#Haspell ###Haskell spell correction based on minimum edit distance calculation ######Developed for a university course by Sebastian J. Mielke 2014
Haspell lets the user correct a text read from a file interactively by comparing words (using the minimum edit distance) to suggestions from a word list.
#Usage
Haspell "wordlist.txt" "myfile.txt"
will read the file wordlist.txt and construct a Trie from the whitespace-separated words in this file. Then Haspell will go through the second file and present the user with correction suggestions for every word it cannot find in the given word list. The user can ignore the spelling error, request more suggestions or exit the program. Once all words are checked, the corrected text is written to myfile.txt.corrected.
###Additional Flags
CLI help is available using the usual --help flag, but let's talk about the available options here, regardless.
Haspell "wordlist.txt" "myfile.txt" --compat
Usually the CLI will use ANSI terminal formatting. However, Windows users and users of less compatible terminals exist, so there is a --compat / -c option to turn off all formatting and instead highlight only the necessary parts of the UI using plain text cues.
Haspell "wordlist.txt" "myfile.txt" --outfile "correctedfile.txt"
The --outfile / -o option allows the user to specify where the corrected output file should be saved to.
#Documentation
The code is documented using standard Haddock syntax, so I will only give a broad overview here.
The WTrie module defines the Trie data structure used to hold the word list as well as methods to import word lists from files and lists.
The usage of a Trie allows us to progressively calculate the minimum edit distance (MED) to a given user word without needing to re-calulate prefixes for every word.
The TrieMED module does precisely that. The calcMEDs entry point calls one of the reduceTries methods, which in turn traverse the Trie, carrying a state that contains mainly the current column of the MED matrix.
At each Trie node the calcNode method is called to transform the old state and the current node into the new state corresponding to the node.
The step function inside finally checks whether insertion, deletion, substitution or reversal give the smallest possible distance and fills the field of the new state accordingly.
The Haspell module finally contains the main. The ArgParser package provides the parsing of the command line arguments (see Usage).
After loading word lists, the correctText function calls segmentText to separate words from punctuation and then maps the monadic action correctSentence over the list of sentences. Here the calcMEDs method explained above is called, the results are printed and user input decides what the function returns. From the list of thus changed sentences the new corrected text can be created by reconstructSentences and then saved to a file.