Language Software - Human Language Technologies

Human Language Technologies

Human Language Technologies (HLT) comprise a number of areas of research and development that focus on the use of technology to facilitate communication in a multilingual information society. Human language technologies are areas of activity in departments of the European Commission that were formerly grouped under the heading Language Engineering (Gupta & Schulze 2011: Section 1.1).

The subset of HLT that is of greatest interest to the language teacher is Natural Language Processing (NLP), especially the areas of speech synthesis, speech recognition and parsing.

Speech synthesis has improved immeasurably in recent years. It is often used in electronic dictionaries to enable learners to find out how words are pronounced. At word level, speech synthesis is quite effective, the artificial voice often closely resembling a human voice. At phrase level and sentence level, however, there are often problems of intonation, resulting in speech production that sounds unnatural even though it may be intelligible. Speech synthesis as embodied in Text To Speech (TTS) applications is invaluable as a tool for unsighted or partially sighted people. Gupta & Schulze (2010: Section 4.1) list several examples of speech synthesis applications.

Speech recognition is less advanced than speech synthesis. It has been used in a number of CALL programs, in which it is usually described as Automatic Speech Recognition (ASR). ASR is not easy to implement. Ehsani & Knodt (1998) summarise the core problem as follows:

"Complex cognitive processes account for the human ability to associate acoustic signals with meanings and intentions. For a computer, on the other hand, speech is essentially a series of digital values. However, despite these differences, the core problem of speech recognition is the same for both humans and machines: namely, of finding the best match between a given speech sound and its corresponding word string. Automatic speech recognition technology attempts to simulate and optimize this process computationally."

Programs embodying ASR normally provide a native speaker model that the learner is requested to imitate, but the matching process is not 100% reliable and may result in a learner's perfectly intelligible attempt to pronounce a word or phrase being rejected (Davies 2010: Section 3.4.6 and Section 3.4.7).

Parsing is used in a number of ways in CALL. Gupta & Schulze (2010: Section 5) describe how parsing may be used to analyse sentences, presenting the learner with a tree diagram that labels the constituent parts of speech of a sentence and shows the learner how the sentence is structured.

Parsing is also used in CALL programs to analyse the learner's input and diagnose errors. Davies (2002) writes:

"Discrete error analysis and feedback were a common feature of traditional CALL, and the more sophisticated programs would attempt to analyse the learner's response, pinpoint errors, and branch to help and remedial activities. Error analysis in CALL is, however, a matter of controversy. Practitioners who come into CALL via the disciplines of computational linguistics, e.g. Natural Language Processing (NLP) and Human Language Technologies (HLT), tend to be more optimistic about the potential of error analysis by computer than those who come into CALL via language teaching. An alternative approach is the use of Artificial Intelligence (AI) techniques to parse the learner's response - so-called intelligent CALL (ICALL) - but there is a gulf between those who favour the use of AI to develop CALL programs (Matthews 1994) and, at the other extreme, those who perceive this approach as a threat to humanity (Last 1989:153)".

Underwood (1989) and Heift & Schulze (2007) present a more positive picture of AI.

Research into speech synthesis, speech recognition and parsing and how these areas of NLP can be used in CALL are the main focus of the NLP Special Interest Group within the EUROCALL professional association and the ICALL Special Interest Group within the CALICO professional association. The EUROCALL NLP SIG also maintains a Ning.