How does Autocorrect work?
The tech behind "Duck"
KCL Fintech Soc - Vishnu Thirumalai, London 10.07.19
Perhaps you are one of the few talented and gifted individuals who can write without the aid of spell and grammar checkers. Or, like the vast majority of us, you take full advantage of the auto correction software that is a standard feature of everything from word processors to messaging applications. Lurking in the background, they’ve for the most part saved us from a world of horribly misspelled emails and messages. It was not until recently, though, as I struggled through the spelling of ‘Polymorphism’ that I began to wonder how they actually worked.
While companies closely guard their own specific implementations, they all share some basic features. They repeatedly scan the text that is being written and extract what they identify as words, though they may be grossly misspelled. From this, the software then compares each word to an existing database (or dictionary) of words to ensure that it matches and is therefore accurate. If not, the spellchecker either highlights the word or automatically corrects it. The latter does have the often unwanted side effect of correcting to the wrong word, sometimes to comedic effect.
A recent improvement in spell checking technology is context sensitive spelling. Rather than just judging the word against its closest dictionary relative, it also takes into account the context the word is used in. A famous example of this is the use of the word “baht” (Thailand’s currency) would often be caught as a misspelling of bath. However, if the words surrounding ‘baht’ includes words related to Thailand, the spellchecker recognises that it’s the currency.
Simple grammar checkers also operate on a similar principle by scanning text and comparing them to a list of commonly misused used words. Simple grammar rules such as capitalization can also be easily caught. However, since grammar is incredibly dependent on context, the similarities end there. Due to the complex and often unstructured nature of human language, not to mention the myriad languages available, a perfect grammar checking program has yet to surface.
The autocorrect on modern phones often combines the two, and adds a predictive element. The predictive aspect comes from analysing large bodies of text (what linguists call the corpus), usually articles from popular media and previous things you’ve typed. This allows your phone to suggest what you’re most likely to write based on both of these, learning from your past habits and subtly shaping it according to the majority.
Collective Equity Ownership Ltd. (CEO) is a secondary fund. We allow founders and shareholders of VC-backed companies to pool together their shares with other late-stage companies to diversify their portfolio.
If you know any cool founders, or someone with a lot of VC experience, get in touch! We love meeting new people!
Other Viewers have also read: