How we understand speech in real time as we hear it
What is it about?
People segment speech into smallish chunks as they make sense of it while listening. This is similar to many other sensory processes. The chunks they come up with are not traditional units of analysis in grammar, neither are they just individual words. The paper presents a model that was developed to account for these chunks. It argues that listeners by and large agree on these and that they can be modelled by a grammar that is not hierarchical but is built on the units that emerge from listeners' perceptions. It can also be applied to written text.
Why is it important?
Fir the first time, we have a model of language which accounts for authentic speech perception and connects it to grammatical description. Grammars are mostly based on written language. Looking at how people process speech in real time makes the model also testable: we can put it to test by experimental methods, which indeed have already confirmed its cognitive validity.