The interplay of Machine learning and predictive texting (v1.1)

To oversimplify: Machine learning, combined with the logic of predictive texting allowing users to type sentences more conveniently.

Text prediction commonly incorporates a dictionary of pre-selected words, allowing the user to type the first few letters and choose from a set of ‘auto-completions.’ This coincides with machine learning when this list is modified by the user’s previous choices, as seen from the autocorrect function on most smartphones.

Earlier functions were more true to the title. on early cell phones with 9 keys, a person had to press a numeric button 1–4 times to get any desired letter. predictive text could see what set of words came from a single string of numbers, such as 4663 for gone. 4663 can, though less frequently, translate to home. if the machine learns, home will overtake gone.Words can even be hidden in these sets of letters, as letters can be swapped so long as they in the same set of three letters.

More advanced forms of the above allow for typos. by taking longer and analyzing possibilities from adjacent letters they can provide even more convenience for the users, at the cost of electrical power and speed.

However this is not without cost. users can be frustrated by a system (especially an over-assertive one) recommending the wrong words, and removing the convenience it attempts to provide. This gets exponentially complex, as more letters mean a much larger file of possibilities the user could want. The addition of too few words causes the same problem, forcing the user to add every word they want to type. The addition of too many causes words that the user would never logically use, an excess of space used up by these words, meaningful or not. This was especially a problem with early phones, especially ones with under a gigabyte of usable storage or low processing power.

The prediction of words to use next in a sentence is more difficult, both to conceptualize and to create. The iPhone’s quicktype function does this, analyzing chains of words the user types and suggesting what they will type next based on what it recognizes. this requires even more storage space to account for the IDs of words and where to use them, but saves even more time, enhancing the convenience that technology provides to humans.

Machine learning, when applied to real life, can greatly improve the quality of life of end users who want to make use of technology. This goes greatly for the… less tech-savvy users, who are unwilling to add words to their own ‘dictionary’ for future use. While doing nothing to increase the actual productivity of a machine; and in fact hindering it, and making it harder to program, it will leave users with a much better impression of any machine that uses it, making them more likely to buy more. To this effect it’s similar to a good-looking user interface, in the effect is has on the end users.