WordQ Writing Aid Software

Word Prediction

As you type, WordQ continuously predicts what you are typing and presents you with a list of correctly spelled words. When you see the word that you want to use, you can choose it with a single keystroke. You can also display the word with its different word endings. If you need help deciding which word to use, each word can be read aloud to you before you make a selection.

The tool learns and adapts to your writing style as you use it. However, if you wish, you can quickly and easily customize your writing vocabulary by importing text from another file.

Using knowledge about the last word that you wrote, WordQ also attempts to predict the next word in your sentence. An understanding of your vocabulary requirements and how you typically combine words ensures that the most relevant words are always presented.

Prediction Dictionary and User Vocabularies

The success of word prediction for spelling assistance depends strongly upon the dictionary and the prediction algorithm. In addition, certain levels of visual and cognitive skills are demanded of the user. These demands are important considerations for individuals who may also have visual deficits, attention problems and other cognitive concerns. Proper configuration of WordQ should help lessen these demands. Flexibility in presentation and interaction of the software is available to match the specific visual-cognitive needs and learning profile of the user.

WordQ uses a self-adapting statistical model of prediction incorporating heuristics or practical rules. Underlying the prediction is a dictionary of approximately 60,000 words developed through linguistic analysis of written text from a variety of sources including samples of writings by students of different age levels. While a larger dictionary is possible, the instances of word usage beyond the 60,000 words are so small that the potential confusion of a truly misspelled word existing as a real obscure word outweighs the value of including obscure words.

The challenge of a word prediction system is not to show a list of all possible words, but to show a list of words that are potentially within the user’s vocabulary and intent. Thus, the developers very intentionally kept the dictionary at a realistically appropriate size for the intended population of users. Both a Canadian dictionary and a US dictionary are included. The Canadian dictionary was validated against the Gage Canadian 2000 dictionary. Both dictionaries also contain hundreds of proper names as well as towns, cities, provinces and states localized for Canada and the US.

Associated with the dictionary is a database of frequency data for each word individually and in combination with other words in the dictionary. There is frequency data for approximately 1.25 million known word combinations. The effect of this linguistic knowledge is that WordQ emulates grammatical knowledge without specific grammatical tagging of words.

Word prediction actually occurs within what is called a user vocabulary. This is a smaller set of words that represents the user’s working memory. Several starting vocabulary templates are included with WordQ at different writing and age levels:

WordQ first requires the user to choose a starting template. Thereafter, WordQ can optionally self-adapt with use. In all cases, the background dictionary and its associated frequency data is always available. Thus, even if a Blank template is used, any words added to it will automatically have known word relationship and frequency information. As users progress through the grades, they can continue with their own user vocabulary, which will have adapted to their use.

As the user begins typing, words that complete the first letter(s) are predicted based upon frequency of use. The next word, however, is predicted based upon the likelihood of following the previous word. In some cases, such as following a function word (e.g., “to”), the number of possibilities is vast and the user would likely need to select another letter. However, at this point the words that complete that letter are chosen from words that follow the previous word. This quickly limits the possibilities to very likely words as well as words in the correct form (e.g., present or past tense). Thus, WordQ will most likely show appropriate words in the correct form. There are additional rules or heuristics that consider the prior use a function or content word to adjust the probabilities.

Simulation testing at Bloorview MacMillan Children’s Centre (Nantais, Shein and Johansson, 2001) with a number of different text sources suggests that WordQ can predict the desired word within 1 keystroke approximately 56% of the time; within 2 keystrokes approximately 72% of the time; and within 3 keystrokes approximately 86% of the time.

When the user types a word outside of their vocabulary, it may be correctly spelled, it may be a spelling error, or it may be a novel word. If the word is correctly spell checked by the WordQ dictionary, the word may be added to the user vocabulary automatically (an optional setting). If the dictionary rejects the word, WordQ holds it for later review by the user. The user can then delete it as a misspelling or add it to their vocabulary as a novel word. Only correctly spelled words and those intentionally added to the vocabulary are shown in the word list.

Prediction options

There are four options associated with WordQ word prediction:

Customized topic-specific vocabularies

Customizing a vocabulary for specific needs is done at two levels. A user vocabulary may be customized for an overall writing style by importing samples of text files. Importing text is similar to the adaptation that occurs naturally over time, but it is done quickly. The user vocabulary may initially start as a blank or with just core function words. When text files are imported, they are automatically analysed for word usage and the word prediction will then take on the writing style of the source documents. There are no limits on the number of user vocabularies that may be created.

You can also easily add and use topic lists of words/phrases that emphasize those words in the prediction process. For example, a topic may be a list of baseball terms, a list of Prime Ministers, or a list of dinosaurs. Topic words are essentially user vocabulary words that are singled out for boosting in the prediction process. Because they are user vocabulary words, they will have word usage data and hence they will be shown only when appropriate. When creating a topic list, novel words are automatically added to the user vocabulary.
While only one topic active can be active at a time, there are no limits on the number of topics that you use. You can switch between topics at any time (e.g., between sentences or paragraphs). When a topic is active, words within that topic are more likely to appear.