Translating the World Wide Web interface into speech
C. Reeves 1, M. Zajicek 2, C. Powell 2and J. Griffiths 2
1IT Services Development,
Royal National Institute for the Blind, 224 Great Portland Street,
London, WIN 6AA, UK.
2The Speech Project,
School of Computing and Mathematical Sciences, Oxford Brookes University,
Oxford OX3 OBP, UK.
ABSTRACT
BrookesTalk, a prototype browser, uses information retrieval to provide a set of complementary options to summarise the Web page. The aim is to enable visually impaired users to effectively browse and use the World Wide Web alone or with sighted co-workers; using either a visual browser, specific items of large text or speech output. Some initial studies and future areas of work are discussed.
KEYWORDS
World Wide Web, browser, visually impaired, information retrieval, HTML, usability
INTRODUCTION
BrookesTalk, a speech output browser using Microsoft speech technology, offers a range of navigation functions for the World Wide Web. These include a list of headings, links, keywords, an abridged version of the page and page summary. It is expected that the user will pick tools which complement one another for the particular type of page under review. Our hypothesis is that improved provision of summary information will increase orientation, navigation and general usability of the World Wide Web for visually impaired users.
This paper briefly describes the essential functionality behind BrookesTalk and then discusses preliminary evaluation of the usability of BrookesTalk together with future areas of work.
HOW BROOKESTALK WORKS
Keywords -The list of extracted keywords consists of words which are assumed to be particularly meaningful within the text (Luhn, 1958).These are found using standard information retrieval techniques based on word frequency (Zajicek and Powell, 1997).
Abridged text - The technique is based on 'Word level n-gram analysis' in automatic document summarisation (Rose and Wyard, 1997). Extraction of three word key phrases, or trigrams, preserves some word position information and then creates a page consisting of the sentences in which the trigrams appeared.Abridged pages on average worked out to be 20% of the size of the original text and, unlike keyword lists, are composed of well formed, comprehensible sentences.
WHAT THE USERS THOUGHT
Keywords -Preliminary experiments with twenty subjects showed that headings and keywords were judged to be roughly comparable in their usefulness as page content indicators; both however were significantly more useful than anchors/links.
Page summary - Perceived by users to be an important tool in Web page orientation. The BrookesTalk summary comprises of title, author name, author defined keywords,
number of words in the page, headings, links and extracted keywords. The number of words in a page was found to be particularly useful for page orientation.
BrookesTalk summary facilities become useful for visually impaired users as a result of the variability in the use of HTML code by Web page authors which can impede standard orientation methods.
General feedback on BrookesTalk was also obtained from a group of visually impaired users including those at the Royal National Institute for the Blind (RNIB). Aspects such as the need for operational simplicity, a small program 'footprint', high flexibility / configurability, usable by sighted co-workers were all deemed important factors and will influence future studies. Further areas of functionality were highlighted, such as a flexible scratchpad to record sections of Web pages for later, integrated retrieval. There are also potential benefits of using the summary functions with other, non Web, formatted documents used during work and leisure.
THE WAY FORWARD WITH BROOKESTALK
It has become apparent that users have varied approaches to the use of BrookesTalk and its functionality. We will attempt to observe and fully understand different conceptual and navigational techniques used both within and between visually impaired and sighted people. We are currently undertaking user mapping to identify the main user groups and stakeholders, which will form the basis for future usability studies. It is likely that a large scale trial will involve users performing tightly controlled tasks on specially chosen types of Web pages. We will seek to establish whether different summarisations are required for different Web page types, the level of activity required to gain a clear 'picture' of a Web page, and the level of enhanced usability achieved by providing a closely linked combination of visual and speech output.
Similarly, on the functionality side, we plan to extend page summarisation to include
