1 PARALLEL CORPUS COLLECTIONTIVELY UNTIL CONVERGENCE. THE TEXTRANK...

3.3.1 Parallel Corpus Collection

tively until convergence. The TextRank score of a

word w in document D at kth iteration is defined as

In Q&A archives, question-answer pairs can be con-

follows:

sidered as a type of parallel corpus, which is used for

estimating the translation probabilities. Unlike the

i,j

^k

_w,D

bilingual machine translation, the questions and an-

∀

l:(j,l)

∈

G

_j,l

^k

_w,D

⁻

¹

∀

j:(i,j)

∈

G

swers in a Q&A archive are written in the same lan-

guage, the translation probability can be calculated

where d is a damping factor usually set to 0.85, and

through setting either as the source and the other as

e

_i,j

is an edge weight between i and j.

the target. In this paper, P (¯ a | q) ¯ is used to denote

We use average TextRank score as threshold:

the translation probability with the question as the

words are removed if their scores are lower than the

source and the answer as the target. P (¯ q| ¯ a) is used

average score of all words in a document.

to denote the opposite configuration.