Integrating Vectorized Lexical Constraints for Neural Machine Translation

We propose to integrate external lexical constraints for the Transformer model through additional attention prefixes to enable controllable neural machine translation. (Read More)

DirectQuote: A Dataset for Direct Quotation Extraction and Attribution in News Articles

We introduce DirectQuote, a new dataset to deal with quotation extraction and attribution. (Read More)

A Variant of Anderson Mixing with Minimal Memory Size

We develop a new variant of Anderson mixing that achieves minimal memory size and has promising results on deterministic optimization and training neural networks. (Read More)

Dynamic Multi-Branch Layers for On-Device Neural Machine Translation

How do we deploy NMT models on a Raspberry Pi 4 device with high inference efficiency? We answer this question in our TASLP paper. (Read More)