In the Autumn of 1989 I came back from my industrial training year at Eurostat in Luxemburg with a view that speech technology was going to be the next big thing. After completing my degree, I set out for Edinburgh the following year to join a new MSc in Speech and Language Processing - with some apprehension. A year seemed like a long time. What if, in the meantime, the smart crowd in Luxemburg had implemented multilingual voice activated database access and even dissuaded Systran from advising drivers to attach children to the back of their cars? Remember, this was the high water mark of Britain’s Alvey Program and everything seemed possible. I needn’t have worried. As any one who has done a course with Ronnie Cann knows : semantics and pragmatics are rather hard. Sixteen years on, there is still a big pot of gold at the end the rainbow.

Speech technology, natural language processing and AI are inextricably linked. They share common components (formal logics, stochastic method and intuitive models), deep division between participants and a linked propensity to get funded. By the time I had finished my PhD at the University of Leeds, the funding tide had turned, attention had shifted and the herd migrated in search of new funding friendly pastures. I was reminded of this the other week when during the MX Alliances’s Mobilizing for Profit event at the Adam Smith College. In response to a question, one of the speakers joked about the potential problems for automatic speech recognition (ASR) of West of Scotland accents. Although it may come as a surprise to many on the east coast, ASR probably wouldn’t have any particular difficulty with west coast speech. The problem is lack of training data. ASR needs a lot of it and marking it up requires skill and effort. SCRIBE (Spoken Corpus of British English) was set up to provide this. Unfortunately, most of it never got marked up because the funding stopped after the first year. It’s a fairly safe bet that, if funding had continued, ASR would probably work a whole lot better. My point is that useful research with the potential to deliver long term customer value was dragged down by unrealistic wider expectations.

Martyn Warwick’s recent TelecomTV article puts Web 3.0 into a similar frame. Broadly, Chris and I would agree with Martyn’s anonymous chum. However, leaving aside the ludicrous notion that semantics can be delivered in an XML wrapper, there are aspects of the Web 3.0 vision (aka Semantic Web) that can be delivered in a reasonable timescale. The key is to target identified needs with decently engineered technology that has a sound theoretical basis. Calico Jack has been working on that for four years.