This is a preview of subscription content, log in via an institution to check access.
Preview
Unable to display preview. Download preview PDF.
References
In fact, HPSG was born at the Hewlett Packard laboratories as a result of research into implementation of GPSG. See, for some details, [Flickinger et al., 1985]; and [Proudian and Polland, 1985].
At this point it may be fair to clarify a few points as to what the terms information-based, feature-based, unification-based refer to. Following Shieber ([Shieber, 1987]), we can conceive of the structure of a linguistic theory as a series of relations. First we have a theory (say, HPSG) which somehow restricts the formalism (say, HPSG F) in which an analysis (say, a grammar of Portuguese) is expressed, which, in turn, predicts the analyses of the constructions of the Portuguese language. Thus, the term information-based is to be related to the theory (i.e., HPSG is an information-based linguistic theory), while the terms (complex) feature-based and unification-based should be related to the formalism (i.e.,HPSG F is a complex-feature-and unification-based formalism). Thus, in principle, an information-based theory need not necessarily be complex-feature-based nor unification-based, and vice versa.
Following the position taken in IBSS by Pollard and Sag I will be rather neutral as to what is the exact nature of signs, whether they are mental objects or real objects. The position of HPSG with respect to such issues as the conceptualism vs. realism debate is to remain silent. They acknowledge that there are good arguments for both positions, but they do not consider any of them to be conclusive. The aim of HPSG is then to work out a theory of linguistic information and of signs as information bearers independently of what is the exact nature of such entities. As Pollard and Sag put it: “we believe it would be premature to make particular claims about the relationship between the information structures that we posit and the structures of those mental objects that actually encode linguistic knowledge inside human brains. If indeed the hypotheses that we set forth about the nature of universal grammar can be correctly construed as being about mental phenomena, then they should be subjected to empirical verification on the basis of psycholinguistic experimentation” (IBSS, p. 6).
The HFP is a close relative of the GPSG Head Feature Convention (HFC), the main difference between them being that the HFC works as a default inheritance mechanism; see [Gazdar, 1987].
The advantages of a hierarchical encoding of grammatical relations are discussed in [Johnson, 1987], [Johnson, 1988]. For a different approach in which no hierarchical encoding is used and grammatical relations are assumed to be primitives, see [Gunji, 1987], where SUBCAT sets are used instead of SUBCAT lists.
Here I follow IBSS in keeping role labels distinct for each verb. This situation seems to have changed recently and traditional role-names like agent, patient, goal, etc. have been introduced in the CONTENT attribute, as for example in [Pollard, 1990].
This is the version of the BIP given in [Pollock, 1989]. I have only changed the name which in the paper is Nonlocal Principle.
Note that the representation in Figure 11 would be rejected in English on independent grounds, since there is no rule in the grammar of English which sanctions a flat structure where a saturated sign immediately dominates its lexical head without it being marked as inverted. The problem remains however. Consider a null subject language like Spanish where subject inversion is possible: (i) a. Juan quiere a María (J loves M) b. quiere Juan a María (loves J M) c. quiere a María (loves M) d. es el tipo de persona que nunca sé a quién quiere (is the kind of person that I never know who loves) (ia) is a simple declarative. (ib) has the subject inverted and it could be the yes-no question version of (ia), but it could also be part of an embedded indirect interrogative like no sé si quiere Juan a María (I do not know whether J loves M), where inversion is not obligatory. (ic) could be the null subject version of either (ia) or (ib). Finally, (id) shows that extraction of the subject is possible from embedded direct interrogatives, where subject inversion is obligatory. Now, it is generally assumed that the structure of inverted sentences is flat, but inversion in null subject languages is rather free and it is possible with any verb, so that we cannot rely on a lexical feature marking verbs as invertible and non-invertible; it is likely then that Spanish has a rule licensing flat structures like the one which will sanction a substructure like that of Figure 11. Thus, a sequence like quiere a María can be analyzed as being flat with one extracted complement, in which case nothing prevents the grammar to provide for (ia) an analysis like that of Figure 11. Borsley ([Borsley, 1987]) offers a similar argument with reference to this problem to which I will return in the next section.
Part of the contents of this section is an extension of the work reported in ([Balari et al., 1990]), and it is the result of close interaction with my colleagues to whom I am very grateful for many helpful suggestions and criticisms.
The fact that certain features like COMP must also be present in the mother category may be accounted for by extending the HFP, as in [Warner, 1989], or just by stipulating it directly in the rule as in [Balari et al., 1990]. Note an important feature of this rule that even though the minor category has some head-like status, the head is still the major phrase to which it is attached, thus, unlike GB we do not have complementizer phrases, but just sentences with complementizers, as in GPSG.
Author information
Authors and Affiliations
Editor information
Rights and permissions
Copyright information
© 1991 Springer-Verlag Berlin Heidelberg
About this paper
Cite this paper
Balari Ravera, S. (1991). Information-based linguistics and head-driven phrase structure. In: Filgueiras, M., Damas, L., Moreira, N., Tomás, A.P. (eds) Natural Language Processing. EAIA 1990. Lecture Notes in Computer Science, vol 476. Springer, Berlin, Heidelberg. https://doi.org/10.1007/3-540-53678-7_4
Download citation
DOI: https://doi.org/10.1007/3-540-53678-7_4
Published:
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-53678-9
Online ISBN: 978-3-540-46975-9
eBook Packages: Springer Book Archive