To prevent confusion in a further direction, it is important to distinguish ambiguity from ambivalence. For our present purpose, ambivalence denotes a phenomenon of uncertainty as to which annotation should apply to a given sentence, not because the sentence itself is ambiguous in this respect, but because the semantics of the annotation scheme is inexplicit (see the section on documentation). For example, in the sentence She's older than her sister, it may be unclear whether than her sister is to be treated as a prepositional phrase or as a comparative clause (or perhaps as some other constituent). Ideally, it may be argued, an annotation scheme should be explicit enough to solve all problems of this kind, by stipulating a single correct analysis, however minor or rare the syntactic phenomena that may occur. In practice, however, the ideal is impossible to achieve, or at least can be achieved only by a long process of narrowing down areas of uncertainty.
An alternative position is to argue that in many areas of syntax it is unrealistic to decide on a single correct analysis: such a position can only be maintained by making arbitrary, artificial decisions. More realistic is the viewpoint that the boundaries of syntactic categories are by the nature unclear, that ambivalence is an inescapable phenomenon, and that 100% consistency in applying a parsing scheme can never be achieved.
Whichever of the two positions above is adopted, the fact remains that ambivalence is an endemic phenomenon to be coped with in corpus annotation. The way to handle ambivalence is not to include all alternative analyses in the annotated corpus, but to reduce areas of ambivalence and inconsistency as far as possible, by providing an explicit annotation scheme.