EPSRC logo

Details of Grant 

EPSRC Reference: EP/C546830/1
Title: Prediction in Human Parsing: Towards a Broad-coverage, Crosslinguistic Model
Principal Investigator: Keller, Professor F
Other Investigators:
Researcher Co-Investigators:
Project Partners:
Department: Sch of Informatics
Organisation: University of Edinburgh
Scheme: First Grant Scheme Pre-FEC
Starts: 01 May 2005 Ends: 30 November 2008 Value (£): 121,802
EPSRC Research Topic Classifications:
Human Communication in ICT
EPSRC Industrial Sector Classifications:
Information Technologies
Related Grants:
Panel History:  
Summary on Grant Application Form
When humans read text (or hear speech), they extract the meaning of what they read or hear incrementally. This means that they do not wait until the end of a sentence before they figure out what the sentence means; rather, they try to guess its meaning on a word-by-word basis. Recently, researchers have discovered that humans not only guess the meaning of what they have read or heard so far, but they also predict the words they are likely to hear next. Evidence for prediction in language processing comes from so-called visual world experiments. In these experiments, participants listen to sentences such as (1):(1) The rabbit will eat the carrot.At the same time, they see a picture on a computer screen that depicts a set of objects related to the sentence; in this example, a rabbit, a fox, and a carrot would be displayed. While the participants listen to the sentence, their eyemovements are recorded using an eye-tracker. This is a device that precisely measures where a person is looking and for how long. For sentences like (1), researchers have found that participants look at the carrot when they hear the word eat , which means that they predict the meaning of the sentence even before they hear the word carrot .There is a growing body of experimental work on prediction in language processing. However, a major challenge remains. Researchers still lack a good theoretical understanding of what kinds of predictions people make. This project will contribute to this understanding by developing detailed theories of prediction, and by testing them against experimental evidence. We know that the prediction process is extremely complicated, since it depends on the relationships between words that the listener has already heard, as well as the listener's real-world knowledge. In order to test such a complicated theory, we need to build a computer program that simulates the predictions a human would make. In the process of building programs whose predictions are closer and closer to those of humans, we come to a better understanding of how people comprehend language moment by moment.The work in this project will consists of four main parts. First, mathematical studies will be carried out to better understand how humans generate predictions and then test them. In particular, we will try to work out why some sentences are more difficult to understand than others; our hypothesis is that this has to do with the kinds of predictions involved.Secondly, we will implement a computer program that can process sentences using prediction. This program will be a large-scale model, which means that it will be designed to work on realistic texts drawn, say, from a daily newspaper. Most simulations that have been attempted so far only work on a very restricted subset of English.The third step is then to evaluate our model by comparing the output of simulation to eye-tracking data obtained from experimental participants. We will test the model against existing data sets (collections of text annotated with eye-tracking data), but also against new experimental data to be collected in this project.The fourth step is then to extend our simulations to languages other than English; this is an area in which there has been very little research so far. We will modify our model so that it can cope with German and French text, and then test it on eye-tracking data sets for these languages. German and French are interesting test cases, as the structures of these languages differ in important ways from the structure of English.
Key Findings
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
Potential use in non-academic contexts
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
Description This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
Date Materialised
Sectors submitted by the Researcher
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
Project URL:  
Further Information:  
Organisation Website: http://www.ed.ac.uk