keyboard_layouts VS SymSpell

So it's possible to have a magic key for same-finger bigrams (SFBs) as seen in Magic Sturdy.

https://github.com/Ikcelaks/keyboard_layouts/blob/main/magic...

Regarding chording, you might find inspiration in the chording schemes in Ikcelaks’ Magic Sturdy, the Ardux project, and precondition's "steno-lite" combos.

Beyond reversing, Ikcelaks's Magic Sturdy layout is a showcase of what else can be done. It is a layout where the Alternate Repeat key is placed in easy reach on the right index finger. This is the "magic" in Magic Sturdy. This key is configured to remove SFBs and type common n-grams, or potentially do anything else you want: "this key can be fully dynamic based on the last key press and need not be constrained to logically reversing that keypress." I've been trying Magic Sturdy for the past few weeks, and definitely this is a fun idea.

I've lately been experimenting with variations on Oxey's Sturdy, which is an attractive layout for an unusually strong combination of low redirects and high rolls, yet keeping SFBs and pinky use low. A very interesting mod that I'm using now is Ikcelaks's Magic Sturdy. It incorporates the Alternate Repeat Key as a "magic" key whose function is determined adaptively based on the previous key. This is leveraged to get around the highest SFBs and as a shortcut to type common n-grams.

The SimSpell algorithm uses deletions to determine edit distance of the input query word compared to a dictionary of correctly spelled words. The Double Metaphone algorithm (or other phonetic algorithms) convert the words to phonetic versions (phonetic "hashes" basically), and you then search based on the input phonetic hash matching the dictionary of phonetic hashes.

Books: a. "Speech and Language Processing" by Daniel Jurafsky and James H. Martin (3rd Edition) - This book covers various aspects of natural language processing, including a section on spelling correction that provides a comprehensive introduction to the topic. b. "Foundations of Statistical Natural Language Processing" by Christopher D. Manning and Hinrich Schütze - This book provides an overview of statistical approaches in NLP, including a chapter on spelling correction. Articles: a. "How to Write a Spelling Corrector" by Peter Norvig - This article demonstrates the development of a simple spelling corrector using statistical algorithms. It's a great starting point for understanding the basics of spell checkers. (Link: https://norvig.com/spell-correct.html) b. "The Design of a Proofreading Software Service" by Michael D. Garris and James L. Blue - This article presents the design and implementation of a spelling correction system that can be integrated into various applications. (Link: https://www.nist.gov/system/files/documents/itl/iad/89403123.pdf) c. "A Fast and Flexible Spellchecker" by Atkinson, K. (2006) - This article details the design of a spell checker that uses a combination of rule-based and statistical approaches for improved performance. (Link: https://aspell.net/0.60.6.1/aspell-0.60.6.1.pdf) Online Resources: a. The Natural Language Toolkit (NLTK) - This is a popular Python library for natural language processing. It includes a spell checker module and various examples of how to use it. (Link: https://www.nltk.org/) b. SymSpell - This is an open-source spell checking library that uses a Symmetric Delete spelling correction algorithm for high performance and accuracy. The GitHub repository includes a detailed description of the algorithm and examples of how to use it. (Link: https://github.com/wolfgarbe/SymSpell) These resources should provide a solid foundation for understanding the design, algorithms, and usage of spell checkers. Happy learning!

Can this github.com/wolfgarbe/SymSpell or this github.com/ruby/did_you_mean or any of these github.com/topics/spell-check?o=desc&s=forks spellcheckers be used as an autocorrection software?

Do you absolutely need to use deep learning? There are tons of way faster autocorrect implementations that use levenshtein distances and non-DL techniques such as SymSpell or Norvig’s algorithm. DL is both expensive and requires tons of data to train on, I would stay away from that unless you’re doing it for your own enrichment or a school project.

This library claims to be orders of magnitude faster: https://github.com/wolfgarbe/SymSpell

If you want to do both at the same time (prefix search, allowing for misspellings), you can use a trie, but rather than just putting all your words in it, you can put everything in the "deletion neighborhood" of each word (that is, each possible variant of each word that has one character deleted), in an approach sort of like what's described here. Fair warning, though, that this gets a little hairy, and you'll have to decide how to weight prefix matches vs. misspellings in your rankings.

SymSpell: 1M times faster spelling correction\ (6 comments)