next up previous contents
Next: Issues in practical application Up: Optional annotations Previous: Syntactic/semantic (functional) annotation



Annotation of deep/logical information  

Under this heading fall a number of phenomena that can be collectively referred to as discontinuity and trace phenomena, i.e. movement phenomena where an element has been moved (or removed) from its `logical' position. In 63 is an example in which a relative clause has been moved from its logical position in an NP:

(63)  A man was at the door who wanted to speak to you
Here, the relative clause who wanted to speak to you is moved to the end of the sentence. An index system may be used to indicate the position that the relative clause `logically' occupies. This is shown in 64 (where a number of details have been left out for the sake of clarity):

(64)  [NP A man [*1] NP] [VP was [at the door] VP] [CL-Rel*1 who wanted to speak to you CL-Rel]
Indices may also be used for zero constituents, such as the zero relative pronoun in the case of some relative clauses. There may also be a need to include such traces in syntactic annotation even where no discontinuity occurs. A classic example is found in English when a preposition occurs in a final position, without a following prepositional complement. In 65, the trace constituent [*1] is indexed to show the discontinuity of the prepositional phrase:

(65)  [S [NP Who NP*1] [Aux does Aux] [NP she NP] [VP work [PP for [*1] PP] VP] ? S]
In 66, a superficially similar construction, there is no discontinuity, but the missing prepositional complement is identified by the unindexed trace constituent [*0]:

(66)  [S [NP He NP] [VP has [NP nothing [VP to live [PP for [*0] PP] VP] NP] VP] . S]
This zero constituent may also be used to indicate ellipsis, where no movement phenomenon occurs.

Indices can also be used to indicate `underlying' Subject or Object postions in the complement of so-called raising constructions and with control verbs. In 67 James is the Subject of tried but logically, it is also the Subject of to go:

(67)  [NP James NP*1] [VP tried [ [CL [*1] to go CL] VP]
In 68 John is the Object of believed and logically, also the Subject of be:

(68)  [NP Mary NP] [VP believed [NP John NP*1] [CL [*1] to be at home CL] VP]
Using indexes as shown here, these logical relations may be made explicit.

Languages differ in the range of possibilities of discontinuity, raising constructions and control phenomena (e.g. in Raising-to-Object constructions, the Dutch equivalent of believe can only take tensed complements, not infinitival ones as in English). However, each language will probably have a number of constructions in which logical relations need to be indicated using an index system. It is beyond the scope of this report to go into further detail, but the ideas set forth here should enable a simple indexing system to be implemented. Details of two actual applications of such systems can be found in the documentation of the SUSANNE corpus (Sampson 1995, Ch. 5) and the Penn Treebank (Marcus et al. 1994).

next up previous contents
Next: Issues in practical application Up: Optional annotations Previous: Syntactic/semantic (functional) annotation