How do Humans parse sentences?

Human sentence parsing continues to be the subject of active research after decades of research. Psycholinguists have found that Working Memory (WM) is a critical resource in the syntactic parsing of sentences ([1]). They have also found that there are significant differences between individuals with reference to their WM spans ([2], [1], [3], [4], [5]).

A number of alternate models have been proposed to predict which sentences will be perceived to be highly complex; the most influential ones include the following:

• Functional Completeness/ segmentation models ([6], [7], [8], [9], [10], [11], [12]),
• Locality Theory ([13]),
• Kimball's Seven Principles ([14]),
• Two-stage Parsing([15]),
• Perspective Shifts ([16])

The underlying logic in sentence processing models is that sentences are read sequentially; the initial part of a clause must be kept in memory while processing the later part of the clause - and both ends must be in memory while assimilating the clause. Experiments have also shown that items stored in short-term Working Memory become inaccessible after a short interval. Experiments also show that incoming items can interfere with items already in Working Memory if they share some features. Clearly, if Working Memory span is limited, the contents of the initial part of the clause may have been lost (or interfered with) by the time the end of the clause is reached.

References