To understand a sentence we need to assign roles to each word in it, i.e. roughly speaking we need to understand who does what to whom (although more recently some scholars have argued that semantics comes first).
If in analytic languages word order is the clue, in synthetic languages morphological characteristics of words are more important. There are, again, different parsing models:
- autonomous (at the initial stage of processing syntactic information is enough to assign roles to words) and
- interactive (syntactic processing is influenced by semantic and other information),
as well as
-
one-stage (syntactic and semantic data are used simultaneously to assign roles to words) and
-
two-stage models (syntactic data first, semantic information afterwards).
Among the most popular models of parsing (also used in computational linguistics to imitate language processing) are:
- a serial autonomous model,
- a parallel autonomous model,
- the garden path model, and
- constraint-based models of parsing.
Arguments are, again, controversial and none of the models gives answers to all questions; it is possible that we use different strategies depending on sentence structure, presence or absence of ambiguity and other factors; it seems though that in general simpler paths are preferred and that verbs normally play central role in parsing.