This implements cosine measure and compares two English sentences (code is Python with libraries from NLTK, NUMPY)

Cosine_Similarity_compare_sentences

import re
import nltk
from numpy import zeros,dot
from numpy.linalg import norm

# Get stop words
stop_words = [w.strip() for w in open('C:\FWs.txt','r').readlines()]

splitter = re.compile ( "[a-z\-']+", re.I )
stemmer = nltk.PorterStemmer()

def add_word(word,d):
#If word not in stop_words list, gets stemmed version of words and adds to dictionary (with count)
 w = word.lower() 
 if w not in stop_words:
  ws = stemmer.stem(w)
  d.setdefault(ws,0)
  d[ws] += 1

def doc_vec(doc,key_idx):
 vec = zeros(len(key_idx))
 for word in splitter.findall(doc):
  keydata=key_idx.get(stemmer.stem(word).lower(), None)
  if keydata: vec[keydata[0]] = 1
 return vec

def compare(doc1,doc2):

 # strip all punctuation but - and '
 # convert to lower case store word/occurance in dict
 all_words = dict()

 for dat in [doc1,doc2]:
  #print dat
  [add_word(w,all_words) for w in splitter.findall(dat)]
 
 # Build an index of keys so that we know the word positions for the vector
 key_idx = dict() # key-> ( position, count )
 keys = all_words.keys()
 keys.sort()
 #print keys
 for i in range(len(keys)):
  key_idx[keys[i]] = (i,all_words[keys[i]])
 del keys
 del all_words

 vec1=doc_vec(doc1, key_idx)
 vec2=doc_vec(doc2, key_idx)
 return float(dot(vec1,vec2) / (norm(vec1) * norm(vec2)))
 
 
if 1==1: 
 print "Running Test..." 
 
 doc1="I see food"
 doc2="I like sea food"
 
 print "Using Doc1: %s\n\nUsing Doc2: %s\n" % ( doc1, doc2 )
 print "Similarity %s" % compare(doc1,doc2)