Word Prediction

If Word Prediction Can Help, Which Program Do You Choose?

Jamie Klund, MS, OTR

Sun Prairie Public Schools
509 Commercial Avenue
Sun Prairie, WI 53590
(608) 837-2541 (Ext. 2157)
email: jkotr@aol.com

Mark Novak, BSEE, PE

Trace R&D Center
1500 Highland Avenue
Madison, WI 53705-2280
(608) 262-6966
email: menovak@facstaff.wisc.edu

Abstract:

Word prediction software programs have been available for IBM and Macintosh personal computer for several years. The concept of word prediction began as a simple assistive tool to reduce the number of keystrokes necessary for individuals with mobility impairment, making it easier to communicate and less fatiguing for the user. However, recent studies have reported the benefits of word prediction techniques for uses other than saving keystrokes and perhaps warranting further examination.

Introduction:

Very important issues to many Alternative Communication and Assistive Technology (AAC/AT) users are the ability to communicate effectively and/or the ability to type effectively on the computer, both for education and competitive employment opportunities. An important characteristic of both effective communication and typing is the "rate" at which one communicates or is able to generate text. Rate is often defined as keystroke efficiency.2

Within the multitude of research and clinical studies that exist in this area, there are several rate enhancement techniques available to the AAC/AT user. Some of the more common strategies include abbreviation expansion, word prediction, and semantic encoding. This paper/presentation will discuss the advantages and disadvantages of word prediction. Also attached to this paper are feature comparison charts of six word prediction programs and a comprehensive listing of word prediction software developers in appendix A and B respectively.

Discussion:

There has been confusion in the rehabilitation community as to the use of the term "word prediction," because word prediction has come to imply both "word completion and word prediction." Word completion is defined as offering the user a list of words after a letter has been typed. Word prediction is defined as offering the user a list of words after a word has been typed or selected, based on previous words rather than on the basis of letters.3 Word prediction programs now offer a multitude of prediction strategies including word lists based on spelling and frequency of word usage in the English language; word lists based on word recency, association, and grammar; and finally, word lists based on topic words.6

Initially, word prediction provided the physically disabled user a simple word selection list for completing their typed or communication selections. This word list offered rate enhancement by decreasing the number of keystrokes required to type any given word, which became known as keystroke savings. Several studies cited in the literature support the use of word prediction to enhance keystroke savings with experimentally determined keystroke savings' ranges of 37-47% and clinical data to support 23-58% keystroke savings.5,10

There are, however, increased cognitive and perceptual loads that accompany the use of word prediction. The most commonly cited activities increasing cognitive and perceptual loads include visually searching the list of words and deciding whether or not the list contains the desired word.10 Other activities that increase perceptual loads include changing the point of gaze between the keyboard and the display6 as well as the cognitive processing that is needed to use and guide the overall activity.7 There are several studies in the literature which demonstrate that even though there is a keystroke savings with the use of word prediction, there is not always an improvement in overall text generation or communication rate due to the costs of increased cognitive and perceptual loads.7,10

Some factors can either increase or decrease the cognitive and perceptual loads associated with word prediction. The continued use or practice of using word prediction over time may actually improve performance and facilitate more consistent and efficient strategies for using the word list. Studies that have documented the increased cognitive and perceptual loads of word prediction gave the participants only one training session before data collection began. Interestingly, user performance was still improving at the end of the studies.9,10 The strategy used to search the word list also has an effect on the text generation rate. A good "all purpose" strategy is to type one letter and then search the word list. However, depending upon the speed of keypress time and list search time, there may be a more efficient strategy.11

There may also be a reduced fatigue factor with the use of word prediction. Due to keystroke savings, the physically disabled user may be able to work longer with more comfort when using word prediction as compared to letters only typing.

Sometimes characteristics of a word prediction system may influence the cognitive and perceptual loads. These characteristics include the length, orientation, placement, and order of the word list. Studies have shown that speed of text generation rate plateaus at a word list length of about five words.12,15 A vertical word list may be easier to visually scan than a horizontal word list,15 as well as easier to cognitively process, depending upon word order. Words lists can be ordered alphabetically, by word length or statistically.

Another important characteristic of the word prediction system is the lexicon. The lexicon, or dictionary of stored vocabulary, effects whether the desired word will appear in the word list. A relatively small dictionary containing the highest frequency words in the English language will be effective, based on normal word usage.1,16 A dictionary can be fixed, having a stored vocabulary that never changes. This fixed list may be more predictable to the user and become more automatic over time.3 An adaptive dictionary is able to collect new words as the user types and may be more apt to generate the desired word sooner since it is capable of learning the user's typing or communication style. Words can also be added to topic dictionaries of a particular subject matter, forcing these specific content area words to appear in the word list.

Other characteristics of a word prediction system which can improve overall keystroke savings may include automatic spacing and automatic capitalization as well as abbreviation expansion.15

In more recent literature, other benefits of word prediction have been explored, showing that use of word prediction can improve the quality and quantity of written work.4 Word prediction may also enhance the development of written literacy skills5 and provide spelling assistance to those with various levels of spelling disorder. The user can often recognize and select a word from a list that he/she would otherwise have been unable to spell, increasing their independence in producing written work.14 Likewise, the use of grammatical word prediction has been shown to improve sentence structure and the grammatical accuracy of text: Specifically, grammatical word prediction offers assistance in remembering not to omit common function words and using the correct form of words as appropriate.13 Other benefits mentioned in the recent literature include increased user attention span, improved confidence, improved ability to write independently, and increased stimulation of language and vocabulary development.4,8,14,17

In summary, the effect of word prediction may be unique to each individual user. The ability to improve effective communication and typing rate is dependent on the characteristics of the user, the costs versus benefits of using word prediction, and the characteristics of the word prediction system itself. These are all important factors to consider when determining if word prediction is appropriate, rather than thinking about word prediction as a rate enhancement tool alone.

Appendix A: Word Prediction Comparison Charts

IBM Focus

FEATURES4 KeyREP Keycache Aurora

Platform PC PC PC

- DOS/Windows No/Yes Yes/Yes Yes/Yes

Self-contained word proc. No No No

Dictionary

- Number of words 5K with reg. 100K

- Add words and phrases Yes Yes Yes

- Choice of dictionaries Yes Yes Yes

- Save/load per. dict. Yes Yes Yes

- Delete/modify dict. Yes Yes Yes

- Auto-learn Yes Yes Yes

- Abbreviation expansion Yes Yes Yes

- Allows for Macros No Yes No

- Dictionary extension *.prd *.dic *.dct

Display Window

- Change size/type Yes No Yes

- Change font Yes No Yes

- Adjust size of window Yes No (auto) Yes

- Adjust location Yes Yes Yes

- Fixed or Dynamic both both fixed

Word Choice Arrangement

- Number of words 1-10 0-9, jump 1-9,plus

- Orientation Hor/ Ver. Ver. Hor/Ver

- Alphabetical order No (freq.) Yes Yes

- Error Correction Key Yes No Yes

Prediction

- Predicts ahead - next No Yes Yes

- Predicts phrases Yes Yes Yes

- Predicts by grammar No No Yes

Preferences

- Cap. first word in Yes Yes Yes

- Space after word Yes Yes Yes

- Space after period Yes Yes Yes

- Choose by number Yes Yes Yes

- Choose by point/click No Yes No

- To type selection type #, <esc> type #, type 0

- On-Screen Keyboard No No No

Speech

- Speak letter/word No No Yes

- Speak word choices No No Yes

- Speak finished sentence No No Yes

Price (~Aug. 1, 95, DOS/Win) $290 $75/$100 (var./$430)

Macintosh Focus

FEATURES4 TELEPATHIC 2.0 CO:WRITER WORDWRITER

Platform Mac Mac Mac

Self-contained word proc. No No No

Dictionary

- Number of words 0-16,000 2, 10 40K 4000

- Add words and phrases Yes Yes Yes

- Choice of dictionaries Yes Yes No

- Save/load per. dict Yes Yes No

- Delete/modify dict Yes Yes Yes

- Auto-learn Yes Yes No

- Abbreviation expansion Yes Yes No

Display Window

- Change size/type Yes Yes No

- Change font Yes Yes No

- Adjust size of window Yes Yes Yes

- Adjust location Yes Yes Yes

Word Choice Arrangement

- Number of words 1-64 0-9 Page

- Fixed or Dynamic Fixed Either Fixed

Arrangement

- Orientation Hor/Ver. Hor/Ver. Page

- Alphabetical order Yes Yes Yes

- Error Correction Key Yes Yes No

Prediction

- Predicts ahead Yes Yes No

- Predicts phrases Yes Yes Yes

- Predicts by grammar No Yes No

Preferences

- Cap. first word in Yes Yes No

- Space after word Yes Yes Yes

- Space after period Yes Yes No

- Choose by number Yes Yes No

- To type selection Set hot key esc. key N/A

- Choose by point/click Yes Yes Yes

- On-Screen Keyboard Yes No Yes

Speech

- Speak letter/word Yes Yes No

- Speak word choices Yes Yes No

- Speak finished sentence No Yes No

Price $295 $290 $150

Appendix B: Word Prediction Software Developer Listing

NOTE: This list is intended as a reference for those who want an overview of word prediction software - rather than a "shopping list." Items appearing with a "M" are intended for the Macintosh operating system, while items appearing with a "D" or "W" are for the PC operating systems DOS and Windows respectively.

Access Technology
1435 N. Meridian
Indianapolis, IN 46202
(317) 686-6093
FAX (317) 686-6094

SofType - W

Aurora Systems, Inc.
2647 Kingsway
Vancouver, BC V5R
5H4 CANADA
(604) 436-2694
(800) 361-8255
FAX (604) 436-2695
aurora@cyberstore.ca

Cape Software
P.O. Box 800
Harwich, MA 02645
(508) 432-2435

RotoTyper - M

Consultants for Communication Technology
508 Bellevue Terrace
Pittsburgh, PA 15202
(412) 761-6062
FAX (412) 761-7336

Don Johnston Incorporated
P.O. Box 639
1000 N. Rand Rd.
Bldg. 115
Wauconda, IL 60084
(800) 999-4660
(708) 526-2682
FAX (708) 526-4177
djde@aol.com

Co:Writer - M

Gus Communications, Inc.
P.O. Box 4362
Blaine, WA 98231
(604) 279-0110
FAX (604) 279-8474
gharris@guscom.com
http://www.direct.ca/gus/index.html

Gus! Multimedia Speech System - W iANSYST Ltd.
The White House
72 Fen Road
CAMBRIDGE
CB4 1UN
(+44)0 1223 42 01 01
100273.1226@compuserve.com

textHELP - W Innovative Designs, Inc.
2464 El Camino Real
Suite 245
Santa Clara, CA 95051
(408) 985-9255
FAX (408) 247-6624
70402.2003@compuserve.com

Finish Line - D/W

Institute on Applied Technology
Children's Hospital
Communication Enhancement Center
Fegan Plaza
300 Longwood Avenue
Boston, MA 02115
(617) 355-6998

WriteAway - D

Kepler Electronic Publishing, Inc.
1525 Centennial Drive
Kingston, ON K7L
4V2 CANADA
(613) 384-1386
FAX (613) 384-3208
kepler@fox.nstn.ca
http://www.novatech.on.ca/kepler/page1.html

PAL - D

Madenta Communications, Inc.
9411A - 20 Avenue
Edmonton, AB T6N
1E5 CANADA
(403) 450-8926
(800) 661-8406
FAX (403) 988-6182

Microsystems Software Inc.
600 Worcester Road
Framingham, MA 01701
(800) 828-2600
(508) 879-9000
BBS (508) 875-8009
FAX (508) 626-8515

NanoPac, Inc.
4833 South Sheridan Road
Suite 402
Tulsa, OK 74145-5718
(918) 665-0329
FAX (918) 665-0361
70524.3306@compuserve.com; dsjb79a@prodigy.com

CINTEX I - D

Olduvai Corp.
9200 s. Dadeland Blvd., Suite 725
Miami, FL 33156
1-800-548-5151
(305) 670-1112
d-283@applelink.apple.com

Magic Typist - M

OMS Development
610-B Forest Avenue
Wilmette, IL 60091
(708) 251-5787
(800) 831-0272
FAX (708) 251-5793
ebholman@netcom.com

Keycache - D/W

Prentke Romich Company
1022 Heyl Road
Wooster, OH 44691
(800) 262-1933
OH (216) 262-1984

Words+, Inc.
40015 Sierra Highway
Bldg. B-145
Palmdale, CA 93550
(805) 266-8500

World Communications
245 Tonopah Drive
Fremont, CA 94539
(510) 656-0911

References:

D.R. Beukelman, K.M. Yorkston, M. Poblete, C. Naranjo, "Frequency of Word Occurrence in Communication Samples Produced by Adult Communication Users," Journal of Speech and Hearing Disorders, vol. 49, pp. 360-367, 1984.
S. Blackstone, Augmentative Communication News, September, 1990, Vol.3, No.5
A.M. Cook, S.M. Hussey, Assistive Technologies: Principles and Practice, Mosby-Year Book Inc., 1995, pp. 493-498.
B. Heinisch, J. Hecht, "A Comparison of Six Programs, Word Prediction Software," TAM Newsletter, pp. 4-9, 1993
D.J. Higginbotham, "Evaluation of Keystroke Savings Across Five Assistive Communication Technologies," Augmentative Alternative Communication, vol. 8, no. 4, 258-272, 1992.
D.J. Higginbotham, "Word Prediction in MS DOS Land," Communicating Together, Part 1, vol. 10, no. 1, pp. 10- 13, 1992, Part 2, vol. 10, no. 2, pp. 9-10, 1992
H.M. Horstmann, S.P. Levine, "The Effectiveness of Word Prediction," Proc. 14th RESNA Conf., 1991, pp. 100-102.
M. Hunt-Berg, J.L. Rankin, D.R. Beukelman, "Ponder the Possibilities: Computer-Supported Writing for Struggling Writers," Learning Disabilities Research and Practice, 9 (3), pp. 169-178, 1994.
H.H. Koester, S.P. Levine, "Learning and Performance of Able-Bodied Individuals Using Scanning Systems with and without Word Prediction," Assistive Technology, vol. 6, no. 1, pp. 42-53, 1994.
H.H. Koester, S.P. Levine, "Modeling the Speed of Text Entry with a Word Prediction Interface," IEEE Transactions on Rehabilitation Engineering, vol. 2, no. 3, pp. 177-187, 1994.
H.H. Koester, S.P. Levine, "Simulations of User Performance with Word Prediction," Proc. 18th RESNA Conf., 1995.
G.A. Miller, "The Magical Number Seven, Plus or Minus Two: Some Limits On Our Capacity For Processing Information," The Psychological Review, vol. 63, no. 2, pp. 81-97, 1956.
C. Morris, A. Newell, L. Booth, I. Ricketts, J. Arnott, "Syntax PAL: A System to Improve the Written Syntax of Language-Impaired Users," Assistive Technology, vol. 4, no. 2, pp. 51-59, 1992.
A.F. Newell, et. al., "Effect of the "PAL" Word Prediction System on the Quality and Quantity of Text Generation," AAC Augmentative and Alternative Communication, vol. 8, pp. 304-311, 1992.
A.L. Swiffin, J.L. Arnott, A.F. Newell, "Adaptive and Predictive Techniques in a Communication Prosthesis," Augmentative Alternative Communication vol. 3, no. 4, pp. 181-191, 1987.
G.C. Vanderheiden, D.P. Kelso, "Comparative Analysis of Fixed-Vocabulary Communication Acceleration Techniques," Augmentative Alternative Communication, vol. 3, no. 4, pp. 196-206, 1987.
W. Waltosz, "Word Prediction Schemes-They're Not All The Same," unpublished work, 1995

FEATURES4	KeyREP	Keycache	Aurora
Platform	PC	PC	PC
- DOS/Windows	No/Yes	Yes/Yes	Yes/Yes
Self-contained word proc.	No	No	No
Dictionary
- Number of words	5K	with reg.	100K
- Add words and phrases	Yes	Yes	Yes
- Choice of dictionaries	Yes	Yes	Yes
- Save/load per. dict.	Yes	Yes	Yes
- Delete/modify dict.	Yes	Yes	Yes
- Auto-learn	Yes	Yes	Yes
- Abbreviation expansion	Yes	Yes	Yes
- Allows for Macros	No	Yes	No
- Dictionary extension	*.prd	*.dic	*.dct
Display Window
- Change size/type	Yes	No	Yes
- Change font	Yes	No	Yes
- Adjust size of window	Yes	No (auto)	Yes
- Adjust location	Yes	Yes	Yes
- Fixed or Dynamic	both	both	fixed
Word Choice Arrangement
- Number of words	1-10	0-9, jump	1-9,plus
- Orientation	Hor/ Ver.	Ver.	Hor/Ver
- Alphabetical order	No (freq.)	Yes	Yes
- Error Correction Key	Yes	No	Yes
Prediction
- Predicts ahead - next	No	Yes	Yes
- Predicts phrases	Yes	Yes	Yes
- Predicts by grammar	No	No	Yes
Preferences
- Cap. first word in	Yes	Yes	Yes
- Space after word	Yes	Yes	Yes
- Space after period	Yes	Yes	Yes
- Choose by number	Yes	Yes	Yes
- Choose by point/click	No	Yes	No
- To type selection	type #,	<esc> type #,	type 0
- On-Screen Keyboard	No	No	No
Speech
- Speak letter/word	No	No	Yes
- Speak word choices	No	No	Yes
- Speak finished sentence	No	No	Yes
Price (~Aug. 1, 95, DOS/Win)	$290	$75/$100	(var./$430)

FEATURES4	TELEPATHIC 2.0	CO:WRITER	WORDWRITER
Platform	Mac	Mac	Mac
Self-contained word proc.	No	No	No
Dictionary
- Number of words	0-16,000	2, 10 40K	4000
- Add words and phrases	Yes	Yes	Yes
- Choice of dictionaries	Yes	Yes	No
- Save/load per. dict	Yes	Yes	No
- Delete/modify dict	Yes	Yes	Yes
- Auto-learn	Yes	Yes	No
- Abbreviation expansion	Yes	Yes	No
Display Window
- Change size/type	Yes	Yes	No
- Change font	Yes	Yes	No
- Adjust size of window	Yes	Yes	Yes
- Adjust location	Yes	Yes	Yes
Word Choice Arrangement
- Number of words	1-64	0-9	Page
- Fixed or Dynamic	Fixed	Either	Fixed
Arrangement
- Orientation	Hor/Ver.	Hor/Ver.	Page
- Alphabetical order	Yes	Yes	Yes
- Error Correction Key	Yes	Yes	No
Prediction
- Predicts ahead	Yes	Yes	No
- Predicts phrases	Yes	Yes	Yes
- Predicts by grammar	No	Yes	No
Preferences
- Cap. first word in	Yes	Yes	No
- Space after word	Yes	Yes	Yes
- Space after period	Yes	Yes	No
- Choose by number	Yes	Yes	No
- To type selection	Set hot key	esc. key	N/A
- Choose by point/click	Yes	Yes	Yes
- On-Screen Keyboard	Yes	No	Yes
Speech
- Speak letter/word	Yes	Yes	No
- Speak word choices	Yes	Yes	No
- Speak finished sentence	No	Yes	No
Price	$295	$290	$150