Simultaneous Speech Translation (SiST) begins translating before the entire source input is received, making it crucial to balance quality and latency. In real interpreting situations, interpreters manage this simultaneity by breaking sentences into smaller segments and translating them while maintaining the source order as much as possible. SiST could benefit from this approach to balance quality and latency. However, current corpora used for simultaneous tasks often involve significant word reordering in translation, which is not ideal given that interpreters faithfully follow source syntax as much as possible. Inspired by conference interpreting by humans utilizing the salami technique, we introduce the Simul-MuST-C, a dataset created by leveraging the Large Language Model (LLM), specifically GPT-4o, which aligns the target text as closely as possible to the source text by using minimal chunks that contain enough information to be interpreted. Experiments on three language pairs show that the effectiveness of segmented-base monotonicity in training data varies with the grammatical distance between the source and the target, with grammatically distant language pairs benefiting the most in achieving quality while minimizing latency.

In Simultaneous Machine Translation (SiMT), training with a simultaneous interpretation (SI) corpus is an effective method for achieving high-quality yet low-latency. However, constructing such a corpus is challenging due to high costs, and limitations in annotator capabilities, and as a result, existing SI corpora are limited. Therefore, we propose a method to convert existing speech translation (ST) corpora into interpretation-style corpora, maintaining the original word order and preserving the entire source content using Large Language Models (LLM-SI-Corpus). We demonstrate that fine-tuning SiMT models using the LLM-SI-Corpus reduces latency while achieving better quality compared to models fine-tuned with other corpora in both speech-to-text and text-to-text settings. The LLM-SI-Corpus is available at https://github.com/yusuke1997/LLM-SI-Corpus.

In Japanese, loanwords are primarily written in Katakana, a syllabic writing system, based on their pronunciation. However, the transliterated loanwords often exhibit spelling variations, such as the word “Hepburn” being written as “ヘボン (hebon)”, “ヘプバーン (hepubaan)”, “ヘップバーン (heppubaan)”. These orthographical variants pose a bottleneck in multilingual Named Entity Recognition (NER), because named entities (NEs) do not have one-to-one matches. In this study, we introduce a rule-based grapheme-to-phoneme (G2P) system for Japanese based on literature in linguistics and a large-scale multilingual NE dataset with annotations of the International Phonetic Alphabet (IPA), focusing on IPA to address the Katakana spelling variations in loanwords. These rules and dataset are expected to be beneficial for tasks such as NE aggregation, G2P system, construction of cross-lingual language models, and entity linking. We hope our work advances research on Japanese NER with multilingual loanwords by solving the spelling ambiguities.

Minimum Bayes risk (MBR) decoding achieved state-of-the-art translation performance by using COMET, a neural metric that has a high correlation with human evaluation.However, MBR decoding requires quadratic time since it computes the expected score between a translation hypothesis and all reference translations.We propose centroid-based MBR (CBMBR) decoding to improve the speed of MBR decoding.Our method clusters the reference translations in the feature space, and then calculates the score using the centroids of each cluster.The experimental results show that our CBMBR not only improved the decoding speed of the expected score calculation 5.7 times, but also outperformed vanilla MBR decoding in translation quality by up to 0.5 COMET in the WMT’22 En↔Ja, En↔De, En↔Zh, and WMT’23 En↔Ja translation tasks.

It is very challenging to curate a dataset for language-specific knowledge and common sense in order to evaluate natural language understanding capabilities of language models. Due to the limitation in the availability of annotators, most current multilingual datasets are created through translation, which cannot evaluate such language-specific aspects. Therefore, we propose Multilingual CommonsenseQA (mCSQA) based on the construction process of CSQA but leveraging language models for a more efficient construction, e.g., by asking LM to generate questions/answers, refine answers and verify QAs followed by reduced human efforts for verification. Constructed dataset is a benchmark for cross-lingual language-transfer capabilities of multilingual LMs, and experimental results showed high language-transfer capabilities for questions that LMs could easily solve, but lower transfer capabilities for questions requiring deep knowledge or commonsense. This highlights the necessity of language-specific datasets for evaluation and training. Finally, our method demonstrated that multilingual LMs could create QA including language-specific knowledge, significantly reducing the dataset creation cost compared to manual creation. The datasets are available at https://huggingface.co/datasets/yusuke1997/mCSQA.

Knowledge graphs (KGs) consist of links that describe relationships between entities. Due to the difficulty of manually enumerating all relationships between entities, automatically completing them is essential for KGs. Knowledge Graph Completion (KGC) is a task that infers unseen relationships between entities in a KG. Traditional embedding-based KGC methods (e.g. RESCAL, TransE, DistMult, ComplEx, RotatE, HAKE, HousE, etc.) infer missing links using only the knowledge from training data. In contrast, the recent Pre-trained Language Model (PLM)-based KGC utilizes knowledge obtained during pre-training, which means it can estimate missing links between entities by reusing memorized knowledge from pre-training without inference. This part is problematic because building KGC models aims to infer unseen links between entities. However, conventional evaluations in KGC do not consider inference and memorization abilities separately. Thus, a PLM-based KGC method, which achieves high performance in current KGC evaluations, may be ineffective in practical applications. To address this issue, we analyze whether PLM-based KGC methods make inferences or merely access memorized knowledge. For this purpose, we propose a method for constructing synthetic datasets specified in this analysis and conclude that PLMs acquire the inference abilities required for KGC through pre-training, even though the performance improvements mostly come from textual information of entities and relations.

Automatic extraction of geographic information, including Location Referring Expressions (LREs), can aid humanities research in analyzing large collections of historical texts. In this study, to investigate how accurate pretrained Transformer language models (LMs) can extract LREs from historical texts, we evaluate two representative types of LMs, namely, masked language model and causal language model, using early modern and contemporary Japanese datasets. Our experimental results demonstrated the potential of contemporary LMs for historical texts, but also suggest the need for further model enhancement, such as pretraining on historical texts.

The natural language understanding (NLU) performance of large language models (LLMs) has been evaluated across various tasks and datasets. The existing evaluation methods, however, do not take into account the variance in scores due to differences in prompts, which leads to unfair evaluation and comparison of NLU performance. Moreover, evaluation designed for specific prompts is inappropriate for instruction tuning, which aims to perform well with any prompt. It is therefore necessary to find a way to measure NLU performance in a fair manner, considering score variance between different instruction templates. In this study, we provide English and Japanese cross-lingual datasets for evaluating the NLU performance of LLMs, which include multiple instruction templates for fair evaluation of each task, along with regular expressions to constrain the output format. Furthermore, we propose the Sharpe score as an evaluation metric that takes into account the variance in scores between templates. Comprehensive analysis of English and Japanese LLMs reveals that the high variance among templates has a significant impact on the fair evaluation of LLMs.

We participated in the constrained track for English-Japanese and Japanese-Chinese translations at the WMT 2024 General Machine Translation Task. Our approach was to generate a large number of sentence-level translation candidates and select the most probable translation using minimum Bayes risk (MBR) decoding and document-level large language model (LLM) re-ranking. We first generated hundreds of translation candidates from multiple translation models and retained the top 30 candidates using MBR decoding. In addition, we continually pre-trained LLMs on the target language corpora to leverage document-level information. We utilized LLMs to select the most probable sentence sequentially in context from the beginning of the document.

Large-scale Vision-Language Models (LVLMs) output text from images and instructions, demonstrating advanced capabilities in text generation and comprehension. However, it has not been clarified to what extent LVLMs understand the knowledge necessary for explaining images, the complex relationships between various pieces of knowledge, and how they integrate these understandings into their explanations. To address this issue, we propose a new task: the artwork explanation generation task, along with its evaluation dataset and metric for quantitatively assessing the understanding and utilization of knowledge about artworks. This task is apt for image description based on the premise that LVLMs are expected to have pre-existing knowledge of artworks, which are often subjects of wide recognition and documented information.It consists of two parts: generating explanations from both images and titles of artworks, and generating explanations using only images, thus evaluating the LVLMs’ language-based and vision-based knowledge.Alongside, we release a training dataset for LVLMs to learn explanations that incorporate knowledge about artworks.Our findings indicate that LVLMs not only struggle with integrating language and visual information but also exhibit a more pronounced limitation in acquiring knowledge from images alone. The datasets ExpArt=Explain Artworks are available at https://huggingface.co/datasets/naist-nlp/ExpArt

In this paper, we describe our NAIST-NICT submission to the WMT’23 English ↔ Japanese general machine translation task. Our system generates diverse translation candidates and reranks them using a two-stage reranking system to find the best translation. First, we generated 50 candidates each from 18 translation methods using a variety of techniques to increase the diversity of the translation candidates. We trained seven models per language direction using various combinations of hyperparameters. From these models we used various decoding algorithms, ensembling the models, and using kNN-MT (Khandelwal et al., 2021). We processed the 900 translation candidates through a two-stage reranking system to find the most promising candidate. In the first step, we compared 50 candidates from each translation method using DrNMT (Lee et al., 2021) and returned the candidate with the best score. We ranked the final 18 candidates using COMET-MBR (Fernandes et al., 2022) and returned the best score as the system output. We found that generating diverse translation candidates improved translation quality using the well-designed reranker model.

In this paper, we describe our NAIST-NICT-TIT submission to the WMT22 general machine translation task. We participated in this task for the English ↔ Japanese language pair. Our system is characterized as an ensemble of Transformer big models, k-nearest-neighbor machine translation (kNN-MT) (Khandelwal et al., 2021), and reranking.In our translation system, we construct the datastore for kNN-MT from back-translated monolingual data and integrate kNN-MT into the ensemble model. We designed a reranking system to select a translation from the n-best translation candidates generated by the translation system. We also use a context-aware model to improve the document-level consistency of the translation.

We introduce a Cyrillic-to-Latin transliterator for the Tatar language based on subword-level language identification. The transliteration is a challenging task due to the following two reasons. First, because modern Tatar texts often contain intra-word code-switching to Russian, a different transliteration set of rules needs to be applied to each morpheme depending on the language, which necessitates morpheme-level language identification. Second, the fact that Tatar is a low-resource language, with most of the texts in Cyrillic, makes it difficult to prepare a sufficient dataset. Given this situation, we proposed a transliteration method based on subword-level language identification. We trained a language classifier with monolingual Tatar and Russian texts, and applied different transliteration rules in accord with the identified language. The results demonstrate that our proposed method outscores other Tatar transliteration tools, and imply that it correctly transcribes Russian loanwords to some extent.