The description of the complementation pattern requires specification of the categorial selection for each slot. However, in most cases, bare categories do not supply enough information and categories must thus be completed with restricting features.
The most acknowledged and general needs concern verbal complementation:
Other needs arise with the encoding of full scale lexicons and the requirement to deal with forms of complementation other than verb complementation.
To provide for all these needs, we allow categories (terminal or non-terminal) to be completed and specialised with restricting features stemming from unification-type formal grammars (HPSG, LFG) and lexicons (GENELEX).
Typed feature formalisms were mostly developed for grammars. Our task is to define a typed feature model for a lexicon. The main part of the job consists in listing the relevant features. What kind of restrictions do we need to encode in a lexicon? Which features will be made available to lexicographers?
Three kinds of features must be clearly distinguished:
There is clear evidence for morphosyntactic restrictions at the phrasal level: infinitivals, impersonal subjects, third person, instrumental case, adverbs of degree, etc.
Morphosyntactic features may be split into:
The EAGLES CLWG Morphosyntax group has already defined, the set of relevant attributes (features) for each category. We decided to rely on these features and to take them as a starting point, an input and a checklist for our phrase restrictions for different reasons:
One of the main advantages of the representation chosen by the morphosyntax group is to take language specificity into account without providing an over-generating union of all possible and `blind-to-language' valued features.
The features are articulated on different levels, corresponding to different degrees of obligatoriness:
Morphosyntactic features to be added on a terminal category realising a slot are directly inherited from the morphosyntax specifications.
A phrase inherits (in linguistic terms, not in unification terms) from the characteristics of its head and bounded elements. In other terms, some of the characteristics of a phrase are linked to the characteristics of its components.
Before entering into details, the reader should be warned about our goal. It is not the point here to decide which category actually instantiates such and such a head (or specifier and so on). This issue is highly controversial and remains open. Take nominal heads, for instance. May a nominal head be a N, Pro, Adj, Adv and even a Det in some cases?
|(73)||Let's call our neighbour|
|(74)||Let's call John|
|(75)||They pleased him|
|(76)||She bought the blue one|
|(77)||She always gives to the poor|
|(78)||I have both|
|(79)||He wants many of them|
In this section, our goal is to decide which features defined at terminal level remain relevant at phrasal level for phrasal specification and subcategorisation. Thus, we are more interested in virtual and prototypical heads (cf. theory) than in actual heads for our specifications. For each phrase, some terminal components must be examined in detail to check whether their morphosyntactic attributes would be relevant for phrasal restrictions.
It is well known that morphosyntactic attributes are language dependent. Consequently, phrasal morphosyntactic restrictions that are linked to terminal categories are also language dependent. The list of phrasal morphosyntactic features cannot be specified in absolute terms.
The method for specifying morphosyntactic restrictions on phrases is the following:
To illustrate these points, we examine what we derive for English verb complementation. Note that different tables would have to be established for noun, adjective and any other subcategorising category. Our first example covers restricted S (table 4.1) as illustrated in the example fragments (80)-(84):
|(80)||To expect that something will happen||(S[Fin:yes])|
|(81)||She lets Mary cook on sundays|
|(82)||I wonder why we should come||(S[WType:inter])|
|(83)||I wonder whether we should come||(S[WType:inter])|
|(84)||I wonder who should come||(S[WType:inter])|
Our next example covers restricted VP (table 4.2) as illustrated in the example fragments (85)-(86):
|(85)||You can lose your wallet||(VP[Fin:no])|
|(86)||They started screaming||(VP[VForm:ING])|
Our third example covers restricted NP/PP (table 4.3) as illustrated in the example fragments (87)-(89):
|noun & det||Numb||no|
Verb is introduced as a component of PP only for infinitivals and participials. Morphological number cannot be expected to represent a semantic contraint such as collective subject -- collective subjects may occur as singular or plural. Impersonal pronoun will be directly specified on the PRONOUN terminal category, not on the non-terminal NP category. This is why the Type attribute associated with pronouns is not relevant for NP specifications.
|(87)||They pleased him||(NP[Case:acc])|
|(88)||They asked him to come||(PP[Fin:no][Mood:inf])|
|(89)||He excels in entertaining people||(PP[Fin:no][Mood:gerund])|
Our final example concerns restricted AP, (table 4.4) as illustrated in the example fragment (90)
|(90)||He looks confident||(AP[Use:pred])|
The selection of a specific lexical item is another important requirement for syntactic subcategorisation.
Bound prepositions such as to or for introducing a NP or a VP, complementisers such as that starting that-clauses or whether starting embedded interrogative clauses -- all these elements are the best-known subcategorised lexicalised items. They can be gathered together under the class of introducers.
An introducer is:
The notion of introducer prevents us from making any statement about the link between the category of the introducer and the category of the introduced phrase. Introducers may equally be considered as true introducers or as phrase headings and heads. This depends on the theory and on the linguistic phenomena analysed.
This feature provides for the means of specification of introducers which play a very special role in the left-to-right expansions of languages (centrifugal in Tesnièrian terminology).
|(91)||He wants to work||(PP[Introd:TO] vs. VP[Introd:TO])|
|(92)||He looked forward to the meeting||(PP[Introd:TO] vs. NP[Introd:TO])|
|(93)||He went to the movie||(PP[Introd:TO] vs. NP[Introd:TO])|
A terminal category realising a slot may also need to be lexicalised for the purpose of subcategorisation. In this case, since we are not at phrasal level but at terminal level, we will use another feature called Lex:
This feature applies to light-verb/noun dependency and any frozen or semi-frozen expression. However, we will not develop this issue here.
|(96)||Mary expected [a friend/to welcome a friend/that a friend came]|
|(97)||A friend is expected by Mary|
Direct objects which may be passivised will bear this property. This feature must be considered as a short cut for a frame alternation. It does not prevent one from describing the passive frame and the alternation between active and passive form.
Some rewriting specificity may be expected for subcategorised phrases. In such a case, and to avoid a complete rewriting, a SubCat feature will be associated with a phrase.
|(99)||Plenty of salt||(PP[Introd:OF][SubCat:NODET])|