Hill, K. J. & Romich, B. A. (2001). A language activity monitor for supporting AAC evidence-based clinical practice. Assistive Technology, 13, 12-22.

A Language Activity Monitor for Supporting AAC Evidence-Based Clinical Practice

Katya J. Hill, Ph.D CCC-slp
Center for Assistive Technology Education and Research
Edinboro University of Pennsylvania
Edinboro, Pennsylvania 16444

Barry A. Romich, P.E.
Prentke Romich Company
1022 Heyl Road
Wooster, Ohio 44691

ABSTRACT
Augmentative and alternative communication (AAC) evidence-based practice requires the collection and analysis of performance data. This paper presents the development, evaluation, and application of automated performance monitoring tools for use in clinical practice. Language activity monitoring (LAM) is the systematic data collection of the actual language activity of an individual who relies on AAC. Work completed to date includes the development and evaluation of the language activity monitor function, which now is commercially available in three forms: 1) a standard feature built into modern high performance AAC systems, 2) an external add-on package for use with older AAC devices based on synthetic speech, and 3) software that allows the PC to serve as a LAM in the clinical environment. The LAM records the time and content of language events (the generation of one or more letters or words). A logging protocol suitable for clinical application has been in use since late 1998. The logged data is uploaded periodically to a computer for editing, analysis, and the generation of a summary measure report. The applications of this work in the areas of clinical service delivery are presented.

BACKGROUND
Evidence-based practice and outcomes measurement are areas of increasing focus for the general assistive technology field (DeRuyter, 1998; Granlund, Blackstone, & Norris, 1996; Logemann, 2000). Both require the collection of data. Specific to the area of augmentative and alternative communication (AAC), little outcomes measurement research of a quantitative nature has been conducted. All stakeholders, however, would value better information (Jutai, Ladak, Schuller, Naumann, and Wright, 1996). Individuals who rely on AAC want to know about their progress and how to improve the effectiveness of their communication. Parents and other family members of individuals who use AAC want to know about progress in communicative competence. The Individuals with Disabilities Education Act Amendments of 1997 (PL 105-17, 1997) includes a requirement of outcomes monitoring. Third party funding sources want to know that their investments are effective and responsible. Speech-language pathologists (SLPs) want to know that their efforts are having the best effect possible and that they are operating in compliance with their code of ethics. Manufacturers of AAC systems want to know how to design systems that are most effective for their customers. Prior to the developments described herein, AAC stakeholders were receiving limited quantitative performance information.

Very few research studies on the actual daily communication of people who rely on AAC exist (Hill and Romich, 1998). In addition, there is insufficient data on an individual’s communication process outside of therapy from the data collection methods currently available. Language sampling analysis provides the most objective and concrete data regarding the language and communication skills of an individual (Paul, 1995). However, only 10% of SLPs report that they use language sampling and observations as part of their standard language assessment battery (Beck, 1996). Conventional methods of monitoring AAC system use are based on personal observation and video or audio recording with subsequent observation, timing, and/or transcription. The cost of this approach is high because of the human time investment and the delay between the collection and analysis of a language sample reduces the usefulness of the data. Consequently, professionals have limited opportunities to collect and analyze data on the actual daily environmental use of AAC systems by consumers. Still, language sampling in the natural setting is considered the best indicator of an augmented communicator’s performance on an AAC system (Light and Binger, 1998).

An obvious solution to this problem is the automation of the language data collection and analysis processes (Romich and Hill, 1999). There have been some efforts made in this area, all integral to specific communication or writing systems. For example, Miller (Miller, Demasco, and Elkins, 1990) proposed a software based system named Meta4 with a feature called “usage tracking”. Ahlsen (Ahlsen and Stromqvist, 1999) presented the possibilities of the ScriptLog computer tool for logging the writing process. The approach of evaluating natural language processing (NLP) techniques in an AAC system using data logged from a user as the primary source for comparison was proposed by Copestake and Flickinger (1999). However, their proposal was primarily interested in using logged data for evaluation purposes, and not as a clinical tool.

Some commercially available AAC devices have included limited features that monitor use, but none has incorporated time information. For example, language content that has left the display of a Liberator can be accessed to allow the clinician to see recent communication. In addition, the Liberator allows communication to be recorded in an area of memory, known as a notebook, that can be uploaded into a computer for clinical analysis. The Dynavox counts the number of times a given key has been selected, but no contextual information or language content is available. These systems do not provide for identification of information such as the language representation method being used, errors generated, or rate of output.

The purpose of this work was the design and development of a tool to facilitate the collection of language samples from people who rely on AAC assistive technology. In support of the design and development processes, a research component consisting of professional and consumer surveys was undertaken to test the feasibility of the design. However, the focus of this article is on the design and development processes.

METHOD
The first design effort for the Language Activity Monitor (LAM) function was directed at developing an external LAM device. This would allow the LAM function to be added to existing AAC devices that have an RS-232c serial port representation of language events (Fig. 1). Thus, the LAM function can quickly be added to individual devices in the large pool of installed AAC systems.

Fig. 1: Diagram showing connection of LAM device to an existing AAC system.

The design was a team effort, including contributions from AAC consumers, AAC service delivery professionals, and a manufacturer. Team consensus was reached that identified the top five features considered design priorities. Table 1 lists the top five features identified by the team.

Table #1: LAM top five design features

Consumer input into the design was extremely valuable from the beginning stages of discussion. In particular, consumer opinions on personal acceptance, usefulness, and effectiveness of the LAM were considered.

Design features of the LAM device that were identified initially included:

· Universal application across many devices (different manufacturers and methods)
· RS-232c input and output
· Sufficient memory for meaningful data logging
· 24 Hour Real Time Clock for time stamp
· Battery power with a charge life of at least one week
· Simple Enable, Disable, Upload, and Erase controls
· Both pushbutton and serial input function commands
· Small size for attachment to AAC devices

LAM Device Technical Development

The LP-3100 low power controller from ZWorld met the technical criteria and eliminated the delay and expense of developing application specific hardware. The LP-3100, which operates on low voltage and low power, has two serial ports, 256K flash memory for program storage, 256K flash memory for data logging, and a real time clock. Twenty-four hour clock data with one second resolution (HH:MM:SS) plus a null character for string termination results in seven bytes of overhead per event recorded. If every language event were a single character (everything is being spelled), this scenario of highest memory use still would allow for the logging of 32,000 characters. Depending on the analysis procedure, a sample size of one hundred utterances is considered large enough for language transcript analysis (Cole, Mills, and Dale, 1989). Pilot studies indicated a typical utterance length averaging on the order of 25 characters, making the minimum sample requirement around 2500 characters. Therefore, the LP-3100 memory size exceeded by more than an order of magnitude the requirement for this application.

Program development was based on the ZWorld Deluxe Dynamic C development tools. C language programming provides for ease of development and maintenance and the potential transport to other hardware. Additional development criteria focused on simplicity of use, such as ease of physical attachment and connecting the LAM to the AAC device, and ease of enabling and disabling the LAM.

The design on which the fabrication of prototype LAM devices was based allowed the LAM device to readily be transferred from one AAC system to another. This facilitated the collection of language samples from an array of AAC devices. The AAC devices needed to be configured such that language events are sent out the serial port at the time they are generated.

LAM Function Built into AAC Systems

Successful pilot studies using the prototype LAM device encouraged the incorporation of the LAM function as a built-in feature in a new AAC system introduced in early 1999. This device was the Axs1600, a pen computer with access features and both Vanguard Minspeak and WiViK software available. It became the first high performance AAC system to include the LAM function. Figure 2 shows the Review Vocabulary Menu screen that includes the LAM functions. The LAM function has become a standard feature of all high performance AAC systems subsequently released by Prentke Romich Company (PRC).

The LAM function should be able to be added as a built-in feature to any high performance AAC device that has a real time clock, available memory, and a serial port or other method of uploading logged data to a computer for no additional manufacturing cost.

Fig. 2: AAC system REVIEW VOCABULARY MENU showing LAM functions

PC-LAM

A further development was undertaken that resulted in software that allows the PC to perform the LAM function. Because the PC must be connected to the AAC system while the language events are being generated, this function is useful in the clinical environment, but not in the natural environment. A window on the computer monitor shows events as they happen.

Protocol Standardization

To facilitate widespread application of language activity monitoring, automated methods of processing the logged data are required (Hill and Romich, 1999). Data periodically is uploaded into a computer for this processing. The uploaded data is preceded by a header that includes a privacy notice and information such as the device sending the logfile, the current software version, etc. The standardization of the data reporting format is necessary for the processing programs to accept data from different sources. A standard protocol ensures data that are: 1) readily interpretable by clinicians using non-technical analysis procedures; and 2) suitable for use by standard language analysis programs.

Subsequent to the start of this initiative, the Rehabilitation Engineering Research Center (RERC) on AAC began the pursuit of automated data logging issues. One of the projects of that center has been the development of a universal logfile standard (Higginbotham, Lesher, and Moulton, 1999). This protocol has the flexibility to provide for the recording of many other events beyond the language activity. These events could include activation of keys or switches, changing of display content, the advance of a scanning indicator, etc. Many research questions could be addressed using such a tool.

The LAM protocol was developed specific to the clinical needs of the AAC practitioner and consumer and focuses on language activity. The protocol design goals were to have as simple a format as possible yet provide for the information needed for clinical use. A time stamp resolution of one second provided sufficient timing accuracy since most summary measures are calculated on the basis of an utterance which is likely to require many seconds to construct. Also, the one second time format is familiar.

Quotations enclosing the content of each event make clear the use of the space character. Thus for a language event, the LAM protocol is:

hh:mm:ss “Any continuous text that is transmitted by the AAC device.”

where hh:mm:ss represents the time of day in hours, minutes, and seconds using the 24 hour clock format. Characters following the hh:mm:ss and one space, inside quotation marks, are the characters that were selected on the AAC system in a continuous sequence with time between characters less than 0.2 seconds. For a non-language event, the protocol is:

hh:mm:ss “*[NON-LANGUAGE INFORMATION IN CONTINUOUS TEXT]*”

For example, consider the individual using an AAC device who is thirsty at a loud party. If he or she is using a language representation method that can access the series of individual core vocabulary words and phrases "I need ", "something ", "to drink " (spaces included) and then spells "immediately " using word prediction, starting at exactly 8:37 PM, then the representation for that series output would be:

20:37:00 “I need ”
20:37:05 “*[VOLUME UP]*”
20:37:06 “*[VOLUME UP]*”
20:37:07 “*[VOLUME UP]*”
20:37:14 “something ”
20:37:16 “to drink ”
20:37:19 “i”
20:37:20 “m”
20:37:24 “m”
20:37:28 “ediately ”
At the beginning of each day, a date notation is made:
*[YY-MM-DD = 00-10-03]*

The LAM device includes certain functional commands that are recognized on the input serial port. The system provides the individual with method(s) for disabling and enabling the recording. It also allows setting the date and clock, such as at daylight savings time transitions or when changing time zones, or if the battery was allowed to fully discharge or needed replacement.

Privacy Protections

Privacy is always a concern whenever the communication of the individual is being recorded. The Code of Ethics of the American Speech-Language-Hearing Association (Principle of Ethics I, 1994) clearly addresses the behavior of SLPs on this point. Language sampling is not an entirely new phenomenon. Audio and video recordings have been used for years. These typically record both sides of the conversation, adding a greater dimension to privacy concerns. Video recordings are reasonably conspicuous. Audio recordings can be quite inconspicuous. Privacy issues using these methods have generally not been common, since they have become accepted methods of language sampling. However, with the addition of the recording system, the high cost of transcription, and the delay in obtaining useful information, these methods are impractical for most practicing clinicians.

The LAM device can be disabled in four ways: pushing the RECORDING OFF pushbutton, sending the *DISABLE* command, turning off the AAC device serial port, and disconnecting the cable. As an internal function, the LAM also can be disabled by a command. The AAC clinician must inform the individual regarding recording and train in how to perform the above functions.

The LAM data header starts with the warning:

*** CAUTION! ***
The following data represent personal communication.
Respect privacy accordingly.

Individuals whose communication is being monitored should be clearly informed and should provide written consent. Public use of recorded communication should assure anonymity.

RESULTS
Twenty LAM devices were fabricated for distribution to beta test sites. Seventeen sites used the devices and returned 33 language samples from 16 individuals.

Tools for Using the LAM data
The two primary elements of the logfile are the time stamp and the recorded language event. These two elements provide enough information for significant logfile analysis to occur. The desired outcome of the logfile or LAM data collection and analysis process is a set of summary measures that characterize the communication performance of the individual. A comprehensive system for supporting evidence-based AAC clinical practice consists of 1) the language activity monitor for collecting the data and 2) the software for converting LAM data into a summary measure report. The topic of this paper to this point has been the language activity monitor.

Summary measures that characterize communication performance are needed to support evidence-based practice. Those that have been considered to be clinically useful are the following:

A. Total utterances
B. Percentage of utterances that are complete
C. Percentage of utterances that are spontaneous
D. Mean Length of Utterance in words (MLUw)
E. Mean length of utterance in morphemes (MLUm)
F. Number of total words
G. Number of different word roots
H. Utterance-based average communication rate (words per minute)
I. Utterance-based peak communication rate (words per minute)
J. Communication rate for each language representation method
K. Selection rate (bits per second)
L. Rate index (communication rate / selection rate)
M. Language representation method usage (LRM)
N. Word selection errors per words selected
O. Spelling errors per words spelled
P. Appendices

1. Raw LAM data
2. Edited utterances
3. Coded utterances
4. Word list in alphabetical order
5. Word list in frequency order
6. Word list by language representation method
7. Word list comparison to reference lists

In order to obtain these summary measures from a language sample, a combination of automated and manual methods is necessary. This is because no single program exists that can generate this list of summary measures from the LAM data. The development of such a program is an area of ongoing work. However, for now the steps of uploading, editing, analysis, and report generation are separate and independent (fig. 3).

Fig. 3: Diagram representing complete LAM process of uploading, editing, analyzing and reporting.

Uploading

The beta test sites used HyperTerminal for uploading the data into the computer. This is a program that is a standard component of the Windows operating system, typically found under Programs/Accessories/Communications. Every site reported difficulty in configuring HyperTerminal to be compatible with the LAM device. This prompted the development of LAMterm, a software program with defaults that match the LAM device and the LAM function as a built-in feature. It is available without cost.

Editing

The purpose of editing is to prepare the sample for the analysis program(s) needed to generate the desired summary measures. A number of the summary measures can be generated using Systematic Analysis of Language Transcripts (SALT) (Miller and Chapman, 1983). SALT expects input in the form of defined utterances with correct spelling, perhaps preceded by the “S” designation for “Subject”. Specific coding procedures have been developed by Hill (2000) to convert LAM data into the format required by SALT.

Other summary measures can be generated using ACQUA (Augmentative Communication Quantitative Analysis) (Lesher, et. al., 2000). ACQUA was developed by the RERC on AAC. The authors have been in close communication with RERC staff associated with this task and have contributed to the functional definition of ACQUA. A set of editing conventions for ACQUA has been reported (Hill 2001).

The output of the AAC device is not well defined utterances, but rather a string of words, generally lacking sentence terminators (“.”, “?”, “!”). In addition, non-language events may be present. Further, the LAM adds time stamps, date transitions, and other non-language information. In addition to automatically stripping out the time stamps and other non-language information, the editor also must provide for the manual segmenting of the language data into defined utterances and allow the application of spell checking and error correction when needed.

Analysis

Language sample analysis programs (SALT, ACQUA, etc.) accept text, utterances, and/or logfiles and produce a report. Of the identified list of summary measures, SALT provides A, B, C, D, E, F, G, M, N, and O. ACQUA provides A, H, and I. Manual methods are used for summary measures such as communication rate by language representation method, selection rate analysis, and rate index (Hill & Romich 2002), that are not yet addressed by available software. Language representation methods commonly used in AAC systems are single meaning pictures, spelling and other alphabet-based approaches, and semantic compaction. Logged data has provided information on analysis of methods used that has never before been available. This is one of the most highly valued of the analyses identified to date.

Report Generation

A concise and useful report of the summary measures is being used clinically at this time. The common use of a standard format facilitates evidence-based AAC practice. The example of the AAC Performance Report is presented in the Appendix. The format presently in use is presented here with information typical of that logged using LAM with people who rely on AAC. Appendices to the summary measure report could include edited utterances, coded utterances and word lists.

DISCUSSION

Currently identified applications for automated language activity monitoring fall into the categories of clinical intervention and outcomes measurement.

Clinical Intervention
The AAC assessment process historically has been idiosyncratic, depending more on the knowledge and experience of the clinician than on the application of proven and scientific methods. The availability of LAM tools and methods has begun to change that situation. Already, clinicians are using LAM data to measure selection rate, for example. This allows for the objective comparison of the various selection techniques that could be considered for an individual (Romich, Hill, and Spaeth, 2001; Hill, Romich, and Ramachandran, 2001). Likewise, communication rate is now being measured and vocabulary use is being monitored, resulting in information never before available to guide the therapy process (Hill, 2001). Clinicians should have more quantitative data available to provide for more effective therapy. Figure 4 shows the prototype LAM device which can be used to collect a language sample for supporting AAC therapy.

Fig. 4: Prototype LAM mounted on AAC system with augmented communicator.

Performance data collection in natural settings is essential for intervention designed to build communicative competence of individuals who rely on AAC (Light and Binger, 1998). The LAM tools provide clinicians with an efficient method to follow the principles of evidence-based practice to support clinical decision-making and evaluate intervention services. Historically, in a typical therapy session, the speech-language pathologist spends less than an hour with the individual who uses an AAC system. That session may include various activities, such as reviewing and practicing old vocabulary, introducing new vocabulary or symbol sequences, adding new words or messages to the system, and/or modeling interactions. At the next therapy session the clinician generally has no quantitative information relative to the effectiveness of the previous session. With LAM tools, clinicians are able to document daily use of not only targeted vocabulary, but also morphologic and syntactic structures, and total number of words used. In addition, information unique to AAC communicative competence can be analyzed to support intervention strategies such as the language representation method used to generate a message, errors in message construction, and rate of output.

Outcomes Measurement
Current best practice in AAC implementation emphasizes communication outcomes based on a team selecting outcomes from a functional curriculum model (Blackstone, 1990; Gray, 1998; Hill, 1996). AAC outcomes can be determined by documenting subjectively positive changes in the attitudes of teachers, classmates, co-workers, and others toward the consumer (Calculator, 1998). The LAM tools provide objective data to analyze systematically the scope and sequence of expected AAC outcomes. In particular, for a school aged AAC user the team now has the tools to develop and monitor Individual Education Program (I.E.P.) goals and objectives. The team can not only quantify the I.E.P. objectives, but also can qualify the implementation strategies and techniques used to facilitate AAC system use.

SUMMARY

The innovative aspect of automated language activity monitoring and analysis is the systematic collection and processing of performance data on the actual language activity of individuals who rely on AAC. Performance measurement tools allow for data collection to occur effectively and efficiently for the first time in both clinical and natural environments. A previous Perspective article for Assistive Technology written by the second author discussed key issues that could facilitate the greater use of assistive technology by people with disabilities (Romich, 1993). The identified areas of concern included awareness, professional preparation and ethics, service delivery, and funding. Many of these issues still need to be addressed by professionals concerned with advancing the field of AAC today. The development of AAC performance measurement tools and procedures provides exciting possibilities and opportunities to contribute valuable information for clinical intervention, outcomes measurement, and research that relate directly to these historic issues.

ACKNOWLEDGEMENT

The authors gratefully acknowledge the support of the National Institute for Deafness and Other Communication Disorders of NIH. NIH Small Business Innovation Research (SBIR) grants have provided funding for the early stages of this work.

REFERENCES

Ahlsen, E., & Stromqvist, S. (1999). ScriptLog: A tool for logging the writing process and its possible diagnostic use. Paper presented at the biennial 1998 ISAAC Research Symposium, Dublin, Ireland.

Beck, A. R. (1996). Language assessment methods for three age groups of children. Journal of Children's Communication Development, 17(2), 51-66.

Blackstone, S. (1990). In pursuit of opportunities and interdependence. Augmentative Communication New, 1-4.

Calculator, S. N. (1998). AAC and individuals with severe to profound disabilities. In S.L. Glennon & D.C. DeCoste (Eds.), Handbook of Augmentative and Alternative Communication. (pp. 445-471). San Diego: Singular Publishing Group.

Cole, K.N., Mills, R., & Dale, P. (1989). Examination of test-retest and split-half reliability for measures derived from language samples of young children. Language, Speech, and Hearing Services in Schools, 20, 259-268.

Copestake, A., & Flickinger, D. (1999). Evaluation of NLP technology for AAC using logged data. Paper presented at the biennial 1998 ISAAC Research Symposium, Dublin, Ireland.

DeRuyter, F. (1998). Concepts and rationale for accountability in assistive technology. RESNA
Resource Guide for Assistive Technology Outcomes: Measurement Tools, USA, 1, 2-14.

Granlund, M., Blackstone, S., & Norris, L. (1996). Measuring the outcome of AAC intervention: Defining outcomes. Communication...Naturally: Theoretical and Methodological Issues in Augmentative and Alternative Communication. Proceedings of the Fourth ISAAC Research Symposium. (pg 203-216). E. L. Bjorck-Akesson, P. Vancouver, Canada: Malardalen University Press.

Gray, S. (1998). AAC in the educational setting. In S.L. Glennon & D.C. DeCoste (Eds.),
Handbook of Augmentative and Alternative Communication (pp. 547-597). San Diego: Singular Publishing Group. 547-597.

Higginbotham, D.J., Lesher, G.W., & Moulton, B.J. (1999). Development of a voluntary standard format for augmentative communication device logfiles. Proceedings of the RESNA 1999, Long Beach (pp25-27). Arlington, VA: RESNA Press.

Hill, K.J. (1996). Welcome to the real world using quality management tools to support the implementation process. Proceedings of the 10th U.S. Minspeak Conference, (pp 138-147) Wooster.

Hill, K.J. (2001). The development of a model for automated performance measurement and the establishment of performance Indices for Augmented Communicators under two sampling conditions. (unpublished dissertation). University of Pittsburgh, Pittsburgh, Pennsylvania. Ann Arbor, Michigan: Bell & Howell Company.

Hill, K., Ramachandran, P., & Romich, B. (2001). Measuring AAC system selection rate. Paper presented at the annual convention of the American Speech Language Hearing Association, New Orleans, LA.

Hill, K. J., & Romich, B.A. (1998). Language research needs and tools in AAC. Paper presented at the Biomedical Engineering Society 1998 Annual Conference, Cleveland.

Hill, K.J., & Romich, B.A. (1999). A proposed standard for AAC and writing system data logging for clinical intervention, outcomes measurement, and research. Proceedings of the RESNA 1999, Long Beach (pp22-24). Arlington, VA: RESNA Press.

Hill, K.J. & Romich, B.A. (2002). A rate index for augmentative and alternative communication. International Journal of Speech Technology. 5(1): 57-64.

Jutai, J., Ladak, N., Schuller, R., Naumann, S., & Wright, V. (1996). Outcomes measurement of assistive technologies: An institutional case study. Assistive Technology, 8, 110-120.

Lesher, G.W., Rinkus, G.J., Moulton, B.J., & Higginbotham, D.J. (2000). Logging and analysis of augmentative communication. Proceedings of RESNA 2000, Orlando (pp 82-84). Arlington, VA: RESNA Press.

Light, J. ,& Binger, C. (1998). Building Communicative Competence with Individuals Who Use Augmentative and Alternative Communication. Baltimore: Paul H. Brookes Publishing. 9-12.

Logemann, J.A. (2000). President's Column: What is evidence based practice and why should we care? ASHA Leader, 5(5), 3.

Miller, J. F., & Chapman, R.S. (1983). Systematic analysis of language transcripts (SALT). San Diego: College Hill Press.

Miller, L. J., Demasco, P.W., & Elkins, R.A. (1990). Automatic data collection and analysis in an augmentative communication system. Proceedings of RESNA 1990, Washington, D.C. (pp 99-100). Arlington, VA: RESNA Press.

Paul, R. (1995). Language Disorders from Infancy through Adolescence: Assessment & Interventon. St. Louis: Mosby. 273-369.

Romich, B. A. (1993). Assistive technology and AAC: An industry perspective. Assistive Technology, 5, 74-77.

Romich, B.A., & Hill, K.J. (1999). A language activity monitor for AAC and writing systems: Clinical intervention, outcomes measurement, and research. Proceedings of the RESNA 1999, Long Beach (pp19-21). Arlington, VA: RESNA Press.

Romich, B.A., Hill, K.J., and Spaeth, D.M. (2001). AAC selection rate measurement: A method for clinical use based on spelling. Proceedings of RESNA 2001, Reno, (pp 52-54). Arlington, VA: RESNA Press.