Learn More
Our approach enables: • accurate prediction of target translation stem and suffix given fixed amount of context • automatic learning of relevant features with neural network architecture Choosing the correct surface form requires linguistic features of source and target context: • in phrase-based SMT, access to source context depends on phrase segmentation(More)
We introduce a manually-created, multireference dataset for abstractive sentence and short paragraph compression. First, we examine the impact of singleand multi-sentence level editing operations on human compression quality as found in this corpus. We observe that substitution and rephrasing operations are more meaning preserving than other operations, and(More)
Lexical sparsity is a major challenge for machine translation into morphologically rich languages. We address this problem by modeling sequences of fine-grained morphological tags in a bilingual context. To overcome the issue of ambiguous word analyses, we introduce soft tags, which are under-specified representations retaining all possible morphological(More)
  • 1