Syntactic category ambiguity
According to the syntactic parsing approach discussed in the preceding section, the syntactic processor cumulatively links words into a syntactic tree structure as they are encountered, as has been illustrated in earlier examples. For this to happen successfully, the processor needs to know what kind of word it is dealing with at each stage. However, a large proportion of words in all languages are ambiguous. Many involve category ambiguity, which is when the same word-form may represent more than one syntactic category. For instance walk can be either a noun or a verb. If it is not clear what the syntactic category of a word is, then how does the parser include it in the tree under construction? Usually, but not always, the appropriate type of word will be apparent from the structure of the tree that has been constructed up to that point. So after x went for a very long…, the noun reading of al would be acceptable, but not the verb reading.
However, there are cases where the partial tree structure would allow words from more than one category. For example, if the word a s is encountered in the fragment in (10.48) it could be either a verb as shown by the continuation in (10.49) or a noun (10.50).
Because of parsing operations like Late Closure, Minimal Attachment or Right Association, the parser may prefer one structure – and therefore one syntactic category for the ambiguous word – over another. Sometimes, however, these parsing strategies do not lead to a structural preference for one analysis. For such cases, one solution that has been proposed is that the processor puts off attaching any additional words into the current syntactic tree until further words in the input clarify the category ambiguity. In reading experiments, such a delay should result in faster reading during ambiguous portions of the sentence, since syntactic attachments are not being made at that point, and slower reading when the disambiguating information is encountered.
Such a result was found in an eye-tracking task with sentences such as (10.49) and (10.50), which were compared with sentences (10.51) and (10.52) where the phrase desert trains is disambiguated by the singular or plural specification in the determiner (this or these) (Frazier Rayner, 1987). Reading times for desert trains (and equivalent sequences in other sentences of this type) were longer in this region for the sentences in (10.51) and (10.52), where the syntactic category of the ambiguous words is clear and where syntactic processing can therefore proceed. By contrast, reading times in the region following desert trains were longer in (10.49) and (10.50). It is precisely in this region that the ambiguity is resolved.
Alternative accounts of category ambiguity resolution include approaches that allow multiple analyses to take place (see the discussion of constraint based approaches in Chapter 11). That is, rather than delaying processing because of category ambiguity, the processor builds different syntactic structures immediately and in parallel, allowing for multiple syntactic analyses of the ambiguous words. Selection from amongst these structures can depend on a number of factors, including the relative frequency with which the different uses of a word are encountered (MacDonald, 1994).
