centering Theory in Discourse (pp. 115-137). Oxford: Clarendon.
Dimitriadis, Alexis. (1996). When pro-drop languages don't: Overt pronominal subjects and pragmatic
inference. In L. M. Dobrin, K. Singer and L. McNair (Eds.), CLS 32: Papers from the Main Session
(pp. 33-47). Chicago: Chicago Linguistics Society.
Downing, Pamela A. (1996). Proper names as a referential option in English conversation. In B. A. Fox
(Ed.), Studies in Anaphora (pp. 95-143). Amsterdam and Philadelphia: John Benjamins.
Eckert, Miriam and Michael Strube. (1999). Resolving discourse deictic anaphora in dialogues,
Proceedings of the 9th Conference of the European Chapter of the Association for Computational
Linguistics (EACL-99) (pp. 37-44). Bergen, Norway.
Edelsky, Carole. (1981). Who's got the floor? Language in Society, 10, 383-421.
Enríquez, Emilia. (1984). El pronombre personal sujeto en la lengua española hablada en Madrid.
Madrid: Consejo Superior de Investigaciones Científicas.
Fais, Laurel. (2004). Inferable centers, Centering transitions and the notion of coherence. Computational
Linguistics, 30 (2), 119-150.
Fernández Soriano, Olga. (1999). El pronombre personal: Formas y distribuciones. Pronombres átonos y
tónicos. In I. Bosque and V. Demonte (Eds.), Gramática descriptiva de la lengua española (Vol. 1:
Sintaxis básica de las clases de palabras, pp. 1209-1273). Madrid: Espasa.
Geluykens, Ronald. (1994). The Pragmatics of Discourse Anaphora in English: Evidence from
Conversational Repair. Berlin: Mouton de Gruyter.
31
Gernsbacher, Morton Ann and David J. Hargreaves. (1988). Accessing sentence participants: The
advantage of first mention. Journal of Memory and Language, 27 (6), 699-717.
Givón, Talmy. (1983). Topic continuity in discourse: An introduction. In T. Givón (Ed.), Topic Continuity
in Discourse: A Quantitative Cross-Language Study (pp. 1-41). Amsterdam and Philadelphia: John
Benjamins.
GNOME. (2000). GNOME Project Final Report. Edinburgh: University of Edinburgh.
Gordon, Peter C., Barbara J. Grosz and Laura A. Gilliom. (1993). Pronouns, names, and the Centering of
attention in discourse. Cognitive Science, 17 (3), 311-347.
Gordon, Peter C., Randall Hendrick, Kerry Ledoux and Chin Lung Yang. (1999). Processing of reference
and the structure of language: An analysis of complex noun phrases. Language and Cognitive
Processes, 14 (4), 353-379.
Grice, H. Paul. (1975). Logic and conversation. In P. Cole and J. L. Morgan (Eds.), Speech Acts. Syntax
and Semantics, Volume 3 (pp. 41-58). New York: Academic Press.
Grosz, Barbara J., Aravind K. Joshi and Scott Weinstein. (1995). Centering: A framework for modelling
the local coherence of discourse. Computational Linguistics, 21 (2), 203-225.
Grosz, Barbara J. and Candace L. Sidner. (1986). Attention, intentions, and the structure of discourse.
Computational Linguistics, 12 (3), 175-204.
Gundel, Jeanette K., Nancy Hedberg and Ron Zacharski. (1993). Cognitive status and the form of
referring expressions in discourse. Language, 69, 274-307.
Hadic Zabala, Loreley and Maite Taboada. (2004). Centering Theory in Spanish: Coding Manual.
Unpublished manuscript, Simon Fraser University.
Halliday, Michael A. K. (1967). Notes in Transitivity and Theme in English. Part II. Journal of
Linguistics, 3, 199-244.
Halliday, Michael A. K. and Ruqaiya Hasan. (1976). Cohesion in English. London: Longman.
Halliday, Michael A. K. and Christian M.I.M. Matthiessen. (2004). An Introdution to Functional
Grammar (3rd ed.). London: Arnold.
Heap, David. (1998). Optimalizing Iberian clitic sequences. In J. Lema and E. Treviño (Eds.), Theoretical
Analyses on Romance Languages (pp. 227-248). Amsterdam and Philadelphia: John Benjamins.
Hudson-D'Zmura, Susan and Michael K. Tanenhaus. (1998). Assigning antecedents to ambiguous
pronouns: The role of the center of attention as the default assignment. In M. A. Walker, A. K. Joshi
and E. F. Prince (Eds.), Centering Theory in Discourse (pp. 199-226). Oxford: Clarendon.
Hurewitz, Felicia. (1998). A quantitative look at discourse coherence. In M. A. Walker, A. K. Joshi and E.
F. Prince (Eds.), Centering Theory in Discourse (pp. 273-291). Oxford: Clarendon.
Jackendoff, Ray. (1990). Semantic Structure. Cambridge, Mass: MIT Press.
Jaeggli, Osvaldo A. (1986). Arbitrary plural pronominals. Natural Language and Linguistic Theory, 4, 4376.
Jefferson, Gail. (1972). Side sequences. In D. Sudnow (Ed.), Studies in Social Interaction (pp. 294-338).
New York: Free Press.
Kameyama, Megumi. (1986). A property-sharing constraint in Centering, Proceedings of the 24th Annual
Meeting of Association for Computational Linguistics (ACL-86) (pp. 200-206). New York, USA.
32
Kameyama, Megumi. (1998). Intrasentential Centering: A case study. In M. A. Walker, A. K. Joshi and E.
F. Prince (Eds.), Centering Theory in Discourse (pp. 89-112). Oxford: Clarendon.
Kuno, Susumu. (1987). Functional Syntax: Anaphora, Discourse and Empathy. Chicago: University of
Chicago Press.
Lambrecht, Knud. (1994). Information Structure and Sentence Form: Topic, Focus, and the Mental
Representation of Discourse Referents. Cambridge: Cambridge University Press.
Levin, Lori, Klaus Ries, Ann Thyme-Gobbel and Alon Lavie. (1999). Tagging of speech acts and dialogue
games in Spanish Call Home, Proceedings of ACL-99 Workshop on Discourse Tagging. College Park,
MD.
Luján, Marta. (1999). Expresión y omisión del pronombre personal. In I. Bosque and V. Demonte (Eds.),
Gramática descriptiva de la lengua española (Vol. 1: Sintaxis básica de las clases de palabras, pp.
1275-1315). Madrid: Espasa.
Mendikoetxea, Amaya. (1999). Construcciones con se: medias, pasivas e impersonales. In I. Bosque and
V. Demonte (Eds.), Gramática descriptiva de la lengua española (Vol. 2: Las construcciones
sintácticas fundamentales; Relaciones temporales, aspectuales y modales, pp. 1631-1722). Madrid:
Espasa.
Miltsakaki, Eleni and Karen Kukich. (2004). Evaluation of text coherence for electronic essay scoring
systems. Natural Language Engineering, 10 (1), 25-55.
Pesetsky, David. (1987). Wh-in-situ: Movement and unselective binding. In E. Reuland and A. G. B. ter
Meulen (Eds.), The Representation of (In)definiteness (pp. 98-129). Cambridge, MA: MIT Press.
Pesetsky, David. (2000). Phrasal Movement and Its Kin. Cambridge, MA: MIT Press.
Poesio, Massimo, Hua Cheng, Renate Henschel, Janet Hitzeman, Rodger Kibble and Rosemary
Stevenson. (2000). Specifying the parameters of Centering Theory: A corpus-based evaluation using
text from application-oriented domains, Proceedings of the 38th Annual Meeting of the Association for
Computational Linguistics (ACL-2000) (pp. 400-407). Hong Kong.
Poesio, Massimo, Rosemary Stevenson, Barbara Di Eugenio and Janet Hitzeman. (2004). Centering: A
parametric theory and its instantiations. Computational Linguistics, 30 (3), 309-363.
Prince, Ellen F. (1981). Towards a taxonomy of given-new information. In P. Cole (Ed.), Radical
Pragmatics (pp. 223-255). New York: Academic Press.
Roberts, Craige. (1998). The place of Centering in a general theory of anaphora resolution. In M. A.
Walker, A. K. Joshi and E. F. Prince (Eds.), Centering Theory in Discourse (pp. 359-399). Oxford:
Clarendon.
Schegloff, Emmanuel, Gail Jefferson and Harvey Sacks. (1977). The preference for self-correction in the
organization of repair in conversation. Language, 53, 361-382.
Schiffrin, Deborah. (1994). Approaches to Discourse. Malden, Mass: Blackwell.
Sharp, Randy. (2005). A unified treatment of Spanish se. In A. Branco, T. McEnery and R. Mitkov (Eds.),
Anaphora Processing: Linguistic, Cognitive and Computational Modelling (pp. 113-136). Amsterdam
and Philadelphia: John Benjamins.
Sidner, Candace L. (1983). Focusing in the comprehension of definite anaphora. In M. Brady and R. C.
Berwick (Eds.), Computational Models of Discourse (pp. 267-330). Cambridge, Mass: MIT Press.
Stevenson, Rosemary, Rosalind A. Crawley and David Kleinman. (1994). Thematic roles, focus and the
representation of actions. Language and Cognitive Processes, 9, 519-548.
33
Stewart, Miranda. (1999). Hedging your bets: The use of yo in face-to-face interaction. Web Journal of
Modern Language Linguistics, 4-5, http://wjmll.ncl.ac.uk/issue04-05/stewart.htm.
Strube, Michael and Udo Hahn. (1999). Functional Centering: Grounding referential coherence in
information structure. Computational Linguistics, 25 (3), 309-344.
Suñer, Margarita. (1988). The role of agreement in clitic-doubled constructions. Natural Language and
Linguistic Theory, 6, 391-434.
Taboada, Maite. (2000). Cohesion as a measure in generic analysis. In A. Melby and A. Lommel (Eds.),
The 26th LACUS Forum (pp. 35-49). Chapel Hill, N.C.: The Linguistic Association of Canada and the
United States.
Taboada, Maite. (2002a). Centering and pronominal reference: In dialogue, in Spanish, Proceedings 6th
Workshop on the Semantics and Pragmatics of Dialog, EDILOG (pp. 177-184).
Taboada, Maite. (2002b). Foco y pronominalización en la lengua hablada: Una primera aproximación.
Documentos de español actual, 3-4, 173-200.
Taboada, Maite. (2004). Building Coherence and Cohesion: Task-Oriented Dialogue in English and
Spanish. Amsterdam and Philadelphia: John Benjamins.
Taboada, Maite and Loreley Hadic Zabala. (2005). What are the units of discourse structure? Segmenting
discourse within Centering Theory. Unpublished manuscript, Simon Fraser University.
Tetreault, Joel R. (2001). A corpus-based evaluation of Centering and pronoun resolution. Computational
Linguistics, 27 (4), 507-520.
Turan, Ümit Deniz. (1995). Null vs. Overt Subjects in Turkish Discourse: A Centering Analysis.
Unpublished Ph.D. dissertation, University of Pennsylvania, Philadelphia.
Walker, Marilyn A. (1998). Centering, anaphora resolution, and discourse structure. In M. A. Walker, A.
K. Joshi and E. F. Prince (Eds.), Centering Theory in Discourse (pp. 401-435). Oxford: Clarendon.
Walker, Marilyn A., Masayo Iida and Sharon Cote. (1994). Japanese discourse and the process of
Centering. Computational Linguistics, 20 (2), 193-232.
Walker, Marilyn A., Aravind K. Joshi and Ellen F. Prince. (1998). Centering in naturally occurring
discourse: An overview. In M. A. Walker, A. K. Joshi and E. F. Prince (Eds.), Centering Theory in
Discourse (pp. 1-28). Oxford: Clarendon.
Walker, Marilyn A. and Ellen F. Prince. (1996). A bilateral approach to givenness: A hearer-status
algorithm and a Centering algorithm. In J. K. Gundel and T. Fretheim (Eds.), Reference and Referent
Accessibility (pp. 291-306).
Wanner, Dieter. (1994). El orden de los clíticos agrupados en castellano. Thesaurus, 49 (1), 1-57.
Webber, Bonnie Lynn. (1981). Structure and ostension in the interpretation of discourse deixis. Language
and Cognitive Processes, 6 (2), 107-135.
Yngve, Victor H. (1970). On getting a word in edgewise. In Papers from the Sixth Regional Meeting of
the Chicago Linguistics Society (pp. 567-577). Chicago: University of Chicago.
Zaenen, Annie, Jean Carletta, Gregory Garretson, Joan Bresnan, Andrew Koontz-Garboden, Tatiana
Nikitina, M. Catherine O'Connor and Tom Wasow. (2004). Animacy encoding in English: Why and
how. In B. Webber and D. K. Byron (Eds.), Proceedings of ACL-2004 Workshop on Discourse
Annotation (pp. 118-125). Barcelona, Spain.
34
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement