We have presented the AGRR-2019 gapping corpus for Russian. Our corpus contains 22.5k sentences, including 7.5k sentences with gapping and 15k relevant negative sentences. The corpus is multi-genre and social media texts form a quarter of it.
In our submission for the WMT18 Multimodal Translation Task, we experimented with the Transformer architecture for MMT. The experiments show that the Transformer architecture outperforms the RNN-based models.
We described SubGram, an extension of the Skip-gram model that considers also substrings of input words. The learned embeddings then better capture almost all morpho-syntactic relations tested on test set which we extended from original described in…
There aren't more papers