MLAT : a set of natural language processing tools in OCAML

MLAT early version can be downloaded at
phimatex sourceforge file list.

A short presentation

I want to achieve something like that to obtain histograms for n-grams:

   let dico = new dictionary range default_parameters;;
   dico#add_to_forget_list file_name;;
   dico#add_source stream_name optional_parameters;;
   dico#info;; (* to get some informations *)
   dico#add_source stream_name optional_parameters;; (* to add another source *)
   dico#quantify forget_levels;;
   dico#info;;

and to quantify the corpus :

   dico#set_feature_space final_dimention internal_dimension;;
   dico#component_cloud format projective_dimension;; (* to get informations, format is in ASCII|XML|HTML|ASCII4MAT *)
   dico#sources_cloud   format projective_dimension;;

or in a more advanced version to produce automatic clustering :

   dico#learn_supervised_clustering new_key flags_set1 flags_set2 algorithm;; (* to clusterize set1 vs set2 *)
   dico#learn_info new_key;;
   let k = dico#add_source stream_name optional_parameters;; 
   dico#clusterize key set_of_source_keys;;

or to provide unsupervised clustering with something like :

   dico#learn_clusters new_key set_of_source_keys;; (* or replacing set_of_source_keys by set_of_flag_keys *)
   dico#learn_info new_key;;
   let k = dico#add_source stream_name optional_parameters;; 
   dico#clusterize key set_of_source_keys;;

some friends already asked me to be able to compute distances like in :

   let d = dico#distance_bewteen source1 source2;;
   let k = dico#closest_to source;;

TODO list

for add_source
- word neutralization
- updating components and flat_components
quantify
set_feature_space
clouds (and a small interface with visualization tools like Maple and Matlab)
supervised learning (one or two basic algorithms : k-means, SOM/Kohonen, SVM)
use of supervised learning
unsupervised learning (one or two simple algorithms too : mixing of gaussians, SOM/Kohonen)
use of unsupervised learning
distances computations

VERSION 0.1 - load it into OCAML with #use "MLAT.ml"

this is a site by CAL