Problem space: s-construct

Operator Overview:

This operator may form part of the s-constructor being constructed. It removes a fuse relation between a semassigner, A and a semreceiver, R set up previously by a fuse operator. Semantic snips are similar to syntactic snips with respect to implementation however the way snips are proposed are different.

Operator Proposal:

Snipping structures that are no longer active

Changes in the utterance model (i.e. due to syntactic snip ) may mean that a constituent in the semantic model is no longer active. Syntactic and Semantic profiles are linked together so that when a syntactic constituent is snipped out of the current u-model, both its syntactic and semantic profiles are marked as being inactive. This category of semantic snip is proposed if there is a fuse between two semantic constituents and one of these constituents is inactive. There are four such proposals, one for checking inactive assigners, one for inactive receivers and one each for inactive assigners and receivers in a property relation. For an example see sentence 2 in the regression corpus. See here for productions.

Special semantic snip proposals.

There are another group of proposals that have been built for more specialized cases. It is not clear how general these proposals are nor if more similar proposals may be necessary.

Proposal for double adjectives/ properties

In some instances the syntactic structure may have changed without leaving syntactic or semantic structures inactive. The same constituents are active as before but the are now arranged differently. One example is in the case of double adjectives where the adjoin and property relations may change. For example see sentence 3 in the regression corpus :"contact is confirmed hostile ". The proposal for this snip is in two parts, the first production notices that a property in semantics is adjoined to something different in syntax. The second notices that the property is no longer justified by the 'is' relation in syntax.

[ I can't remember the details of this and why I had to split the proposals up. The whole area of adjectives and properties is a bit unclear as there are many syntactical ways a semantic property may be justified : this leads to complicated constraints checks and snip proposals ]

Proposals for incorrect subcategorisation

Another instance when syntax changes but leaves the same constituents active is when the subcategorisation of a verb is changed from one category to another.

One proposal looks for a semantic constituent that assigns an internal of category thing while the corresponding syntax assigns something of category CP. For example see sentence 29 in the regression corpus : " the man wanted the horse to like the apple".

Another proposal looks for a semantic constituent that assigns an internal of category action or state while the corresponding syntact entry assigns a noun complement. [There may be more proposals needed of this type]

Special proposal for dative verbs

This is a special semsnip proposal for when the non-dative form of a usually dative verb is used. For example see sentence 17 of the regression corpus: "The man gave the apple to the horse". Semantics assumes that the dative form is being used (i.e. in the case of 'give' the first complement is the receiver of the second complement)

Operator Application:

Implementation is very similar to the implementation of syntax snips. Slight variations are the fact that adding things back on to the sem-ar-set is much simpler than in the syntactic case and that some productions have to be mirrored for a property snip.

Operator Reconsider:

Upon application. Termination of semsnip operators is the same as in syntactic snips except an additional termination production is required for the semsnip of a property relation.

Back to the operator hierarchy.

This page written by Mark H. Smith, April 1997.