IBM 1

from collections import defaultdict, Counter
from itertools import chain

src = """dogs are in house .
dogs are in house .
dogs are in house .
dogs are in house .
cat is in house .
cat is in house .
is house in town ?
is house in town ?
is house in town ?
is house in town ?"""

trg = """koirat on talossa .
koira t on talo ssa .
koirat on talo ssa .
koira t on talossa .
kissa  on talossa .
kissa  on talo ssa .
onks talo kaupungissa ?
onks talo kaupunki ssa ?
on ko talo kaupungissa ?
on ko talo kaupunki ssa ?"""

'''
src = """dogs are in house .
cat is in house .
is house in town ?"""

trg = """koirat on talossa .
kissa  on talossa .
onks talo kaupungissa ?
"""
'''

# Convert into docstream.
src = [i.split() for i in src.split('\n')]
trg = [i.split() for i in trg.split('\n')]

src_vocab = Counter(chain(*src))
trg_vocab = Counter(chain(*trg))


num_probs = len(src_vocab) * len(trg_vocab)

default_prob = 1.0 / len(src_vocab)
t = defaultdict(lambda: default_prob) # probability table.

convergent_threshold=1e-2
globally_converged = False
probabilities = []
iteration_count = 0

while not globally_converged:
    count = defaultdict(float) # count(e|f)
    total = defaultdict(float) # total(f)

    for srcline, trgline in zip(src, trg):
        s_total = {} # Sum of probabilities for this sentence pair.
        for srcword in srcline:
            s_total[srcword] = 0.0
            for trgword in trgline:
                s_total[srcword] += t[srcword, trgword]


        for srcword in srcline:
            for trgword in trgline:
                # Normalize probabilities.
                cnt = t[srcword, trgword] / s_total[srcword]
                # Summing the prob of srcword given trgword.
                count[srcword, trgword] += cnt
                total[trgword] += cnt

    num_converged = 0
    for trgword in trg_vocab:
        for srcword in src_vocab:
            ##print engdeu_corpus.dictionary[trgword], engdeu_corpus.dictionary[srcword], count[srcword, trgword], total[trgword]
            new_prob = count[srcword, trgword] / total[trgword]
            delta = abs(t[srcword, trgword] - new_prob)
            if delta < convergent_threshold:
                num_converged += 1
            t[srcword, trgword] = new_prob

    iteration_count += 1
    if num_converged == num_probs:
        globally_converged = True

for i,j in sorted(t):
    print i,j, t[i,j]