Trip Report: RFB Symposium on Math and Science Access

This trip report covers the RFB Symposium on Math and Science Access (May 12--13, 1994) held at the RFB headquarters in Princeton.

May 12: Pre-Lunch Session

I described the design and implementation of AsTeR in a one-hour talk ( slides: dvi, slides: SliTeX) followed by a half-hour discussion on some of the design decisions and implementation issues. This was followed by a one-hour panel discussion on future directions. The panel discussion set the theme for the rest of the symposium. The issues that were identified for further discussion were:

Obtaining electronic encodings of documents from publishers in a form that lends itself for appropriate processing.
- What kind of SGML DTD's should be used?
- Is the current Math DTD (ISO 1208.3) appropriate?
- How should the above DTD be modified/extended?
What kinds of internal representations are needed by systems like AsTeR, and how can we derive them from particular document encodings?
How do we use such internal representations to display the information in a manner appropriate for different modalities?
What is involved in making AsTeR more widely available?
- Installing AsTeR at some of the RFB sites.
- AsTeR typically needs to be extended to recognize and render specialized notations used in particular documents. Though this is straight-forward, there is a need for people other than me to have done this.
- Porting to different platforms.
How do we extend the user-interface for systems like AsTeR? Goals:
- Provide a consistent interface across platforms and modes of interaction.
- Interact using multiple modalities, i.e. use both speech and visual output, or speech and enlarged output etc. How do we effectively synchronize the different displays?

May 12: Post-Lunch Session

Following lunch, the participants broke up into different working groups to discuss the issues identified above. There were three working groups:

Porting.
Document encoding and representation.
User-interface design.

The working groups met for three hours and prepared a list of recommendations outlining future work. Since I felt I belonged in all of the groups, I spent an hour with each group.

Porting

Prof. Krishnamoorthy (RPI) proposed to design an AsTeR server that would be accessible over the Internet. This would be a UNIX workstation running the current implementation of AsTeR, and would allow clients to send TeX/LaTeX documents to be rendered in audio, as well as provide a library of documents that a client could read. The output would be sent back to the client in the form of a text file with embedded Dectalk control codes. This would provide immediate access to the system to a user with Internet access and a serial-line Dectalk. (It would be trivial to make all this work with the software Dectalk.) The next stage would be to develop a protocol for sending across commands to generate non-speech audio.

Bill Barry (Oregon State) agreed to start working on porting AsTeR to other Lisps, and eventually perhaps to C/C++. Bill and others at Oregon will probably work on developing more documentation for the entire system. (At present, the only thing available is my thesis.) We all agreed that porting to the DOS platform would not happen immediately since it involved too much work.

Rich Cox (RFB Board Member and ATT Speech Group) outlined RFB's short-term plans to install AsTeR at RFB and use the system to produce audio-tapes of technical documents. This involves RFB acquiring and installing a UNIX workstation (either DEC 5000 or SPARC ) workstation running AsTeR under Lucid Common Lisp. The RFB programmers would learn to write rendering rules in AFL in order to be able to extend AsTeR as appropriate for recording specific books.

Document encoding and representation

We discussed extensions to the SGML MATH DTD (ISO 12083) in terms of putting more semantic information into the document encoding. The current specification only allows for Math to be encoded using pure visual layout operators. Though it is impossible to develop a general DTD that captures mathematical semantics, we wanted to introduce a framework that would allow an author to introduce his/her own abstractions to the encoding, i.e. analogous to defining macros in TeX or LaTeX. Having such user-defined elements in the DTD would allow the document to encode mathematics by using visual layout operators by default, but also enable encoding of semantics wherever such information is available. This would make such document encodings easier to maintain and process.

As for the internal representation, we decided to go with the what is currently used in AsTeR, i.e. an attributed tree structure for document content and the quasi-prefix form for mathematics.

User Interfaces

The group discussed long-term goals such as how different forms of interactive output could be synchronized. No clear design strategy emerged from these discussions.

May 13: Pre-lunch

RFB Tour

We received a half-hour tour of the RFB facilities. I was extremely impressed by the setup. We visited the recording studios where the books are taped. A trained reader records the books in a sound-proof booth while a monitor listens to the recording at a station outside. The monitor verifies that the reading is unambiguous, tone-indexes the recording (inserting low-pitched tones at page breaks etc.) and generally makes sure that everything is working.

We were also given a tour of the master-tape library, where RFB catalogues and stores its collection (currently over 80,000 titles). They have a conveyor-belt system that allows their staff to pull out a specific master-tape automatically, and have it moved to the copying stations. The copying stations can copy upto four cassettes in one minute, and the copies are then sent out after doing a spot-check to verify the copies.

Working Groups

The remaining time until lunch was taken up by each working group presenting its report. We had some useful discussions. The final reports from each working group will be made available in a couple of weeks.

May 13: Post-Lunch

We concluded the workshop by discussing models for future collaboration.

raman@crl.dec.com

Last modified: Mon May 16 13:54:55 1994