This version of Swype contains a new component that we call the Context Prediction Engine (or "CPE" for short). CPE provides Swype with more sophisticated linguistic processing that improves both swyping and tapping accuracy. It does this by looking at previous words that you have entered (i.e., context) and which words tend to follow. For example, if you enter the phrase "I jumped into the mosh" and then swyped from 'p' to 't', you would expect Swype to generate "pit" as the default choice. Without CPE, Swype would have generated "put" because it is a more common word, but with CPE it favors "pit". The overall effect that you should see is improvements in Swype’s choices when you swype or tap. CPE also learns as you use Swype; there will be additional improvements as you continue to use it.You should see the most benefit from CPE on words which have an ambiguous path, such as pit/pot/put, our/or, I'm/in/on, at/st, yeah/trash, etc. This is especially the case when the preceding word(s) help to disambiguate the path.English (US), English (UK), French, Italian, German, and Spanish have full CPE support in this release, which includes an initial (aka static) language model. Canadian French is supported with a learning model only (aka dynamic model). Our research shows that the learning-only model should provide a similar experience after about 500-1000 words of text.The following are some examples of English (US) CPE that we use internally:
- Swype "I" then "an". Without CPE, Swype will output "an" - a common Swypo. With CPE, if you're near the 'm' even a little, "am" should be the default word, because "I am" is a much more common phrase than "I an".
- Swype "mosh" then "pit". Without CPE, Swype usually prefers "put" due to word frequency. With CPE, it should prefer "pit".
- initial model, static model - language usage information that's built into the release and used until the learning component takes over.
- learning model, dynamic model - language usage information learned from text you type using Swype.
- language model - both static and dynamic models are kinds of language model, so we might say "static language model" from time to time.
FAQFeel free to post questions and I'll answer as many as I can.
- The dynamic model learns from text you enter. You'll get the most out of CPE by correcting typos/Swypos (so that it doesn't learn the wrong text). You'll get slightly better accuracy out of the learning model if you make corrections before starting a new paragraph or switching fields.