Semantically transparent linking in HPSG

Most researchers now agree that subcategorization correlates significantly with semantics. But this semantic component of linking has proved elusive. Most, if not all, theories of linking have, in pratice, resorted to constructs that are syntactic diacritics. We show in this paper that the implicit syntactic diacritics that plague the basic linking constraints posited in at least some of these theories can be eliminated, provided that (i) the metalanguage in which linguistic constraints are written allows for true implicational statements; (ii) one is willing to slightly increase the number of linking constraints. We focus in particular on the linking theory presented in Davis and Koenig 2000, Davis 2001, and Koenig and Davis 2000, but we maintain that our arguments apply, mutatis mutandis, to many other linking theories. We note some of the consequences of this view of linking, including: linking constraints are stated in terms of semantically natural classes of situations, a single entailment of a verb’s argument is sufficient to determine its linking, and interaction among linking constraints restricts the range of possible lexical items. Most researchers now agree that subcategorization correlates significantly with semantics (see, among others, Foley and Van Valin (1984), Pinker (1989), Jackendoff (1990), Levin (1993), Goldberg (1995), Wechsler (1995b), Davis and Koenig (2000b)). To put it in motto form, knowing the meaning of a verb is to a large extent knowing its context of occurrence. But this semantic component of linking has proved elusive. Most, if not all, theories of linking have, in pratice, resorted to constructs that are syntactic diacritics. We show in this paper that the implicit syntactic diacritics that plague the basic linking constraints posited in at least some of these various theories can be eliminated, provided that (i) the metalanguage in which linguistic constraints are written allows for true implicational statements; (ii) one is willing to slightly increase the number of linking constraints. Because of space considerations, we focus in particular on the linking theory presented in Davis and Koenig (2000b), Davis (2001), and Koenig and Davis (2001). But we believe our arguments apply, mutatis mutandis, to many other linking theories. 1 Syntactic diacritics in semantically-based linking theories We first briefly present the approach to linking in HPSG described in Davis and Koenig (2000) and Davis (2001). This linking theory is based on three crucial ideas: (1) A multiple inheritance hierarchy of semantic relations; (2) a multiple inheritance hierarchy of predicator types defined by how they link attribute values within their CONTENT to members of the ARG-ST list (more precisely, to the situational nucleus of their CONTENT); (3) a metatheoretical constraint on the relationship between the hierarchy of semantic relations and the hierarchy of predicator types. We thank Detmar Meurers for discussing some of the issues raised in this paper. All remaining errors are solely ours. Consider how this theory accounts for the linking of arguments displayed in the following simple transitive sentence. (1) Sandy moved the ball. The fact that, for all English transitive verbs that denote causes changing the states of entities, the cause is realized as the subject of its active form, and the entity changing state is realized as the direct object, is modeled through the interaction of three constraints. First, the CONTENT of move includes a semantic relation which is a subtype of both act-rel and und-rel. This is illustrated in figure 1 where lines between nodes labelling semantic relations indicate a subtype-supertype relation. Such a semantic hierarchy, which encodes the (linguistically relevant) relations between categories of situations, helps restrict the grammatical constraints on the realization of semantic arguments to the proper semantically-defined class of verbs. rel