Computational Linguistics 8, Exercise 3: saying "unsure"

Code (run with py -X utf8, after the chapter's lesk code):

def lesk2(word, context):
    sc, b = lesk(word, context)
    top = sorted(sc.values(), reverse=True)
    return sc, ("unsure" if top[0] == top[1] else b)
print(lesk2("light", "she took the light"))
print(lesk2("light", "the light guides the boats at night")[1])

Output from the run:

({'lamp-light': 0, 'not-heavy': 0}, 'unsure')
lamp-light

The tie (0 against 0) now returns "unsure" instead of silently picking the
first sense. The sentence with real overlap (3 against 0) still returns
lamp-light.

Another fallback: the WSD article reports that always choosing the most
frequent sense scored 51.4% (SemEval-2007) and 57% (Senseval-2), so when the
context gives no evidence a system can return the most frequent sense
instead of guessing at random. I did not implement that here, since my two
glosses have no frequency counts.