Thomas R. Kovacs

Critical Issues Using Brain-Computer Interfaces for Augmentative and Alternative Communication

Brain-computer interfaces (BCIs) may potentially be of significant practical value to patients in advanced stages of amyotrophic lateral sclerosis and locked-in syndrome for whom conventional augmentative and alternative communication (AAC) systems, which require some measure of consistent voluntary muscle control, are not satisfactory options. However, BCIs have primarily been used for communication in laboratory research settings. This article discusses 4 critical issues that should be addressed as BCIs are translated out of laboratory settings to become fully functional BCI/AAC systems that may be implemented clinically. These issues include (1) identification of primary, secondary, and tertiary system features; (2) integrating BCI/AAC systems in the World Health Organizations International Classification of Functioning, Disability and Health framework; (3) implementing language-based assessment and intervention; and (4) performance measurement. A clinical demonstration project is presented as an example of research beginning to address these critical issues.

Reliability of Brain-Computer Interface Language Sample Transcription Procedures

We tested the reliability of transcribing language samples of daily brain-computer interface (BCI) communication recorded as language activity monitoring (LAM) logfiles. This study determined interrater reliability and interjudge agreement for transcription of communication of veterans with amyotrophic lateral sclerosis using a P300-based BCI as an augmentative and alternative communication (AAC) system. KeyLAM software recorded logfiles in a universal logfile format during use of BCI-controlled email and word processing applications. These logfiles were encrypted and sent to our laboratory for decryption, transcription, and analysis. The study reports reliability results on transcription of 345 daily logfile samples. The procedure was found to be accurate across transcribers/raters. Frequency of agreement ratios of 97.6% for total number of words and 93.5% for total utterances were found as measures of interrater reliability. Interjudge agreement was 100% for both measures. The results indicated that transcribing language samples using LAM data is highly reliable and the fidelity of the process can be maintained. LAM data supported the transcription of a large number of samples that could not have been completed using audio and video recordings of AAC speakers. This demonstrated efficiency of LAM tools to measure language performance benefits to BCI research and clinical communities.

A Tutorial on Reliability Testing in AAC Language Sample Transcription and Analysis

Abstract Establishing reliability is an essential step in language sample transcription and analysis. This tutorial provides an illustration of replicable procedures for reliability testing during transcription and analysis of language samples generated by people who use augmentative and alternative communication (AAC) systems. Statistical measures used for testing agreement between raters coding categorical data are summarized. Detailed procedures for reliability testing in AAC language sample transcription and analysis are provided, beginning with the collection of raw language sample data. Procedures include guidelines for (a) establishing inter-judge agreement during the transcription process, and (b) using Cohens kappa to establish inter-rater reliability during deeper analysis of transcribed utterances. All procedures are demonstrated in a case example using language samples from children who use AAC.

Language Samples From Children Who Use Speech-Generating Devices: Making Sense of Small Samples and Utterance Length

Purpose Mean length of utterance in morphemes (MLUm) is underreported in people who use augmentative and alternative communication (AAC). MLUm is difficult to measure in people who use AAC because of 2 challenges described in literature: the challenge of small language samples (difficulty collecting representative samples) and the challenge of transcribing short utterances (difficulty transcribing 1-morpheme utterances). We tested solutions to both challenges in a corpus of language samples from children who use speech-generating devices. Method The first challenge was addressed by adjusting the length of the sampling window to obtain representative language samples. The second challenge was addressed by using mean syntactic length (MSL) as an alternative to MLUm. Results A 24-hour sample window consistently failed to yield representative samples. An extended 1-month sample window consistently yielded representative samples. A significant positive prediction of MLUm by MSL was found in a normative sample. Observed measures of MSL were used to predict MLUm in representative language samples from children who use AAC. Conclusions Valid measures of utterance length in people who use AAC can be obtained using extended sampling windows and MSL. Research is needed to characterize the strengths and limitations of both solutions.