Fix LC cutoff policy of text tiling

[richarddwang]

According to the paper (Hearst, Marti A.. “Text Tiling: Segmenting Text into Multi-paragraph Subtopic Passages.” Comput. Linguistics 23 (1997): 33-64.) that the NLTK TextTiling algorithm is probably based on,

One version of this function entails drawing a boundary only if the depth score exceeds s_hat - sigma (the liberal measure, LC). This function can be varied to achieve correspondingly varying precision/recall trade-offs. A higher precision but lower recall can be found by setting the limit to be depth scores exceeding s_hat - sigma/2 (the conservative measure, HC) instead of s_hat - sigma

I believe LC/HC uses a threshold which is (s_hat - sigma)/(s_hat - sigma/2) respectively.

[stevenbird]

@richarddwang - thanks for the fix (and answering my question about this conditional).

[richarddwang]

@stevenbird - Is it good to be merged? Or we need some extra modification?

[stevenbird]

Thanks @richarddwang. Sorry for the delay!