37.A meeting... The last meeting but two
W ith the restrictions she imposes, Sidner misses cases like A conference ... the first talk, or C lark's example I swung three times. The first swing... Again, it is not clear if these would be cases treated by the rules of associated or inferred specification. When no relation can be established, Sidner says that definite NPs with no modifiers are odd uses. For those descriptions that have modifiers she says that they specify outside the discourse context.
8. Strand's taxonomy of linking relations
S trand's approach (Strand, 1996) is also mainly concerned with those cases of definite description use in which an explicit contextual relation (link) holds between the description and an antecedent (anchor). Strand, as Sidner, assumes the availability of a semantic representation of the text (in this case, DRT) and inference mechanisms. He proposes taxonomy of linking relations in which five main classes are distinguished along with fifteen subclasses. They are as follows:
C o-referentiality
T he antecedent and definite refer to the same entity through identical or different description.
I dentical head The anchor and the definite description share the same head noun
38.A yellow car... The car...
G eneralization The definite description is more general than or is a synonymous of the antecedent.
39.A car... The vehicle...
S pecification The description is more specific than the antecedent.
40.A car... The sedan...
R edescription The definite description is a fully alternative description of the antecedent which neither entails nor is entailed by any conditions on (properties of) the antecedent. 41.A car... The notorious wreck...
S trand's co-referential class differs from Clark's direct reference. Whereas Clark classifies the house was large... The size surprised me as direct reference, Strand does not. Other differences in their taxonomies are discussed below.
N arrowing
T he definite is part/member or an argument/role of the antecedent.
S et-member The description is a member or a subset of the set indicated by the antecedent.
42.A school class... The girls...
W hole-part The description constitutes a part of its antecedent.
43.A car... The engine...
E vent-argument The description is an argument of an antecedent event.
44.John was murdered. The murderer...
W idening
T hese are cases which expand on familiar sets.
M ember-set The description is a set of which the antecedent is a member or a subset.
45.John and his nephew... The family...
P art-whole The description has the antecedent as its part.
46.A wall... The building...
Adjoining
P art-part The antecedent and description are members of the same state or parts of the same whole.
512f4f2f7375a417866f8f19st Wednesday... The next day...
P ossessor-thing The antecedent possesses the description.
48.A professor... The car...
D elimitation
I n these cases the anchor may be seen as an argument to the description.
A rgument-event The description is an event in which the antecedent is an argument delimiting its denotation.
49.Israel and Egypt... The peace agreement...
S ubcategorization The description subcategorizes for something of the antecedent’s type. This applies to so called relational nouns like father, weight, price, owner, driver, etc. 50.A bicycle... The price...
T ime-anchored The time region indicated by the antecedent gives a more delimited or unambiguous reading to the description.
512f4f2f7375a417866f8f19st Wednesday... The news...
S pace-anchored The space region indicated by the antecedent delimits the description. 52.A Greek village... The taxi drivers...
S trand also mentions the existence of implicit or inferred anchors: for instance, when someone is telling about a visit to a Greek village, the (implicit) time of visit may be an anchor to the referents of the descriptions.
S trand acknowledges the problem of multiple anchors/links being available for a description resolution. He says that one should give preference to the most informative link and that identity should be preferred whenever possible. However, besides the problem of deciding between identity and non-identity, it seems hard to find a way of identifying a `most informative' link. Strand mentions that an opposite approach is one like Sidner's, where a saliency order is followed.
11. Comparison of terminology
I n this section we compare the classifications presented in the previous sections. We present tables which relate examples of definite description use with the classes identified by the authors we have discussed. Tables 1 (a and b) and 2 (a, b, c and d) describe the anaphoric and associative uses respectively. Seven different schemes are listed (Hawkins, 1878; Prince, 1992; Fraurud, 1990, L?bner, 1985; Clark, 1977; Sidner, 1979; Strand, 1997). The other tables (Tables 3 and 4) consider only four of the authors, since not all authors refer to the phenomena presented there (situational and unfamiliar uses). Although Sidner notices that definite descriptions may specify outside the linguistic context, she does not explain in which different ways. Strand briefly mentions the existence of implicit and inferred anchors. Clark is only concerned with discourse relations. The terms that
appear in the tables in Italics are our guesses for the examples not explicitly discussed by the author of the corresponding scheme. Question marks were placed where the authors were generally silent about the case, and it was not clear whether their classification would apply or not to the example. We can see that those authors who present a more comprehensive characterisation of uses of definite descriptions (Hawkins, Prince, Fraurud and L?bner) do not discriminate anaphoric and associative descriptions in as much detail as the others (Clark, Sidner and Strand) do. On the other hand the first authors pay special attention to situational and unfamiliar uses. Also note that there is no absolute consensus about the sub-classifications of the various uses.
T ables 1 (a and b) lists the anaphoric uses. Hawkins and Prince do not make any distinction among them. It is not clear whether Prince would consider the definite description in a sequence like he travelled... the journey as textually evoked, nor if Fraurud would consider that as subsequent mention. L?bner only refers explicitly to “direct anaphora”, those cases based on an antecedent with identical head. But what he calls pragmatic anaphoric seems to apply well for all examples in the table. L?bner says that the construction of a universe of discourse is comparable to the braiding of a complex multi-dimensional network, with object and event nodes; every node is a potential discourse referent, and anaphoric descriptions are used to refer to nodes in the net, usually providing only sortal information for the retrieval of their referent. Clark, on the other hand, distinguishes among four different ways in which a co-reference relation may be realised, but he is silent about the cases that Strand calls specification and widening. Sidner considers only two types of co-reference: identical head and generalization. Both Fraurud and Strand observe for a sequence like a man, a woman ... the couple a difference in the entities represented by the description and antecedent. Strand (1997) explains that events in his framework are represented by discourse referents and a link for cases like he travelled... the journey would be of the coreferentiality class. For named entities he explains that usually the relation is coreference, and the subclass specification or redescription. Clark and Strand give the most comprehensive account for the anaphoric use.
T ables 2 (a, b, c and d) summarises the classifications of associative uses and is the most complex of all. It is difficult to complete the table for each different author, since usually they are not explicit about all the possible associations capable of linking bridging descriptions with their anchors. Hawkins reckons the difficulty in providing the defining parameters for the set of possible associates; he then comments on the more general defining characteristics of these associations. He says that speaker and hearer share general knowledge of relationships between triggers and associates, usually part-of relations and attributes. It is not clear if he would consider event roles or cases involving hearers' supposition as associative anaphora.. Hawkins does not explicitly refer to a description as being associated to a previous VP, although this is considered in the examples of
