Preliminary Recommendations

 

Semantic class

Information concerning semantic class is considered in the present context only insofar as it enters into the account of lexically-governed syntactic processes (such as frame alternation). The kind of semantic class information that is of use mainly for disambiguation purposes (e.g. PP attachment) is not relevant at the level of linguistic description we are concerned with here, on the grounds that it involves processing issues which pertain to the role of semantics in NLP and which are to be dealt with accordingly by a prospective EAGLES Lexicon/Semantics group.

For each slot filler, semantic class information is optional, although it is highly recommended whenever it has a bearing on a proper account of phenomena which are already represented in the lexicon. A case in point is frame alternation (see diathesis alternation). By way of exemplification, consider the case of causative-inchoative alternation illustrated in (148)-(153):

These examples show that a given verb does not always undergo the alternation. For each example, possible and impossible alternations are contrasted. Two different cases can be distinguished: (148)-(149) and (150)-(151) show that not all possible objects of the transitive reading can play the role of subject of the intransitive reading of the same verb; conversely, (152)-(153) illustrate the case in which not all possible subjects of the intransitive reading can be the object of the transitive one.

From this it follows that the representation of the causative-inchoative alternation of different verbs in the lexicon requires that semantic class information concerning the argument alternating between two possible syntactic realisations (i.e. as object of the causative reading and as subject of the inchoative one) be also considered (Montemagni et al., 1995; Montemagni & Pirrelli, 1995). For instance, with the verb begin, the alternation is possible only with event-denoting objects (e.g. concert, lesson, show) and with gather, only with inherently volitional ones (e.g. animals, friends). With ring, more subtle lexico-semantic distinctions have to be considered: for instance, bell and door bell can both be subject and object of the causative and inchoative readings respectively, but telephone and alarm clock can only play the subject role within the inchoative reading.

Starting from the simplest case, consider the possible representation of gather in an EAGLES-conformant lexicon, as shown in tables 4.18-4.20:

 

frame_1 ([slot_1 ([slot_real_1(CatLabel: np, Subject: subj)]),

slot_2 ([slot_real_1(CatLabel: np, Subject: comp)])])

 

frame_2 ([slot_1 ([slot_real_1(CatLabel: np, Subject: subj)])])

 

frame_list ([related_1(FrameIndex: 1, SlotIndex: 2, FrameIndex: 2, SlotIndex: 1)])

Here, the condition is tacitly enforced that relatedness, as expressed at the frame_list level, holds if and only if the `semclass' values specified at the slot level are met by both related slots. To be more concrete, in the case of gather the alternation holds only with volitional_animate entities, but fails to apply with entities that are not inherently volitional. In terms of extensional semantics, this means that an alternating argument must belong to the intersection between the sets of entities and the set of volitional animates.

A more complex case is represented by the verb ring where the causative-inchoative alternation appears to be sensitive to rather specific conceptual nuances like those distinguishing bells from telephones or alarm clocks. This example, when compared with the previous ones, shows that the semantic class information needed to account for frame alternation ranges from general restrictions, corresponding to broad semantic classes (such as `animate', `human' and the like) on the one hand, to highly specific restrictions on the other hand. It is not always easy to isolate linguistically relevant distinctions from cognitively possible ones as, in some cases, the distinction is not so clear-cut. This difficulty relates to the more general issue of how semantic class information should be represented in the lexicon. Although this problem is not immediately relevant here (arguably, a final decision on the matter should be taken at the level of semantic description), we raise this point at this juncture as an issue worthy of consideration, with a view to the development of standards for the representation of semantic information in the lexicon.