Monolingual Features Sample Clauses

Monolingual Features. ‌ To capture basic monolingual parse quality, we begin with a single source and a single target feature whose values are the log likelihood of the source tree t and the target tree tr, respectively, as given by our baseline monolingual parsers. These two features are called SOURCELL and TARGETLL respectively. It is certainly possible to augment these simple features with what would amount to monolingual reranking features, but we do not explore that option here. Note that with only these two features, little can be learned: all positive weights θ cause the jointly optimal parse pair (t, tr) to comprise the two top-1 monolingual outputs (the baseline). NP NP PP NP JJ NNS IN NN CC NN 0.1 0 0.1 0.3 0 0.8 0 0.7 0.4 0.2 0 0 0.8 0.3 0.2 0 0.1 0.1 High levels of product and project a) InSrcOutTrg 0.4 InsideBoth 3.3 InTrgOutSrc 0.4 3.8 0.7 0.6 NP NP NN PU NP NP NN NN NP IP VA VP PP NP