One of the most annoying problems faced by people who phone a call centre is when the interactive voice response (IVR) system that handles the call misunderstands what the caller has said, leading them down the wrong path.
IVR systems rely on speech recognition. But the scope of speech technology deployed with an IVR system remains limited, as it generally works by recognising key words in a conversation. Researchers at the University of Edinburgh have been looking for ways to improve the way computers interpret human communication.
The focus of the research being conducted by the Human Communication Research Centre (HCRC) at Edinburgh is dialogue systems. These could allow a voice-driven application such as a call centre menu to react more flexibly to unexpected input by the user.
Potential uses of this technology include voice interaction with devices, training, customer relationship management and customer support. A dialogue system could also drive interactive entertainment applications such as life-like computer characters for gaming and virtual reality environments.
Johanna Moore, the professor leading the research, said, "We are working on technology that has a natural language capability and can learn as it goes along."
According to Moore, the challenge in developing a dialogue system is to move away from the prescriptive voice interface, as used in today's IVR systems, to a more flexible approach, capable of reacting to unexpected voice commands.
An example of this flexibility would be in detecting when someone is unsure what response is required. Moore said the systems in development at HCRC would be able to tell from voice characteristics such as hesitancy that more explicit confirmation is required.
"We are also trying to make systems more adaptive," Moore said. "For instance, people habitually speak using similar terminology throughout the conversation. The system would be able to align its use of language with that of the user."
The HCRC has also been looking at improving language technology when applied to the way people search for information. Moore said the HCRC has been developing computational techniques to extract information from text, sustain a dialogue between a human and a computer and generate texts that are personalised for individual readers. Such techniques could be used on internet search engines.
"Services such as Google are good but they could be so much better, as you do not get answers in the way you want them packaged," Moore said. "We want applications that go out and find information and package it for you. We have had a lot of interest in this area, particularly from companies that offer general goods."
The aim of another project is to build and evaluate a simple, adaptable language system that could generate textual dialogue incorporating subtle linguistic features that create the impression of a personality. The goal is for the linguistic personalities to be clearly identifiable, and for the interaction between them to be believable and engaging for both researchers and the public. The user of such a system should not need to adapt their behaviour.
"Human communication is all about gestures, intonational patterns and body language, as well as what is being said," Moore said. "A public information kiosk with a computerised agent that can point and gesture would be easier to understand than simple text."
The dialogue systems being developed by HCRC could enable new types of interactions that would be useful in business environments. The HCRC has been talking to call centre operators and call centre software companies about using its technology, Moore said.
However, she stressed that implementing improved dialogue technology is not about replacing call centre staff, but ensuring they only have to deal with complex calls.
"The technology could possibly integrate into a lot of the conversation flow software that is being used already so that certain aspects of the call can be automated," she said. "It is about using people's skills in the right way, not cutting jobs."
The Human Communication Research Centre
The work of the University of Edinburgh's Human Communication Research Centre is focused on improving the understanding of human communicative processes, and developing computer applications based on this knowledge. It has showcased its research before an audience drawn from the Scottish business and IT community at a technical briefing organised by Connect Scotland.
The HCRC has secured £6m of funding from Scottish Enterprise to develop a new research programme with the Centre for the Study of Language and Information at Stanford University in the US.
Connect Scotland is a non-profit-making organisation founded in 1996 to nurture the creation, development and growth of emerging technology companies. It provides a nationwide support infrastructure which brings together universities, venture capitalists, banks and technology experts, as well as corporates, local enterprise companies, lawyers and individuals with specific management or sector experience.
A spinout of the University of Edinburgh, Connect Scotland is supported by all 14 of Scotland's universities.
This was first published in October 2003