Professional Documents
Culture Documents
North-Holland
443
1. Introduction
The proposals initially offered in Ariel (1985) were designed to account for the
process of accessing mental representations used in interpreting utterances.
Broadly speaking, the question posed by Accessibility theory (see Ariel 1985,
1988, 1990) is how context is brought into use while processing sentences. The
proposal, as first formulated, attempts to account for all the grammatical
categories used when marking the need to access context, i.e., NPs, VPs, and
Ss. I have since concentrated specifically on the accessing of NP antecedents.
Assuming that mental representations are not equally accessible to addressees at any given stage of the discourse, I have suggested (Ariel 1985, 1988,
* Author's address: M. Ariel, Department of L;aguistics, TeI-Aviv University, Ramat-Aviv, TelAviv 69978, Israel.
0378-2166/91/$03.50 199! -- Elsevier Science Publishers B.V. (North-Holland)
444
M. Ariel / Accessibility
2. Discourse-level references
M. Ariel / Accessibility
445
446
M. Ariel / Accessibility
Table I
Names and definite descriptions in first vs. subsequent mentions.
Expression
Definite descriptions
Full names
Context
First mention
Subsequent mentions
180 = 33.8%
118 = 63.8%
353 = 66.2%
67 = 36.2%
A similar specialization is revealed when we compare long definite descriptions (more than two content words) versus short definite descriptions (one to
two content words), as can be seen in table 2.1
Table 2
Long and short definite descriptions in first vs. subsequent mentions.
Expression
Context
First mention
Subsequent mentions
84 = 21.8%
96 = 65.3%
302 = 78.2%
51 = 34.7%
Last names
First names
135 = 85.2%
2 3 = 14.8%
~=0%
M. Ariel / Accessibility
447
Table 4
Distribution of anaphoric expressions in text (popular contexts in bold face).
Expression
Pronoun
Demonstrative
Definite description
Context
Same S
Previous S
Same para.
Across para.
110=20.8%
4 = 4.8%
4 = 2.8%
320=60.5%
50 -- 59.5%
20 = 14.1%
75= 14.2%
17-- 20.2%
65 = 45.8%
24= 4.5%
13 = 15.5%
53 = 37.3%
The conclusions regarding the popular contexts (80%) of each expression are
that pronouns mostly occur within the closest distance (same or previous S),
demonstratives are concentrated in the intermediate distance (previous S,
same paragraph), and definite descriptions occur at the furthest contexts (same
and across paragraph boundary). 3 Under the proposal at hand, they will be
classified as low, intermediate and high Accessibility markers respectively.
I have suggested that unity (how related the unit in which the antecedent
occurs is to the unit the anaphor is in) is also relevant when current
Accessibility is determined. Findings supporting the unity criterion discuss the
crucial role of same versus different points of view, worlds or frames.
Anderson et al.'s (1983) experiments are characteristic in this respect. They
found that referring to a frame-dependent entity (an entertainer in a children's
birthday party) is easier when the discourse is still embedded in the same
frame. Thus, when they compared reading times of the same anaphoric
expression referfng to the same antecedent when ~till withi~ ;.he birthday
party scene and outside it (five hours later), reading time of the now frameexternal antecedent was slower. Distance was, of course, kept constant.
Another result of the unity criterion is the distribution of pronouns versus
zeros in Chinese discourse, as discussed by Li and Thompson (1979). They
refer to degree of unity as degree of conjoinability, and what they found is
that pronouns, rather than zeros tend to occur in relatively independent
sentences (in the pragmatic sense). Zeros occur in the complementary context.
The following is one of their examples:
(la)
(Ib)
(Ic)
(id)
3 Two of the texts examined were fictional discourse: the opening section of a novel (E.H.
Young, The curate~ wife. London: Virago Press, 1934/1985, 7-13), and a short story (G. Paley,
"The pale pink roast', in Little disturbances qfman. London: Virago Press, 1956/1980, 43-52). For
non-fiction I chose 'Trouble on the set: An analysis of female characters on 1985 television
programs'. National Commission on Working Women: 1-12, and J.E. Tucker, Women in
Nineteenth Century Egypt. Cambridge: Cambridge University Press, 1985, I--6.
448
M. Ariel / Accessibility
M. Ariel / Accessibility
449
support of the Accessiblity theory has been quite brief. It pertains to names
(ar.d name types), definite descriptions (contrasting long versus short ones as
well), demonstratives, pronouns and zeros. In each case, a comparison of the
context where there is some preference for one expression over another shows
them to be different in the degree of Accessibility they signal. The reader is
referred to Ariel (1988, 1990) for further support for the role of Accessibility
theory in dictating appropriate choices among referring expressions in discourse. There, based on additional distributional findings drawn from a
variety of unrelated languages, I have established the following hierarchy of
Accessibility markers:
(3) LOW ACCESSIBILITY
Full name + Modifier
Full name
Long definite description
Short definite description
Last name
First name
Distal demonstrative (+ Modifier)
Proximal demonstrative ( + Modifier)
Stressed pronouns + Gesture
Stressed pronouns
Unstressed pronouns
Zeros
HIGH ACCESSIBILITY
450
M. Ariel / Accessibility
linguist).
The consequences of applying the rigidity criterion are the distinctions listed
in (6):
M. Ariel / Accessibility
451
4. Sentence-level anaphora
Up till now, we have limited our discussion to discourse anaphora. However,
if ,Accessibility is a cognitive concept guiding choices among referring and
anaphoric expressions in discourse, it is only plausible to argue that it is also
functional for grammatical forms proper. Indeed, it is my belief that grammatical anaphora obeys Accessibility theory too. Note that such a thesis can be
452
M. Ariel / Accessibility
The initial data motivating the Avoid Pronoun principle were cases where
native speakers show preferences for zero forms when both a pronoun and a
zero are grammatical. Relevant examples with P R O and pro are presented
below:
(8a) PRO: Maya wants 0/? for herself to win.
(8b) pro: 7ani/0 kamti meuxar ha - boker.
I
0 got up late
this morning.
I would like to argue that the Avoid Pronoun principle is redundant. Not
because I intend to dispute examples such as (8) above. I agree with the
judgments, but I find the principle as stated superfluous, even for the
, This is the approach adopted by Kuno in his recent book (Kuno 1987).
M. Ariel / Accessibility
453
distributional facts as they were taken to be. The optionality of the zero vs.
overt form coupled with any pragmatic theory, such as Kasher's (1976, 1982)
Rationality, Sperber and Wilson's (1986) Relevance or Levinson's (1985)
Minimization principle predict that whenever grammatically possible, the zero
form should be preferred. There is, therefore, no need for a special principle in
order to account for the zero choices. A problem does arise when pronouns
are preferred.
I claim that in fact, when the Avoid Pronoun seems to work it is because the
Accessibility associated witil the representation is extremely high, hence calling
for the use of a zero, an obviously attenuated form. Under such circumstances,
Accessibility theory, the pragmatic principles mentioned above and the Avoid
Pronoun principle all agree. The problem with the Avoid Pronoun principle
arises in those cases where it does not apply. Whereas Accessibility theory has
an explanation for the 'failure' to use the zero form, proponents of the Avoid
Pronoun principle offer no explanation. All they can offer is that such cases are
marked. Note, however, that ev'~ if this were the case, one still needs to explain
when and why marked forms are ever preferred over unmarked ones. Moreover, claiming that it is marked to prefer a pronoun over a zero form in a
language like English (Unlike Chinese, e.g.) is quite unmotivated.
The problem with the Avoid Pronoun principle (as well as the other theories,
when not mediated by Accesibility theory) is the prediction that preferences are
unidirectional (only zeros are favored over pronouns, never the reverse). This is
not true in actual discourse. The only reason why grammarians saw only one
side of the coin, namely, a preference for the higher Accessibility mark,,r, is ~he.:
they tend to concentrate on minimal units. No wonder speakers prefer higher
Accessibility markers when the unit involving both the antecedent and the
anaphor is the S domain, or even the more restricted, governing category
domain. Grammarians simply do not encounter examples where preferences for
lower Accessibility forms (to use my terminology) are at work.
Why, then, does the speaker bother to use a pronoun when it is redundant?
This is especially puzzling in cases, such as in Hebrew, where the choice
between an overt pronoun and a zero form entails no differences in the
information supplied on the referent intended by the speaker (this is so
because the verb often incorporates the same gender number and person
distinctions as the pronominal system). I suggest that Accessibility theory can
provide an account for this phenomenon. Relying on examples from Hebrew,
it will be shown that once the Accessibility of a given antecedent is seen as
relatively low, speaker prefers to use the relatively lower Accessibility marker,
a pronoun in this case. s
-" Another disadvantage of the Avoid Pronoun principle is that it is too specific. The same
regularity responsible for C-pronoun distributions is responsible for other forms: intermediate
pronominal forms (to be discussed below), 1 vs. reflexives, pronouns vs. reflexives, pronouns vs.
deicties, etc.
454
M. Ariel/Accessibility
'True' 0
(no agreement at all, as in Chinese)
'Poor' AGR (gender and number inflection, but no person marker, as
in Hebrew present tense)
'Rich' AGR ('Italian' inflection, where gender, number and person
are marked).
Although the mere existence of tense inflections does not guarantee zero
subject optionality (e.g., French), given the possibility to have empty subjects,
we should expect its distribution to obey Accessibility theory. Based on the
scale in (9), we derive the following universal Constraint on Distribution:
(10) The grammar is not expected to allow zero subjects with higher but not
lower Accessibility markers.
In other words, I predict that no language allows zero subjects with no
inflection whatsoever, but does not allow zero subjects to occur when AGR is
partially or fidly informative.
Factors related to the nature of the antecedent also play a role in sentencelevel anaphora. Since antecedent prominence facilitates its being a preferred
M. Ariel / Accessibility
455
M. Ariel / Accessibility
456
Table 5
The effect of informativity.
Applying the rigidity criterion, we can distinguish the two "poor' expressions:
3rd person ('anaphoric' AGR) versus present tense AGR. While in the former
person is not marked, it is still uniquely identified. The latter is truly
ambiguous. (I 3) has the relevant examples:
(13a) 0 halax; 0 halxa
~,ent-he went-she
VS.
(13b) 0
holex;
0
holexet
go-I (masc.)/you (masc.)/he go-I(fem.)/you(fem.)/she
The reason is that historically, future tense inflectioas by far pr~ate past tense inflections.
M. Ariel / Accessibility
457
( 15 Accessibility Hierarchy:
Full pronoun (Ist > 2nd > 3rd)
Cliticized pronoun (Ist > 2nd > 3rd)
Past tense person inflection (ist > 2nd)
Future tense person inflection (lst > 2nd)
Third person inflection (Past > Future)
'Degenerate' AGR (Present tense AGR)
I shall next present evidence in support of the Accessibility Hierarchy in (15).
The evidence will be of two types: intuitions concerning acceptability judgments, and patterns of distribution in actual discourse. We start with judgments concerning sentences in the abstract.
A split antecedent is an inferior antecedent. It is not as salient as a single
antecedent. Indeed, Borer (1985) finds that while an 'anaphoric' A G R (3rd
person inflection) can in general be anaphoric across S boundaries (16a),
when the antecedent is a split one (16b) the sentence is unacceptable ((16) and
(I 7) are Borer's examples and judgments):
(16a) Talila~ amra le-itamaq ~e- Om
hiclixa/hicliax.
Talila said to-ltamar that [she/he] succeeded.
VS.
(16b) *Talilai
amra le-itamarj ~e- Oi+j hiclixu.
Talila [did] said to-Itamar that [they] succeeded.
Note, however, that an anaphoric reference to a split antecedent is possible
with a lower Accessibility marker (a 'pronominal' A G R - 1st or 2nd person
inflection):
(17) Talilai Io sipra lexaj ~,e- O ~ . j
huzmantem la-mesiba.
Talila not tell you that [she&you] were-invited to-the party.
This is precisely the prediction of Accessibility theory. A lower Accessibility
marker can be anaphoric ~o a relatively less accessible antecedent for it signals
to the addressee that the antecedent is less than highly accessible.
That the ban on a link between an 'anaphoric' A G R and split antecedents
is nm gra~'mnatic:01 can be seen from the following example. Once we raise the
Accessibility .of the split antecedent by tightening the unity of the two clauses,
anaphoric AGR can be anaphoric to the inferior split antecedent (it is the
verbs of saying which block the anaphoric relation in the examples from
Borer, I claim):
M. Ariel / Accessibility
458
et
~imon-al
maamaro ha-guvenisti k~e(18a) Noga bikra
Noga criticized ACC Simon about his article chauvinistic when
nasu li-yrugalaim.
O
[they] went to Jerusalem.
ge-ha-paam O
yecu
Noga asked from Simon that this time [they] will-leave
la-mesiba
ba-zman.
for-the party on time.
M. Ariel / Accessibility
459
roakol
seret ~,alog peamim.
see every movie three times.
kol seret g,aio~, peamim.
every movie three times.
Note that in line with Accessibility theory, 'Pronominal' AGR (lst and 2rid
person future and past inflections), which is a lower Accessibility marker, is
allowed under such circumstances:
(21c) 0
raiti/rait/ere/tiri kol seret galog peamim.
I/you saw/will see
every movie three times.
We now move to supporting evidence which comes from actual data. The
difference in Accessibility between a full versus a cliticized pronoun (marked
by an apostrophe) can be seen in (22). In (22a), the topic is referred to using a
cliticized pronoun (lst person in this case), but when the discourse topic (or
frame) shifts, speaker changes to a full pronoun. In (22b), we see a similar
phenomenon with a third person referent. Speaker starts by contrasting a
name (a low Accessibility marker) for the non-topic versus a full pronoun for
the topic, and later, a full pronoun serves to refer to the accessible though
non-topic entity (Nubar=j), while a cliticized pronoun is used for the
accessible entity which is also the topic (indexed by i): 7
oto be-yom sheni, an' 1o
yodea ...
(22a) Anixogev ... an' oci
I think ... I' will print it on Monday, I' don't know ...
[Change of Discourse Topic]: ani xogev ge-ulay
ani ectarex
I think that maybe I will-have
li-nsoa le-london be-yore ~,eni.
to go to London on Monday.
(22b) Hul diber imnubarj, nubarj amar ... nubarj haya ... h'l pashut diber
He spoke to Nubar. Nubar said ... Nubar was ... He'simply spoke
itoj.
hu e xasav kge huj ...
to-him, he thought when he ...
The following examples show the sensitivity of the 0/pronoun choice to
questions of unity. (23a) and (23b) are practically a minimal pair, taken from
the same story. All along the story, the narrator is extremely accessible. She
'The data for spoken Hebrew here and elsewhere in this paper is based on a taped kitchen
conversation between wife and husband (Jan. 8, 1987).
460
M. Ariel / Accessibility
constitutes the discourse topic. This is why the translator mostly chose O's for
the original English L The interesting difference is the one between the
underlined 0 vs. pronoun. In both cases, the previous discourse topic is the
mother, rather than the narrator. Still, in (23a) she resumes talking about
herself using a 0. In (23b), a full pronoun is used for the same referent.
Neither distance nor topicality can explain the difference. I suggest the reason
is the sharp shift in the nature of the flow of events which dictates the
different strategy: s
i~a
le- heanes.
(23a) Ze haya davar Eel ma bexax bi~vil yalda o
It was nothing
for (a) girl or (a) woman to get raped.
ani acmi neenasti, k~e- 0 hayiti bat ~tem esre.
I myself was raped, when [I] was twelve years old
ima af paam 1o yadaa, u- 0 meolam 1o siparti le-iL (Walker 1985)
Mama never knew,
and [I] never
told anybody.
ve-lifamim,
(23b) Hu pagut him~ix le-nasot le-alec oti la-cet ito,
He just kept trying to make me go
with-him, and sometimes,
mi-tox hergel, ani xogevet, 0 halaxti ito.
gufi
asa ma ~e
out of habit, I guess, [I] went with-him. My body did what (it)
~'alam ~e
yaase, ve- ima
meta. ve-ani haragti
was-being-paid to do. And Mother died. And I killed
et
buba. (Walker 1985)
ACC Bubba.
I found that examples like these are statistically significant. In my data for
spoken Hebrew, only 17.5% of the past tense verbs were accompanied by
overt pronouns. All of them are of the (b) type, i.e., manifesting a relatively
low unity between antecedent and anaphor clauses.
The difference between past versus future person inflections is mainly
revealed in the frequency of allowing 0 subjects. In spoken Hebrew, 82.5% of
the cases had a 0 rather than a 1st or 2nd person pronoun in past tense
inflections No zeros occurred in future inflections. In literary Hebrew, the
gap is smaller: 90% ~'s in past, 76.5% in future. This is not a grammatical
'finding', but the large gap in popularity does mean that speakers distinguish
between past arid future inflections.
Even 'anaphoric' AG R (3rd person past/future inflection) (in 24a)
and a 'degenerate' A G R (present tense) (24b, c,d) can refer extrasententially, provided the antecedent is highly salient, a topic in the following
examples:
s See notes I and 3 and for precise references of written Hebrew examples, i have specified the
author's name at the end of each example.
M. Ariel / Accessibility
461
(24a) Hu mitxaten,
buba [el baxura, g
hayta gam yaxad ito.., beten
He is marrying. A darling girl.
[she] was there with him.., little
gtuxa-- ~ ulay bat 22, aval niret bat 17. (Paley !980).
tummy-- [she] must be 22, but looks seventeen.
(24b) Ba-zman ha-axaron ani mitoreret be-~eva, ani menasa li-~,on.
Recently
I wakc-up at seven. I try
to sleep.
be-dere-klal ~ mityae~et. (Spoken Hebrew).
Usually
[I] give up
(24e)
af paam io moce kan klum
ii-~tot ...
[I] Never
find here anything to drink ...
(Literary Hebrew) (Paley 1980)
(24d) Kogadol
~uv ha-merxak beynehem.
~
medabrim
So large [is] again the distance between them. [They] speak
be-milim teomot u-mexavnim li-dvarim axerim.
compatible words and meaa
different things.
(Literary Hebrew) (Savion Liebrecht, Applesfrom the Desert, !986; TelAviv: Sifriyat Ha-Poalim)
The above have been a selection of examples aimed at showing that O
pronoun choices are actually choices among a richer variety of forms, each
specialized for a different degree of Accessibility, which dictates its distribution. More data can be found in Ariel (1990), where the evidence regarding
sentence-level anaphora is not limited to Hebrew, nor to O/pro choices.
Regarding the latter, the claim is that Chinese zero subjects, for example, are
not quite as free as is often assumed. For grammarians it is of the utmost
importance that Chinese O's can refer extra-textually That feature immediately turns (some) O's into an extra-grammatical phenomenon, not to be
handled by the grammar. However, the mere fact that such O's can refer
deictically does not mean that they are completely free. Actually they too are
heavily restricted - to highly salient objects from the discourse setting, to
topics, speaker or addressee. As such, they are not as different from the
Hebrew O's. Even English, I would claim, is not as different from Chinese as
grammarians make it out to be (on this point). English and Chinese allow O's
under the same circumstances, except that in English O's are marked. The
following shows that English too allows extra-textual references by 0:
(25) Careful! 0 contains Methanol
5. In conclusion
462
M. Ariel / Accessibility
expressions, definite NPs among them, are expressions each specialized for a
specific degree of memory Accessibility. Accessibility, therefore, is not only a
cognitive concept. It is a linguistic feature. In section 3, I proposed that the
codification of the cognitive concept into linguistic form is mediated for the
large part (excluding markedness considerations) by three criteria: informativity, rigidity and attenuation. Although a universal scale of Accessibility is
predicted, each language carves its particular scale, excluding those expressions it has no correlates for, and establishing preferred (unmarked) forms
versus marginal (marked) forms.
Section 2 provided some examples from discourse anaphora. Section 4 was
devoted to arguing that the scale of Accessibility marking is essential not only
for discourse anaphora. Sentential anaphora obeys Accessibility theory too. I
have mainly concentrated on optional choices regarding both discourse and
sentential anaphora, arguing that speakers make consistent choices in their
anaphoric expressions, favoring relatively higher Accessibility markers when
anteceder, t is highly available (when it is a topic, or when the anaphor appears
in a unit highly connected to the ore where the anteceden~ occurred, etc.).
The only strictly grammatical poJ nt I made was the proposal that languages
are restricted in how 'liberal' their options regarding a given Accessibility
marker can be: a language can ne eer license the use of a high Accessibility
marker in a context where it does not license the use of a lower Accessibility
marker. Hebrew, I have argued, o3eys this restriction. 'Degenerate' AGR is
more restricted in occurrence thar 'anaphoric' Agr which is, in turn, more
restricted than 'pronominal' AGR. However, a qualification is here in order:
being more restricted does not necessarily mean occurring less frequently! The
more heavily-defined context may 'be quite frequent, at least in some genres.
I have argued that Accessibility, an obviously cognitive concept, has a
linguistic correlate. Can this claim leave the grammar-pragmatics borderline
intact? I believe it can, if we adopt Kasher's (1976, 1982) and Sperber and
Wilson's (1986) view of pragmatics. According to this view, a phenomenon is
pragmatic if it is not language-specific. Grice's (1975) 'be cooperative', Kasher's (1976, 1982) 'be rational', 'be polite', and Sperber and Wiison's (1986)
'be relevant' are general behavior patterns.
Accessibility conventions, on the other hand, are language-specific in that they
are formulated over specific linguistic expressions ,~nd are therefore part and
parcel of the grammar. Under one version of the Chomskyan view, where,
linguistic vs. non-linguistic, sentence vs. discourse and licensing principles vs.
principles governing optional choices all boil down to one and the same
distinction, that of grammar vs. pragmatics, Accessibility theory does conflate
the two. However, Asa Kasher (personal communication) has drawn my
attention to a few passages in Chomsky's writings where Chomsky does seem
to take the former position regarding the grammar-pragmatics division of
labor. Following Chomsky, Kasher (1991) proposes to distinguish between
M. Ariel / Accessibility
463
general pragmatics and linguistic pragmatics, a component within the grammar. Accessibility theory, then, belongs in linguistic pragmatics.
References
Anderson, A., A.C. Garrod and A.J. Sanford, 1983. The accessibiliq, of pronominal antecedents
as a function of episode shifts in narrative texts. Quarterly Journal of Psychology 35A: 427440.
Ariel, Mira, 1985. Givenness marking. Ph.D. Thesis, Tel-Arty University.
Ariel, Mira. 1988. Referring and accessibility. Journal of Lir,guistics 24: 65-87.
Ariel, Mira, 1990. Accessing NP antecedents. London: Routledge (Croom Helm Linguistics
Series.)
Borer, H., 1985. Anaphoric AGR. Ms., University of California at Irvine.
Broadbent, D.E., 1973. in defence of empirical psychology. London: Methuen
C::omsky, Noam. 1981 Lectures on government and binding. Dordrecht: Forts.
Clancy, Patricia M., 1980. Referential choice in English and Japanese narrative discourse. In:
Wallace Chafe, ed., The Pear Stories, 127-202. Norwood~ NJ Ablex.
Clark. Herbert H. and C.R. Marshall. 1981. Definite reference and mutual knowledge. In: A.
Joshi et ai., eds., Elements of discourse understanding. 10-63. Cambridge: Cambridge University Press.
Clark, Herbert H., R. Schreuder and S. Buttrick, 1983. Common ground and the understanding
of demonstrative reference. Journal of Verbal Learning and Verbal Behavior 22: 245-258.
Giv6n, Talmy, 1983. Topic continuity in discourse: An introduction. In: T. Giv6n, ed., Topic
continuity in discourse: A quantitative cross-language study, 1-42. Amsterdam: Benjamins.
Kosher, Asa. 1976. Conversational maxims and rationality, in: A. Kasher, ed., Language in
focus: Foundations, methods and systems, i97-216. Dordrecht: Re~del.
Kasher~ Asa, 1982. Gricean inference reconsidered. Philosophica (Gent) 29: 25--44.
Kasher Asa, 1991. On the pragmatic modules: A lecture. Journal of Pragmatics 16:381-397 (this
issue).
Kuno, Susumu, 1987. Functional syntax. Chicago, IL: The University of Chicago Press.
Levinson, Stephen C., 1985. Minimization and conversational inference. Paper delivered at the
International Pragmatics Conference at Viareggio, Italy, September 1985.
Li, Charles N. and Sandra A. Thompson, 1979. Third-person pronouns and zero-anaphora in
Chinese discourse. In: T. Giv6n. ed., 1979, Syntax and semantics, Vol. 12: Discourse and
syntax, 311-335. New-York, Academic Press.
Marslen-Wilson, W., E. Levy and L.K. Komisarjevsky Tyler, 1982. Producing interpretable
discourse: The establishment and maintenance of reference. In: Robert Jarvella and Wolfgang
Klein, eds., Speech, place and action: Studies in deixis and related topics, 339-378. Chichester:
Wiley.
Prince, Ellen F.. 1981. Toward a taxonomy of given-new information. In: P. Cole, ed., Radical
pragmatics, 223-255. New-York: Academic Press.
Sanford, A.J. and S.C. Garrod, 1981. Understanding written language. Chichester: Wiley.
Sperber, Dan and Deirdre Wilson, 1986. Relevance. Oxford: Blackwell.