The aim of this thesis was to answer two main questions concerning language comprehension:- What strategies do readers use when an ambiguity is encountered? - What kind of information do they use in such a case?A particular attention was granted to Relative Clause (RC) attachment preferences, for which no satisfactory model has been presented so far in the literature, especially if one takes into account all the cross-linguistic data that exists for this type of construction. A series of experiments was thus conducted in order to determine which factors are responsible for attachment decisions in the case of ambiguous RCs. Eye movement recording was used in order to track down the time course of processing. Experiments 1-4 suggested that constituent length influences attachment decisions. In order to account for this result, we hypothesised that some decisions can be delayed, or left under-specified, until the disambiguating information becomes available. We implemented a simulation model based on this idea and were able to account for our results as well as to make new predictions. These predictions were verified in Experiment 5. The idea syntactic decisions can be delayed is accepted in Frazier and Clifton's (1996) Construal theory. However, unlike our model, the Construal theory states that only one class of syntactic relations, namely the non-primary relations can be under-specified. As for primary relations, they are assumed to immediately receive a totally specified description. Experiments 6-11 suggested that the primary/non-primary distinction is probably irrelevant and that under-specification applies to both types of constructions, in line with our model.
