Barry A. Romich1,3, Katya J. Hill2,3, and Donald M. Spaeth3
1 Prentke Romich Company, Wooster, OH 44691
2 Edinboro University of Pennsylvania, Edinboro, PA 16444
3 University of Pittsburgh, Pittsburgh, PA 15260

The rate at which a person who relies on AAC (augmentative and alternative communication) can make choices from an array determines the speed of communication. Automated language activity monitoring (LAM) tools allow the measurement of selection rate based on logged data generated by actual AAC system operators. A method for measuring selection rate based on spelling speed is presented. Implications are significant for clinical assessment and intervention.

The AAC clinical processes initiate with a team commitment to work toward the goal of AAC. The team agrees to provide the supports and services that result in the most effective communication possible for the individual. Generally, effective communication is accomplished through spontaneous novel utterance generation (SNUG). For people who rely on AAC, one measure of communication effectiveness is communication rate, which is almost always far slower than natural speech. Therefore, every effort must be made to maximize communication rate for this population. Communication rate is influenced by many factors. By far the most significant factor can be the language representation method (LRM) employed for accessing core vocabulary. However, the speed of making selections also can be an important factor.

Once the team has determined the LRM(s), decisions regarding the most appropriate selection technique must be made. Research on how different AAC configurations affect speed and efficiency is needed to facilitate the clinical decision-making process for motor access (1). For most teams, determination of selection techniques (e.g., keyboard vs. headpointing) has been qualitative ratings of speed and reliability. Decisions relative to selection rate have been based on clinical intuition, trial-and-error counts, or not documented.

Recent work has resulted in AAC language activity monitoring (LAM) (2, 3) for clinical use, funded in part by the National Institute for Deafness and Other Communication Disorders on NIH, and a comprehensive universal logfile standard (4). These developments have made available tools to collect language sample quantitative data on which to base assessment and intervention decisions. LAM data is being used to produce quantitative AAC performance summary measures.

A method for measuring selection rate proposed in late 1999 has been used clinically (5) since that time. While the results have been useful, the method requires that the subject produce a particular pattern of selections. While this is not onerous, it does require an intentional setup and thus may be done infrequently in the course of normal therapy.

The objective of this work is to provide a method of extracting selection rate measurement from normal LAM data. Selection rate can be used for the comparison of different selection techniques on systems of different array sizes, the measurement of progress in learning to use a particular technique, and changes in rate that might occur as a result of other short term (e.g., fatigue) or long term (e.g., learning curve, physical improvement or deterioration) factors.

The human interface information transfer rate has historically been measured in terms of bits per second. (The rate for able-bodied people is generally considered to be under 100 bits per second. (6)). Consistent with historical measures, for this work the selection rate also is being reported in terms of bits per second.

The size of the array (A) (e.g., number of keys on a keyboard) from which choices are being made determines the number of bits (N) that are available with each choice. A = 2(exponentN). N = ln(A) / ln(2). The integer number of bits (N) for various array sizes (A) is presented in this chart:


Thus, if a person were able to make one choice per second from a keyboard with 128 keys, the selection rate would be seven bits per second. It is important to make the distinction here that the selection rate is not the same as the AAC communication rate. While the selection rate influences the communication rate, other factors, such as the language representation methods employed, do as well.

Language samples from LAM are reported in the following format: 20:37:00 “content”. The time stamp is a 24 hour format with one second resolution. A space and two quotation marks follow the time stamp. The content of the language event being recorded is between the quotation marks. Each event is presented on a new line.

The original selection rate measurement required the subject to enter a series of specific selections. The LAM would time stamp and record these selections. For individuals who spell as a normal part of their communication, and the letters are selected with a predictable number of selections (generally one), spelled word LAM data can be used to determine selection rate.

The timestamp of the event of the first letter(s) of the word is the start time (S). The time stamp of the last letter (or SPACE) of the word is the end time (E). The selection rate (SR) in bits per second is defined as follows, where L is the number of letters (including SPACE) following the first event selected during the spelling process, A is the number of locations in the selection array used for normal communication, and NS is the number of selections required per letter. SR = NS ((L) x ln(A) / ln(2)) / (E – S). For example, if the LAM data showed that a word of eight letters (L = 8) following the first event was spelled (from an array (A) of 128 keys) with direct access to the letters (NS = 1), the first event was at 09:37:17 (S), and the last letter was selected at 09:37:27 (E), then the selection rate would be 5.6 bits per second.

The above process is applied to all spelled single words with no multiple or repeated letters and no error correction in the language sample. Considering that spelling may be interrupted or erratic, the reported selection rate is the weighted average (by L) of all calculated selection rates above the mean. The need for multiple calculations makes this potentially much more time consuming than the original method if done manually. However, as a feature in an automatic analysis program it provides routine selection rate measurement with no additional clinical procedure.

AAC professionals, in order to increase the selection rate, generally will want to maximize the number of keys available to the user. However, range of motion and pointing skill put limits on what can be done in this area. Fitts’ Law (7) offers some theoretical predictions on how quickly an individual can make choices of targets of a given size located a given distance from a starting point. Since the application of Fitts’ Law in the clinical setting would require information not generally available, the more practical approach is actual trials on keyboards of different sizes.

The development of selection skills requires training time. With quantitative measurement of performance, rational decisions can be made relative to level and stability of performance.

Consideration of error correction is not provided in the above procedure. This is because the number of selections necessary to correct an error is a function of features of the AAC system. The time used in generating errors is included in the test, but no consideration of correction time.

The AAC clinician is cautioned that other factors can influence communication rate. Some of these factors can be far more significant than the typical differences between some selection techniques. For example, motor planning can be important in developing communication speed. Also, an alphabet-based language representation method may produce only a single letter per selection while a whole word per selection may be produced by other methods. And navigating from screen to screen to access single meaning pictures can be problematic and time consuming.

The availability of methods for measuring selection rate has implications in the areas of clinical intervention, outcomes measurement, and research. The end result of the use of these tools is the enhanced communication and higher personal achievement of people who rely on AAC.

By having measurement tools and methods, additional information on the acquisition of personal AAC performance will be forthcoming. Understanding the characteristics of the learning process should impact AAC assessment approaches.

(1) Glennen SL & DeCoste D (1997). Handbook of Augmentative and Alternative Communication. Singular Publishing Group, Inc. San Diego. pg. 249.

(2) Romich BA & Hill KJ (1999). A language activity monitor for AAC and writing systems: Clinical intervention, outcomes measurement, and research. Proceedings of the RESNA ’99 Annual Conference. Long Beach, CA. pp 19-21.

(3) Hill, K & Romich B (1999). AAC language activity monitoring and analysis for clinical intervention and research outcomes. CSUN. Los Angeles, CA.

(4) Higginbotham DJ & Lesher GW (1999). Development of a voluntary standard format for augmentative communication device logfiles. Proceedings of the RESNA ’99 Annual Conference. Long Beach, CA. pp 25-27.

(5) Romich, B, Hill, K & Spaeth, D (2000). AAC selection rate measurement: tools for clinical use. Proceedings of the RESNA 2000 Annual Conference. Orlando, FL. pp 61-63.

(6) Lucky RW (1989). Silicon Dreams New York, NY: St. Martin’s Press.

(7) Fitts P (1954). The Information Capacity of the Human Motor System in Controlling the Amplitude of Movements. Journal of Experimental Psychology. Vol. 47, No 6.

Barry A. Romich, P.E.
Prentke Romich Company
1022 Heyl Road
Wooster, OH 44691
Tel: 330-262-1984 ext. 211
Fax: 330-263-4829
Email: bromich@aol.com