User Interface of a Nonvisual Table Navigation Method
Chieko Asakawa, Takashi Itoh · 1999 · CHI '99 Extended Abstracts on Human Factors in Computing Systems · doi:10.1145/632716.632850
Summary
This two-page CHI '99 late-breaking-results extended abstract addresses a problem that was central to the early accessible web: HTML tables were everywhere — used both for tabular data and, at the time, as a layout device — but talking web browsers read them strictly horizontally, cell by cell across rows. A blind user trying to find a column value several rows deep had to count columns, read across, and hold the row-column structure in working memory. The authors, from IBM Tokyo Research Laboratory, propose a nonvisual table-navigation method implemented on top of their Home Page Reader voice browser. The design has three components: a 'table cursor' that the user moves horizontally (numeric keys 4/6) and vertically (2/8) through cells; a 'table pointer' that reads the contents of the cell at the top of the current column or the leftmost cell of the current row without moving the cursor, so users can look up row/column headers from any position; and a 'cell-jumping key' that skips compound cells created by HTML COLSPAN or ROWSPAN attributes, so that irregular ('ungridded') tables can still be traversed row-by-row. The system internally converts ungridded tables into an off-screen gridded representation, and uses two voices (male for content, female for system announcements such as 'M, consisting of two columns') to separate structural cues from cell data.
Key findings
Three blind participants completed three tasks — finding the type of CPU in a spec table, the opening time of an amusement park, and the noontime program of a broadcasting station — with the new method and the prevailing horizontal-read method. The new cursor/pointer/jumping interface cut task time roughly in half on the two gridded tables (Task 1: 16 s vs 29 s; Task 2: 16 s vs 32 s) and rescued Task 3 entirely: the complex TV-listings table with nested COLSPAN/ROWSPAN structure could not be completed correctly under the horizontal method (all participants gave an incorrect answer after 139 s on average), whereas the new method produced correct answers in 55 seconds. Participants reported that ungridded tables, which they had learned to avoid with standard talking browsers, no longer blocked their browsing. The dual-voice announcement of header context (e.g. 'M, second column' spoken in the female voice whenever a new cell is entered) was highlighted as important for orientation, reinforcing a broader design principle that non-visual interfaces benefit from separating structural metadata from primary content by audio channel or voice.
Relevance
This short paper is an early articulation of what eventually became standard practice in screen readers and in the HTML/ARIA table accessibility model: the ability to move a cursor by row or column, to announce row and column headers on demand, and to handle merged cells gracefully. The table cursor + pointer + header-echo pattern described here is essentially the interaction model later adopted by JAWS, NVDA, VoiceOver, and Narrator for data-table navigation, and it anticipates the role of scope, headers, and th elements in HTML and of the grid role in ARIA. For practitioners today, the paper's main lesson is not the specific keypad layout but the insistence that 'a table which is effective as a visual representation can also be accessible and useful to the blind' given adequate user-side interaction — provided authors supply the structural markup that lets assistive technology distinguish headers, cells, and spans. Limitations are those of a late-breaking abstract: tiny sample (n = 3), no long-term study, and no handling of layout tables, nested tables, or author-side guidance.
Tags: web accessibility · screen readers · voice browser · table accessibility · blindness and low vision · non-visual interaction · assistive technology · text-to-speech