Gradience in natural language denotes the existence of degrees of acceptability of an utterance, i.e. the existence of an acceptability gradient in human language processing. This basically means that an ill-formed utterance can be considered as more or less acceptable. For instance, the following sentences are increasingly unacceptable:

(1) I am the singer of my band.

is perfectly correct and acceptable, whereas

(2) I are the singer of my band.*

is incorrect and slightly unacceptable, while

(3) Me are the singer of my band.*

is more unacceptable, and

(4) Me are singer the me's band of.**

is even more unacceptable.

My work aims to develop a computational model for gradience, and to show that it can be used as a prediction model for acceptability judgement, by confronting its predicting power with empirical results obtained for human judgement. In other words, the goal is to develop a model capable of computing degrees of acceptability, which will have to match empirical results obtained for experiments about human judgement. Such experiments have been carried out by linguists independently from the present work for both French and English.

The model that I have been developping in this regard relies on two distinct but compatible approaches to language modelling, which are the Construction Grammar (CxG) and the Property Grammars (PG). When brought together these two frameworks provide the environment along with the computational tools necessary to compute a figure for the degree of acceptability of an utterance. The machinery involved in calculating such a figure is based on an appropriate account of the different linguistic properties, which are satisfied or violated by the input to be assessed.

In this talk I will present an overview of how the Platypus system works and I will show why it provides a suitable environment for computing degrees of acceptability of an utterance.

Please send comments or queries about this web site to jpprost@ics.mq.edu.au