Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.430258
Title: Automated assessment of handwritten scripts
Author: Allan, Jonathan
ISNI:       0000 0001 3415 0195
Awarding Body: Nottingham Trent University
Current Institution: Nottingham Trent University
Date of Award: 2005
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Abstract:
In this thesis, the automatic assessment of handwritten responses to formal exam questions is introduced as a novel research area. This work highlights difficulty of recognising children's handwriting since the handwritten samples used are often of a poorer quality compared to that given by an adult. The work also shows that the errors that are introduced through recognition of the handwriting can be overcome by using a Specific Word Assessment Technique (SWAT). This technique utilises the nature of the assessment medium to concentrate on scoring responses according to how well the handwritten images match to the actual correct answer. This is in direct contrast to a Conventional Lexical Approach (CLA), which is required to match the handwritten image against all possible answers. In the CLA, the automatic assessment is reliant upon the handwriting recognition stage producing a perfect reconstruction of the written responses before being compared to the model answer. The performance of both the CLA and the novel SWAT is evaluated when each method is employed to assess a number of different question response styles. In the first instance, a preliminary investigation is carried out using the CLA in order to determine the practicality of automatically assessing highly constrained adult handwritten responses. SWAT is then introduced as an alternative method for automatically assessing children's single word handwritten responses and a retrospective experiment is then carried out, employing SWAT to automatically assess the adult's handwritten responses from the preliminary investigation. The thesis will show that the generalised CLA is not robust enough to be able to cope with the errors introduced at the recognition stage and therefore the overall automatic assessment system incurs a large inaccuracy. This is shown to its fullest extent when the CLA is employed to automatically assess children's handwritten sentence responses. The CLA automatically assessed 8 8% of all the responses, but this was at the expense of an overall assessment accuracy of only 37%. The use of a questions history is also exploited in order to give greater assessment accuracy. It is used to help assess the recognised responses from both of the two methods and was compared to baseline results where the history has not been used. Results show that the SWAT with History (SWATH) has the better performance with an overall assessment accuracy of 100%. The high accuracy has been achieved at the expense of the total number of responses assessed, 33%. The approach was not sufficiently confident of 67% of the responses to automatically assess them however the system was able to automatically set aside the responses for human intervention. The work in this thesis illustrates the potential for automatically assessing handwritten responses using current handwriting recognition systems and provides a basis for future research in the area of automatic assessment of handwritten scripts.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.430258  DOI: Not available
Share: