Topic modeling in fringe word prediction for AAC

Trnka, Keith; Yarrington, Debra; McCoy, Kathleen F.; Pennington, Christopher

doi:10.1145/1111449.1111509

Cited by 20 publications

(26 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Many researchers are currently investigating ways to improve the simplistic word prediction model [10,11,15]. These methods include predictions based not just on characters entered, but also on previous whole words entered and even topics that a user is talking about.…”

Section: Introductionmentioning

confidence: 99%

The effects of word prediction on communication rate for AAC

Trnka

Yarrington

McCaw

et al. 2007

Human Language Technologies 2007: The Conference of the North American Chapter of the Association for Computational Linguistics

Self Cite

View full text Add to dashboard Cite

Word prediction systems can reduce the number of keystrokes required to form a message in a letter-based AAC system. It has been questioned however, whether such savings translate into an enhanced communication rate due to the additional cognitive load (e.g., shifting of focus and scanning of a prediction list) required in using such a system. Our hypothesis is that word prediction has great potential in enhancing communication rate, but the amount is dependent on the accuracy of the word prediction system. Due to significant user interface variations in AAC systems and significant differences between communication rates achieved by different users on even the same device, this hypothesis is difficult to verify. We present a study of communication rate and word prediction systems using "pseudo-impaired" participants and two different word prediction systems compared against letter-by-letter entry. We find that word prediction systems can in fact speed communication rate, and that a more accurate word prediction system can raise communication rate higher than is explained by the additional accuracy of the system alone.

show abstract

Section: Introductionmentioning

confidence: 99%

The effects of word prediction on communication rate for AAC

Trnka

Yarrington

McCaw

et al. 2007

Human Language Technologies 2007: The Conference of the North American Chapter of the Association for Computational Linguistics

Self Cite

View full text Add to dashboard Cite

show abstract

“…A list of 5 words seems to be the most common [5,14]. In our past experience [21], we found that the differences between methods using the same ngram order were roughly the same regardless of window size.…”

Section: Discussionmentioning

confidence: 67%

“…Furthermore, the combination of in-domain training data with a much larger amount of out-of-domain data is more useful than either data set alone, even when the two training sets are combined naïvely. Beyond this, we show how language modeling improvements can still hold even when the training and testing languages are very different -we apply a topic model from [21] both in-domain and out-ofdomain and show that the topic model significantly improves keystroke savings despite the topical differences in training and testing corpora. Section 2 will give an overview of highly related work.…”

Section: Introductionmentioning

confidence: 86%

Corpus studies in word prediction

Trnka

McCoy

2007

Proceedings of the 9th International ACM SIGACCESS Conference on Computers and Accessibility

Self Cite

View full text Add to dashboard Cite

Word prediction can be used to enhance the communication rate of people with disabilities who use Augmentative and Alternative Communication (AAC) devices. We use statistical methods in a word prediction system, which are trained on a corpus, and then measure the efficacy of the resulting system by calculating the theoretical keystroke savings on some held out data. Ideally training and testing should be done on a large corpus of AAC text covering a variety of topics, but no such corpus exists. We discuss training and testing on a wide variety of corpora meant to approximate text from AAC users. We show that training on a combination of in-domain data with out-of-domain data is often more beneficial than either data set alone and that advanced language modeling such as topic modeling is portable even when applied to very different text.

show abstract

“…An individual using a communication aid that has a fixed set of words (vocabulary) may not have access to the very personalised, low-frequency, or context-specific words required to adapt their communication 35 . For those using alphabetic typing, non-adaptive word prediction may result in a more static use of language and may poorly predict fringe vocabulary 36 . Finally, in all cases, constructing a message using a communication aid is slow and this is likely to mean that the utterances are shorter (or telegrammatic) and that the message is thus less likely to be adapted to the context of the conversation.…”

Section: The Use Of Context For Augmentative Communicationmentioning

confidence: 99%

What is the potential for context aware communication aids?

Judge

Hawley

Cunningham

et al. 2015

Journal of Medical Engineering & Technology

View full text Add to dashboard Cite

Use of voice output communication aids (VOCAs) can be a very effective strategy to assist people with speech impairments in communicating. Despite this, people who use communication aids often express frustration with VOCAs -desiring devices that are simpler, quicker and more effective to use.Whilst it is not possible to resolve all these issues with technology, we argue that significant progress can be made. The use of contextual information is one development that could improve the simplicity and effectiveness of communication aid design. Individuals using a VOCA also face constraints related to the selection of these items.Those using text representation will need to select around 27-40 items (unless using an ambiguous keyboard 5 ), whereas those using graphic symbols may need to select from as many as 4000 items. To navigate this number of items the user is required to make a series of selections Individuals use a range of strategies to select items on a VOCA -some may be able to press directly onto a screen or keyboard (known as direct access), whereas others may have difficulties with co-ordination or control and require alternative access devices. These devices can include alternative keyboards (for example those with key 'guards'), alternative mice (for example, eye gaze or head movement selection), or other controls such as switches. Switches are binary (non-latched) 'buttons' that come in a range of types -with varying size, shape, activation force or method. When using a switch, the user will either press it to select the desired item from a list of items through which the computer is scrolling, or use multiple switches to manually 'move' through the possible selections and then select the desired item 7 .There are few studies measuring actual input or output (speaking) rates of VOCA users, however for text production an output (measured in words per minute -wpm) of less than 5 wpm for switch access 8 , or less than or around 10 wpm for AAC users using any mode 9,10 has been suggested. This can be compared to studies of typing on 'soft' keyboards that suggest rates of between 8-20wpm 11 . As the speaking rate of a typical speaker during conversation may be between 100 and 200 wpm 12,13 it can be seen that current AAC output is in the region of an order of magnitude slower than spoken speech.People who use AAC and VOCAs often cite challenges and frustrations related to the speed of communication using these systems and also in their simplicity of use 14,15 . This has been a focus of prior work carried out as part of the NIHR Devices for Dignity (D4D) Healthcare example, the iOS8 word prediction keyboard claims to take into account " your text messaging style", "who you're writing to", "what the conversation is about" and "how you email" 34 . The Use of Context for Augmentative CommunicationPeople who use AAC may be able to process and use the context of a conversation receptively in the same way as any other individual. However, their operational use of this contextual information is much mo...

show abstract

Topic modeling in fringe word prediction for AAC

Abstract: Word prediction can be used for enhancing the communication ability of persons with speech and language impairments. In this work, we explore two methods of adapting a language model to the topic of conversation, and apply these methods to the prediction of fringe words.

Cited by 20 publications

References 11 publications

The effects of word prediction on communication rate for AAC

The effects of word prediction on communication rate for AAC

Corpus studies in word prediction

What is the potential for context aware communication aids?

Contact Info

Product

Resources

About