Text Prediction FAQ

Text prediction works by suggesting the most likely word, words, or sentences, based on what the user has already entered. Using Natural Language Processing (NLP), and Machine Learning (ML), text prediction tools can give accurate predictions of what a user is most likely to write next. AI text prediction software can learn from the user, so that the predictions become more and more accurate with use.

Using a text prediction tool or writing assistant speeds up text entry by suggesting the next word or words, and even completing entire sentences. This means a user can complete an email, message or document with fewer keystrokes and therefore save time. Some text prediction tools can also improve the quality of writing by suggesting words or phrases the user may not have considered.

Some text prediction tools have grammar checking functionality built in. Both Microsoft Outlook and Google’s Smart Compose do suggest words or corrections based on grammar rules. Typewise’s text prediction AI also comes with a grammar checker as well as a spelling checker (autocorrect) functionality built in.

Most text prediction tools come with autocorrect functionality – but not all. Typewise text prediction has autocorrection built in.

Text prediction is a growing trend. As AI improves, we’re likely to see text prediction AI in more and more text input situations – such as the writing of emails, documents and support chats. Currently, text prediction functionality is seen in Gmail and some other Google products, as well as Microsoft Outlook and Excel. There are also browser based tools such as Typewise Text Prediction which enables the functionality to be used across many different browser-based platforms.

Text prediction currently tends to work best in situations where the text input is most predictable. For example, the composition of emails is an ideal candidate because there are generally words and phrases which are formulaic. This is why both Google and Microsoft have applied their text prediction AI to their email products first. As the AI increases in sophistication, expect to see text prediction being rolled out across more and more complex text entry situations.

Yes. Although billions of people spend hours each day typing on their computers and smartphones, the speed at which we can type is vastly slower than the speed at which we can think. Text prediction AI is enabling us to write much faster and as the AI improves, people will be able to type and text faster and faster – bringing us closer and closer to being able to type at the speed of thought.

Currently Google Gmail and Microsoft Outlook offer some level of autocompletion and next word suggestion. It’s likely that as text prediction AI improves, we will see wider adoption of predictive text tooling and eventually such functionality will become ubiquitous across all text input interfaces.

Yes. Languages that have longer average word lengths typically make text prediction even more useful. For example, German has an average word length of 10.55 characters, whereas English has 8.93. This means that when using text prediction to type German, on average, more characters are saved than in English (assuming the text prediction requires the same number of character input in both languages).

Any professional who enters text into a computer will save time with predictive text software. Professionals who need to type words that are longer than average (for example, doctors needing to type out long Latin names for body parts, illnesses or medicines) stand to save even more keystrokes and therefore more time.

Sentence completion is a type of text prediction where words are suggested to complete the sentence you have started to type. If the suggested words complete your sentence accurately – you can simply accept them in one keystroke (normally by pressing ‘tab’), or you can continue typing to finish the sentence manually. As the accuracy of sentence completion increases, the number of keystrokes saved increases significantly.

The accuracy of predictive text depends on what writing assistant tool is being used, and in what situation. In data comparing Typewise text prediction and autocorrect vs Google Gboard vs Apple Keyboard, Typewise's AI outperformed both, saving users a higher percentage of keystrokes.

As the current speed of text input occurs at only around 10% of the speed of human thought, there is still considerable ground to be gained in terms of writing assistant AI. As the sophistication of text prection tools increases, the number of keystrokes required will decrease, therefore reducing typing time and closing the gap between the speed of thought and the speed of typing.

Some writing assistants offer AI text prediction that can learn. This means the accuracy of the text it suggests improves with the amount of text entered by the user. Typewise AI for example can learn users slang and dialect, as well as words and phrases that are used frequently.

Both Microsoft Word and Google Docs offer some level of text prediction tooling.

Autocompletion in web browsers is a type of text prediction seen in the URL bar of browsers such as Google Chrome, Mozilla FireFox and Microsoft Edge, and is now standard. As many URLs are typically long and difficult to remember, autocompletion here is particularly useful, and simple character matching makes it an excellent candidate for the technology.

Email platforms were some of the first to add next word and sentence competition functionality. The formulaic nature of email composition (at least the opening and closing of correspondence) make email a good candidate for predictive text functionality. This is currently seen in Gmail’s ‘Smart Compose’ and Outlook's 'Text Suggestion' tooling.
‍
We also see autocomplete in the email address bar, where entering the first character of the intended recipient's address immediately brings up a suggestion of the full email address. Clicking on the address or tapping enter typically adds the recipient email to the ‘Send to’ bar – removing the need for the author to have to type out the full address.