Hacker News Reader

Comment by kevincox | Hacker News Reader

This seems very focused on spelling. What I find is key for a good spell correction system is how the words sound. For example I find that Firefox often can't find the word I meant for it's suggestion list but pasting the misspelt word into Google gets the right result 99% of the time as the one provided option.

I wonder how difficult it would be to adapt this to work on sounds or other frequent typos and misspelling sources instead of just characters. It seems it should he possible if you can define a decent "normalization" function.

View on HN

I used an old algorithm in a text search, called Metaphone[0]. It works relatively well, for English (There's a Spanish version, too). I used Double Metaphone.

[0] https://en.wikipedia.org/wiki/Metaphone

There's a pretty cool python library with a huge number of these if you want to experiment (GPLv3): https://github.com/chrislit/abydos

Metaphone is good if you typed in something that sounds like the correct word. But often spelling typos are missing letters, or pressing a nearby letter instead of the one you meant.

For those the sound of the word isn't right, you need something that checks for nearby keys (I guess it needs to know the keyboard layout), and for missing or extra letters.

There are definitely two related concerns here:

- I mistyped the word.

- I don't know how to spell the word.

The errors in these two look quite different and may need different methods of correcting.

Same here, Apple often auto correct my word into something else, or failed to even suggest the correct word. I had to use Google to get a correct spelling. ( I am not sure if I am alone in this sometimes I dont think I can spell any more. )

Comparing the spellcheck quality of Firefox to Google (search, not Chrome) might not be completely fair. The browser most likely uses a client-side spelling model with tight memory & performance constraints. The suggestions from Google come from high quality models hosted on their servers.

https://en.wikipedia.org/wiki/Soundex#:~:text=Soundex%20is%2....

Soundex is a rough phonetic "spelling" used for such things. It's also something I never see mentioned.

Soundex was designed as a way to index last names before computers. It's primary benefit it that it's designed to do by hand. It performs pretty lousy as a general purpose natural language hashing scheme.

The precision and recall is too bad to use as a spell checker.

It's also pretty simple, I'm sure there are much better options readily available as libraries these days.

Firefox's spelling correction is very poor. For instance, try spelling the name of a country starting with a lowercase letter and omitting a single letter.

"argetina" suggests "retina", Argentina isn't in the list.

"nethrlands" suggests "fatherland", Netherlands isn't in the list.

"ukriane" suggests "hurricane", Ukraine isn't in the list.

It fails for every country I try.

Reply on news.ycombinator.com