RNNLG

Includes four original domains and counterfeited datasets for cross-domain adaptation, enabling standardized evaluation as detailed in Wen et al., 2016, which facilitates reproducible research.

Provides RNN-based generators like Semantically Conditioned LSTM and Attentive Encoder-Decoder LSTM, plus baselines (kNN, N-gram), allowing for comparative studies in NLG.

Supports Maximum Likelihood and Discriminative Training (Expected BLEU), offering different optimization objectives for model training, as shown in the configuration parameters.

Includes counterfeited datasets and adaptation methods for cross-domain scenarios, crucial for research on multi-domain spoken dialogue systems.