NAIST-SIC-Aligned: an Aligned English-Japanese Simultaneous Interpretation Corpus

TL;DR

This work tackles the lack of large-scale parallel SI data by introducing NAIST-SIC-Aligned, an automatically aligned English–Japanese SI corpus derived from NAIST-SIC. A two-stage alignment pipeline (coarse grouping with vecAlign and LASER, followed by intra- and inter-sentence filtering) converts non-parallel SI data into parallel training material, with qualitative and quantitative validation at each step and a manually curated SI test set. Experiments with wait-k SiMT demonstrate that models trained on SI data substantially improve translation quality and latency relative to baselines, with the Inter and Inter$^{Srank}$ variants delivering the strongest gains. The dataset provides a practical resource for SiMT research and a methodological blueprint for constructing SI corpora across language pairs, with the data and testing framework available at the project site.

Abstract

It remains a question that how simultaneous interpretation (SI) data affects simultaneous machine translation (SiMT). Research has been limited due to the lack of a large-scale training corpus. In this work, we aim to fill in the gap by introducing NAIST-SIC-Aligned, which is an automatically-aligned parallel English-Japanese SI dataset. Starting with a non-aligned corpus NAIST-SIC, we propose a two-stage alignment approach to make the corpus parallel and thus suitable for model training. The first stage is coarse alignment where we perform a many-to-many mapping between source and target sentences, and the second stage is fine-grained alignment where we perform intra- and inter-sentence filtering to improve the quality of aligned pairs. To ensure the quality of the corpus, each step has been validated either quantitatively or qualitatively. This is the first open-sourced large-scale parallel SI dataset in the literature. We also manually curated a small test set for evaluation purposes. Our results show that models trained with SI data lead to significant improvement in translation quality and latency over baselines. We hope our work advances research on SI corpora construction and SiMT. Our data can be found at https://github.com/mingzi151/AHC-SI.

Figures (1)

• Figure 1: Translation quality and latency for wait-k systems trained on Must-C and various SI data.