SALM: Suffix Array tool kit for empirical Language Manipulations. By Joy, joy@cs.cmu.edu

1) Download the source code from: http://projectile.is.cs.cmu.edu/research/public/tools/salm/salm.htm or http://www.sourceforge.net/projects/salm 2) Build binaries: a) For Linux platform: cd Distribution/Linux make allO32 (for 32-bit platform) or make allO64 (for 64-bit platform)

    Binaries are created under Bin/Linux

b) For Win32 platform
    open project files under Distribution/Win32 and use Visual C++ to build executables.
    Executables are placed under Bin/Win32

3) Index a corpus. The first step is to index a corpus using IndexSA program. There is no limitation to the size of the corpus as long as there is enough RAM. A corpus of N words requires 9N bytes memory during indexing.

Another constraint is that no sentence can have more than 254 words.

Synposis of IndexSA:
    IndexSA corpusFileName [existingIDVocabularyFile]

Optional existingIDVocabularyFile can be used to specify an existing vocabulary.
It will be updated if the words in the corpus are new to the exising vocabulary.
This is useful if several corpora want to share a common vocabulary.

4) Applications The key functions to suffix array applications are provided in class C_SuffixArraySearchApplicationBase and C_SuffixArrayScanningBase Please check the documentation and API for more details.

Sample programs such as:

    FrequencyOfNgrams: 
        Output the frequency of an n-gram in the training corpus

    NGramMatchingStat4TestSet       
        Output the n-gram token matching statistics of a testing data

    NgramTypeInTestSetMatchedInCorpus
        Output the n-gram type matching statistics of a testing data

    NgramMatchingFreq4Sent
        Output the frequencies of all the embedded n-grams in a sentence

    NgramMatchingFreqAndNonCompositionality4Sent
        Output the non-compositionalities of the embedded n-grams in a sentence

    FilterDuplicatedSentences
        Filter out duplicated sentences in the training corpus and output the unique ones

    CollectNgramFreqCount
        Given a list of n-grams and a list of traing corpus indexed by their suffix array, collect counts of n-grams in these corpus. E.g. given a Chinese word list, one can collect the frequency of these words (as character n-grams) from several large corpora (segmented into characters).

    CalcCountOfCounts
        Output the count-of-counts information of a corpus

    OutputHighFreqNgram
        Specified by a configuration file, output the n-gram types that have frequencies higher than the threshold

    TypeTokenFreqInCorpus
        Output the type/token statistics of the corpus

5) Questions, comments and suggestions? Please email joy+salm@cs.cmu.edu