Today's large language models (LLMs) are capable of supporting multilingual scenarios, allowing users to interact with LLMs in their native languages. When LLMs respond to subjective questions posed by users, they are expected to align with the views of specific demographic groups or historical periods, shaped by the language in which the user interacts with the model. Existing studies mainly focus on researching the opinions represented by LLMs among demographic groups in the United States or a few countries, lacking worldwide country samples and studies on human opinions in different historical periods, as well as lacking discussion on using language to steer LLMs. Moreover, they also overlook the potential influence of prompt language on the alignment of LLMs' opinions. In this study, our goal is to fill these gaps. To this end, we create an evaluation framework based on the World Values Survey (WVS) to systematically assess the alignment of LLMs with human opinions across different countries, languages, and historical periods around the world. We find that LLMs appropriately or over-align the opinions with only a few countries while under-aligning the opinions with most countries. Furthermore, changing the language of the prompt to match the language used in the questionnaire can effectively steer LLMs to align with the opinions of the corresponding country more effectively than existing steering methods. At the same time, LLMs are more aligned with the opinions of the contemporary population. To our knowledge, our study is the first comprehensive investigation of the topic of opinion alignment in LLMs across global, language, and temporal dimensions.

Evaluating meeting effectiveness is crucial for improving organizational productivity.Current approaches rely on post-hoc surveys that yield a single coarse-grained score for an entire meeting.The reliance on manual assessment is inherently limited in scalability, cost, and reproducibility.Moreover, a single score fails to capture the dynamic nature of collaborative discussions.We propose a new paradigm for evaluating meeting effectiveness centered on novel criteria and temporal fine-grained approach.We define effectiveness as the rate of objective achievement over time and assess it for individual topical segments within a meeting.To support this task, we introduce the AMI Meeting Effectiveness (AMI-ME) dataset, a new meta-evaluation dataset containing 2,459 human-annotated segments from 130 AMI Corpus meetings.We also develop an automatic effectiveness evaluation framework that uses a Large Language Model (LLM) as a judge to score each segment's effectiveness relative to the overall meeting objectives.Through substantial experiments, we establish a comprehensive benchmark for this new task and evaluate the framework's generalizability across distinct meeting types, ranging from business scenarios to unstructured discussions.Furthermore, we benchmark end-to-end performance starting from raw speech to measure the capabilities of a complete system.Our results validate the framework's effectiveness and provide strong baselines to facilitate future research in meeting analysis and multi-party dialogue.Our dataset and code will be publicly available.

Prompt sensitivity, which refers to how strongly the output of a large language model (LLM) depends on the exact wording of its input prompt, raises concerns among users about the LLM's stability and reliability.In this work, we consider LLMs as multivariate functions and perform a first-order Taylor expansion, thereby analyzing the relationship between meaningpreserving prompts, their gradients, and the log probabilities of the model's next token.We derive an upper bound on the difference between log probabilities using the Cauchy-Schwarz inequality.We show that LLMs do not internally cluster similar inputs like smaller neural networks do, but instead disperse them.This dispersing behavior leads to an excessively high upper bound on the difference of log probabilities between two meaning-preserving prompts, making it difficult to effectively reduce to 0. In our analysis, we also show which types of meaning-preserving prompt variants are more likely to introduce prompt sensitivity risks in LLMs.In addition, we demonstrate that the upper bound is strongly correlated with an existing prompt sensitivity metric, PromptSensiScore.Moreover, by analyzing the logit variance, we find that prompt templates typically exert a greater influence on logits than the questions themselves.Overall, our results provide a general interpretation for why current LLMs can be highly sensitive to prompts with the same meaning, offering crucial evidence for understanding the prompt sensitivity of LLMs.Code for experiments is available at https://github.com/ku-nlp/ Understanding_the_Prompt_Sensitivity.

Zhen Wan, Chao-Han Huck Yang, Jinchuan Tian, Hanrong Ye, Ankita Pasad, Szu-Wei Fu, Arushi Goel, Ryo Hachiuma, Shizhe Diao, Kunal Dhawan, Sreyan Ghosh, Yusuke Hirota, Zhehuai Chen, Rafael Valle, Chenhui Chu, Shinji Watanabe, Boris Ginsburg, Yu-Chiang Frank Wang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

This study investigates the interaction between personality traits and emotion expression, exploring how personality information can improve speech emotion recognition (SER). We collect the personality annotation for the IEMOCAP dataset, making it the first speech dataset that contains both emotion and personality annotations (PA-IEMOCAP), and enabling direct integration of personality traits into SER. Statistical analysis on this dataset identified significant correlations between per sonality traits and emotional expressions. To extract finegrained personality features, we propose a temporal interaction condition network (TICN), in which personality features are integrated with HuBERT-based acoustic features for SER. Experiments show that incorporating ground-truth personality traits significantly enhances valence recognition, improving the concordance correlation coefficient (CCC) from 0.698 to 0.785 compared to the baseline without personality information. For practical applications in dialogue systems where personality information about the user is unavailable, we develop a front-end module of automatic personality recognition. Using these automatically predicted traits as inputs to our proposed TICN model, we achieve a CCC of 0.776 for valence recognition, representing an 11.17% relative improvement over the baseline. These findings confirm the effectiveness of personality-aware SER and provide a solid foundation for further exploration in personality-aware speech processing applications.

Zhen Wan, Chao-Han Huck Yang, Yahan Yu, Jinchuan Tian, Sheng Li, Ke Hu, Zhehuai Chen, Shinji Watanabe, Fei Cheng, Chenhui Chu, Sadao Kurohashi. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Large language models (LLMs) can rewrite the N -best hypotheses from a speech-to-text model, often fixing recognition or translation errors that traditional rescoring cannot.Yet research on generative error correction (GER) has been focusing on monolingual automatic speech recognition (ASR), leaving its multilingual and multitask potential underexplored.We introduce CoVoGER, a benchmark for GER that covers both ASR and speech-to-text translation (ST) across 15 languages and 28 language pairs.CoVoGER is constructed by decoding Common Voice 20.0 and CoVoST-2 with Whisper of three model sizes and Seam-lessM4T of two model sizes, providing 5-best lists obtained via a mixture of beam search and temperature sampling.We evaluated various instruction-tuned LLMs, including commercial models in zero-shot mode and open-sourced models with LoRA fine-tuning, and found that the mixture decoding strategy yields the best GER performance in most settings.CoVoGER will be released to promote research on reliable language-universal speech-to-text GER.

This paper extends Whisper’s automatic speech recognition (ASR) capabilities to perform speech-based emotion recognition (SER) by incorporating word-level emotion classification alongside ASR output. We generate four emotion pseudo-labels (neutral, happy, sad, angry) for each word using a pretrained frame-level SER model, and Whisper is fine-tuned for joint ASR and emotion classification at the word level. Sentence-level emotion labels are masked during training to encourage the transformer to use the ASR output for word-level emotion prediction. During inference, word-level predictions are combined with sentence-level predictions through majority voting to generate the final sentence-level label. When evaluated on the IEMOCAP dataset, our method maintains Whisper’s ASR word error rate while improving the SER weighted accuracy from 74.4% to 76.4% and the unweighted average recall from 77.1% to 79.0%.

Abstract The use of AI systems for managing large-scale cultural heritage artifacts has become possible due to the rise of digitization. To classify such content, machine learning is typically used, where contextual information is important to structure the data. One way to capture context is through a knowledge graph. In this study, we propose a newd graph neural networks, we can improve artwork classification by utilizing the relationships between entities of the knowledge graph. Our experiments demonstrate that this approach achieves state-of-the-art results on multiple classification tasks across three datasets (SemArt paintings, Buddha statues, and Ukiyo-e woodblock prints). Moreover, our approach is effective in dealing with unbalanced data and we explore the use of both graph attention mechanisms and focal loss functions.

Machine translation-mediated communication can benefit from pre-editing source language texts to ensure accurate transmission of intended meaning in the target language. The primary challenge lies in identifying source language expressions that pose difficulties in translation. In this paper, we hypothesize that such expressions tend to be distinctive features of texts originally written in the source language (native language) rather than translations generated from the target language into the source language (machine translation). To identify such expressions, we train a neural classifier to distinguish native language from machine translation, and subsequently isolate the expressions that contribute to the model’s prediction of native language. Our manual evaluation revealed that our method successfully identified characteristic expressions of the native language, despite the noise and the inherent nuances of the task. We also present case studies where we edit the identified expressions to improve translation quality.

The creation of instruction data and evaluation benchmarks for serving Large language models often involves enormous human annotation. This issue becomes particularly pronounced when rapidly developing such resources for a non-English language like Japanese. Instead of following the popular practice of directly translating existing English resources into Japanese (e.g., Japanese-Alpaca), we propose an efficient self-instruct method based on GPT-4. We first translate a small amount of English instructions into Japanese and post-edit them to obtain native-level quality. GPT-4 then utilizes them as demonstrations to automatically generate Japanese instruction data. We also construct an evaluation benchmark containing 80 questions across 8 categories, using GPT-4 to automatically assess the response quality of LLMs without human references. The empirical results suggest that the models fine-tuned on our GPT-4 self-instruct data significantly outperformed the Japanese-Alpaca across all three base pre-trained models. Our GPT-4 self-instruct data allowed the LLaMA 13B model to defeat GPT-3.5 (Davinci-003) with a 54.37% win-rate. The human evaluation exhibits the consistency between GPT-4’s assessments and human preference. Our high-quality instruction data and evaluation benchmark are released here.

This paper introduces a new task, abstractive multi-video captioning, which focuses on abstracting multiple videos with natural language. Unlike conventional video captioning tasks generating a specific caption for a video, our task generates an abstract caption of the shared content in a video group containing multiple videos. To address our task, models must learn to understand each video in detail and have strong abstraction abilities to find commonalities among videos. We construct a benchmark dataset for abstractive multi-video captioning named AbstrActs. AbstrActs contains 13.5k video groups and corresponding abstract captions. As abstractive multi-video captioning models, we explore two approaches: end-to-end and cascade. For evaluation, we proposed a new metric, CocoA, which can evaluate the model performance based on the abstractness of the generated captions. In experiments, we report the impact of the way of combining multiple video features, the overall model architecture, and the number of input videos.

Key-value relations are prevalent in Visually-Rich Documents (VRDs), often depicted in distinct spatial regions accompanied by specific color and font styles. These non-textual cues serve as important indicators that greatly enhance human comprehension and acquisition of such relation triplets. However, current document AI approaches often fail to consider this valuable prior information related to visual and spatial features, resulting in suboptimal performance, particularly when dealing with limited examples. To address this limitation, our research focuses on few-shot relational learning, specifically targeting the extraction of key-value relation triplets in VRDs. Given the absence of a suitable dataset for this task, we introduce two new few-shot benchmarks built upon existing supervised benchmark datasets. Furthermore, we propose a variational approach that incorporates relational 2D-spatial priors and prototypical rectification techniques. This approach aims to generate relation representations that are more aware of the spatial context and unseen relation in a manner similar to human perception. Experimental results demonstrate the effectiveness of our proposed method by showcasing its ability to outperform existing methods. This study also opens up new possibilities for practical applications.

Subword regularized models leverage multiple subword tokenizations of one target sentence during training. However, selecting one tokenization during inference leads to the underutilization of knowledge learned about multiple tokenizations.We propose the SubMerge algorithm to rescue the ignored Subword tokenizations through merging equivalent ones during inference.SubMerge is a nested search algorithm where the outer beam search treats the word as the minimal unit, and the inner beam search provides a list of word candidates and their probabilities, merging equivalent subword tokenizations. SubMerge estimates the probability of the next word more precisely, providing better guidance during inference.Experimental results on six low-resource to high-resource machine translation datasets show that SubMerge utilizes a greater proportion of a model’s probability weight during decoding (lower word perplexities for hypotheses). It also improves BLEU and chrF++ scores for many translation directions, most reliably for low-resource scenarios. We investigate the effect of different beam sizes, training set sizes, dropout rates, and whether it is effective on non-regularized models.

This study investigates the effective finetuning of a pretrained model using adapters for speech emotion recognition (SER). Since emotion is related with linguistic and prosodic information and also other attributes such as gender and speaking style, a framework of multi-task learning (MTL) has been shown to be effective for SER. However, the learning targets of automatic speech recognition (ASR) and other attribute recognition are apparently in conflict. Therefore, we propose to employ different adaptation methods for different tasks in multiple finetuning stages. Since ASR is the most challenging and also influential for SER, in the first stage, we finetune all parameters of the pretrained model for ASR and SER. In the second stage, we incorporate adapters to finetune the model for gender and style recognition in addition to SER by freezing the parameters of the main Transformer model tuned for ASR. Experimental evaluations which extensively compare different adaptation methods using the IEMOCAP dataset demonstrate that the proposed approach achieves a significant improvement from the simple MTL.

IEEE Automatic Mean Opinion Score (MOS) prediction is employed to evaluate the quality of synthetic speech. This study extends the application of predicted MOS to the task of Fake Audio Detection (FAD) as we expect that MOS can be used to assess how close synthesized speech is to the natural human voice. We propose MOS-FAD, where MOS can be leveraged at two key points in FAD: training data selection and model fusion. In training data selection, we demonstrate that MOS enables effective filtering of samples from unbalanced datasets. In the model fusion, our results demonstrate that incorporating MOS as a gating mechanism in FAD model fusion enhances overall performance.

Recent approaches for empathetic response generation mainly focus on emotional resonance and user understanding, without considering the system's personality.Consistent personality is evident in real human expression and is important for creating trustworthy systems.To address this problem, we propose StyEmp, which aims to stylize the empathetic response generation with a consistent personality.Specifically, it incorporates a multi-grained prefix mechanism designed to capture the intricate relationship between a system's personality and its empathetic expressions.Furthermore, we introduce a personality reinforcement module that leverages contrastive learning to calibrate the generation model, ensuring that responses are both empathetic and reflective of a distinct personality.Automatic and human evaluations on the EMPATHETICDIALOGUES benchmark show that StyEmp outperforms competitive baselines in terms of both empathy and personality expressions.Our code is available at https://github.com/fuyahuii/StyEmp.

Massively multilingual sentence representation models, e.g., LASER, SBERT-distill, and LaBSE, help significantly improve cross-lingual downstream tasks. However, the use of a large amount of data or inefficient model architectures results in heavy computation to train a new model according to our preferred languages and domains. To resolve this issue, we introduce efficient and effective massively multilingual sentence embedding (EMS), using cross-lingual token-level reconstruction (XTR) and sentence-level contrastive learning as training objectives. Compared with related studies, the proposed model can be efficiently trained using significantly fewer parallel sentences and GPU computation resources. Empirical results showed that the proposed model significantly yields better or comparable results with regard to cross-lingual sentence retrieval, zero-shot cross-lingual genre classification, and sentiment classification. Ablative analyses demonstrated the efficiency and effectiveness of each component of the proposed model. We release the codes for model training and the EMS pre-trained sentence embedding model, which supports 62 languages (https://github.com/Mao-KU/EMS).

The early-stage Alzheimer’s disease (AD) detection has been considered an important field of medical studies. Like traditional machine learning methods, speech-based automatic detection also suffers from data privacy risks because the data of specific patients are exclusive to each medical institution. A common practice is to use federated learning to protect the patients’ data privacy. However, its distributed learning process also causes performance reduction. To alleviate this problem while protecting user privacy, we propose a federated contrastive pre-training (FedCPC) performed before federated training for AD speech detection, which can learn a better representation from raw data and enables different clients to share data in the pre-training and training stages. Experimental results demonstrate that the proposed methods can achieve satisfactory performance while preserving data privacy.

Existing multimodal machine translation (MMT) datasets consist of images and video captions or instructional video subtitles, which rarely contain linguistic ambiguity, making visual information ineffective in generating appropriate translations.Recent work has constructed an ambiguous subtitles dataset to alleviate this problem but is still limited to the problem that videos do not necessarily contribute to disambiguation.We introduce EVA (Extensive training set and Videohelpful evaluation set for Ambiguous subtitles translation), an MMT dataset containing 852k Japanese-English (Ja-En) parallel subtitle pairs, 520k Chinese-English (Zh-En) parallel subtitle pairs, and corresponding video clips collected from movies and TV episodes.In addition to the extensive training set, EVA contains a video-helpful evaluation set in which subtitles are ambiguous, and videos are guaranteed helpful for disambiguation.Furthermore, we propose SAFA, an MMT model based on the Selective Attention model with two novel methods: Frame attention loss and Ambiguity augmentation, aiming to use videos in EVA for disambiguation fully.Experiments on EVA show that visual information and the proposed methods can boost translation performance, and our model performs significantly better than existing MMT models.The EVA dataset and the SAFA model are available at: https://github.com/ku-nlp/video-helpful-MMT.git.

Extracting meaningful entities belonging to predefined categories from Visually-rich Formlike Documents (VFDs) is a challenging task.Visual and layout features such as font, background, color, and bounding box location and size provide important cues for identifying entities of the same type.However, existing models commonly train a visual encoder with weak cross-modal supervision signals, resulting in a limited capacity to capture these nontextual features and suboptimal performance.In this paper, we propose a novel Visually-Asymmetric coNsistenCy Learning (VANCL) approach that addresses the above limitation by enhancing the model's ability to capture finegrained visual and layout features through the incorporation of color priors.Experimental results on benchmark datasets show that our approach substantially outperforms the strong LayoutLM series baseline, demonstrating the effectiveness of our approach.Additionally, we investigate the effects of different color schemes on our approach, providing insights for optimizing model performance.We believe our work will inspire future research on multimodal information extraction.

We address the effective finetuning of a large-scale pretrained model for automatic speech recognition (ASR) of lowresource languages with only a one-hour matched dataset. The finetuning is composed of domain adaptation and language adaptation, and they are conducted by using heterogeneous datasets, which are matched with either domain or language. For effective adaptation, we incorporate auxiliary tasks of domain identification and language identification with multi-task learning. Moreover, the embedding result of the auxiliary tasks is fused to the encoder output of the pretrained model for ASR. Experimental evaluations on the Khmer ASR using the corpus of ECCC (the Extraordinary Chambers in the Courts of Cambodia) demonstrate that first conducting domain adaption and then language adaption is effective. In addition, multi-tasking with domain identification and fusing the domain ID embedding gives the best performance, which is a CER improvement of 6.47% absolute from the baseline finetuning method.

Low-resource speech recognition has been long-suffering from insufficient training data. In this paper, we propose an approach that leverages neighboring languages to improve low-resource scenario performance, founded on the hypothesis that similar linguistic units in neighboring languages exhibit comparable term frequency distributions, which enables us to construct a Huffman tree for performing multilingual hierarchical Softmax decoding. This hierarchical structure enables cross-lingual knowledge sharing among similar tokens, thereby enhancing low-resource training outcomes. Empirical analyses demonstrate that our method is effective in improving the accuracy and efficiency of low-resource speech recognition.

Recent approaches to empathetic response generation try to incorporate commonsense knowledge or reasoning about the causes of emotions to better understand the user's experiences and feelings.However, these approaches mainly focus on understanding the causalities of context from the user's perspective, ignoring the system's perspective.In this paper, we propose a commonsense-based causality explanation approach for diverse empathetic response generation that considers both the user's perspective (user's desires and reactions) and the system's perspective (system's intentions and reactions).We enhance ChatGPT's ability to reason for the system's perspective by integrating in-context learning with commonsense knowledge.Then, we integrate the commonsense-based causality explanation with both ChatGPT and a T5-based model.Experimental evaluations demonstrate that our method outperforms other comparable methods on both automatic and human evaluations.

Sub-word segmentation is an essential pre-processing step for Neural Machine Translation (NMT). Existing work has shown that neural sub-word segmenters are better than Byte-Pair Encoding (BPE), however, they are inefficient, as they require parallel corpora, days to train, and hours to decode. This article introduces SelfSeg, a self-supervised neural sub-word segmentation method that is much faster to train/decode and requires only monolingual dictionaries instead of parallel corpora. SelfSeg takes as input a word in the form of a partially masked character sequence, optimizes the word generation probability, and generates the segmentation with the maximum posterior probability, which is calculated using a dynamic programming algorithm. The training time of SelfSeg depends on word frequencies, and we explore several word frequency normalization strategies to accelerate the training phase. Additionally, we propose a regularization mechanism that allows the segmenter to generate various segmentations for one word. To show the effectiveness of our approach, we conduct MT experiments in low-, middle-, and high-resource scenarios, where we compare the performance of using different segmentation methods. The experimental results demonstrate that, on the low-resource ALT dataset, our method achieves more than 1.2 BLEU score improvement compared with BPE and SentencePiece, and a 1.1 score improvement over Dynamic Programming Encoding (DPE) and Vocabulary Learning via Optimal Transport (VOLT), on average. The regularization method achieves approximately a 4.3 BLEU score improvement over BPE and a 1.2 BLEU score improvement over BPE-dropout, the regularized version of BPE. We also observed significant improvements on IWSLT15 Vi→En, WMT16 Ro→En, and WMT15 Fi→En datasets and competitive results on the WMT14 De→En and WMT14 Fr→En datasets. Furthermore, our method is 17.8× faster during training and up to 36.8× faster during decoding in a high-resource scenario compared to DPE. We provide extensive analysis, including why monolingual word-level data is enough to train SelfSeg.

We participated in the mean opinion score (MOS) prediction challenge, 2022.This challenge aims to predict MOS scores of synthetic speech on two tracks, the main track and a more challenging sub-track: out-of-domain (OOD).To improve the accuracy of the predicted scores, we have explored several model fusion-related strategies and proposed a fused framework in which seven pretrained self-supervised learned (SSL) models have been engaged.These pretrained SSL models are derived from three ASR frameworks, including Wav2Vec, Hubert, and WavLM.For the OOD track, we followed the 7 SSL models selected on the main track and adopted a semi-supervised learning method to exploit the unlabeled data.According to the official analysis results, our system has achieved 1 st rank in 6 out of 16 metrics and is one of the top 3 systems for 13 out of 16 metrics.Specifically, we have achieved the highest LCC, SRCC, and KTAU scores at the system level on main track, as well as the best performance on the LCC, SRCC, and KTAU evaluation metrics at the utterance level on OOD track.Compared with the basic SSL models, the prediction accuracy of the fused system has been largely improved, especially on OOD sub-track.

Existing multimodal machine translation (MMT) datasets consist of images and video captions or general subtitles which rarely contain linguistic ambiguity, making visual information not so effective to generate appropriate translations. We introduce VISA, a new dataset that consists of 40k Japanese-English parallel sentence pairs and corresponding video clips with the following key features: (1) the parallel sentences are subtitles from movies and TV episodes; (2) the source subtitles are ambiguous, which means they have multiple possible translations with different meanings; (3) we divide the dataset into Polysemy and Omission according to the cause of ambiguity. We show that VISA is challenging for the latest MMT system, and we hope that the dataset can facilitate MMT research.

Visually grounded paraphrases (VGPs) are different phrasal expressions describing the same visual concept in an image. Previous studies treat VGP identification as a binary classification task, which ignores various phenomena behind VGPs (i.e., different linguistic interpretation of the same visual concept) such as linguistic paraphrases and VGPs from different aspects. In this paper, we propose semantic typology for VGPs, aiming to elucidate the VGP phenomena and deepen the understanding about how human beings interpret vision with language. We construct a large VGP dataset that annotates the class to which each VGP pair belongs according to our typology. In addition, we present a classification model that fuses language and visual features for VGP classification on our dataset. Experiments indicate that joint language and vision representation learning is important for VGP classification. We further demonstrate that our VGP typology can boost the performance of visually grounded textual entailment.

We propose a multimodal neural machine translation (MNMT) method with semantic image regions called region-attentive multimodal neural machine translation (RA-NMT). Existing studies on MNMT have mainly focused on employing global visual features or equally sized grid local visual features extracted by convolutional neural networks (CNNs) to improve translation performance. However, they neglect the effect of semantic information captured inside the visual features. This study utilizes semantic image regions extracted by object detection for MNMT and integrates visual and textual features using two modality-dependent attention mechanisms. The proposed method was implemented and verified on two neural architectures of neural machine translation (NMT): recurrent neural network (RNN) and self-attention network (SAN). Experimental results on different language pairs of Multi30k dataset show that our proposed method improves over baselines and outperforms most of the state-of-the-art MNMT methods. Further analysis demonstrates that the proposed method can achieve better translation performance because of its better visual feature use.

In the present study, we propose novel sequence-to-sequence pre-training objectives for low-resource machine translation (NMT): Japanese-specific sequence to sequence (JASS) for language pairs involving Japanese as the source or target language, and English-specific sequence to sequence (ENSS) for language pairs involving English. JASS focuses on masking and reordering Japanese linguistic units known as bunsetsu, whereas ENSS is proposed based on phrase structure masking and reordering tasks. Experiments on ASPEC Japanese–English & Japanese–Chinese, Wikipedia Japanese–Chinese, News English–Korean corpora demonstrate that JASS and ENSS outperform MASS and other existing language-agnostic pre-training methods by up to +2.9 BLEU points for the Japanese–English tasks, up to +7.0 BLEU points for the Japanese–Chinese tasks and up to +1.3 BLEU points for English–Korean tasks. Empirical analysis, which focuses on the relationship between individual parts in JASS and ENSS, reveals the complementary nature of the subtasks of JASS and ENSS. Adequacy evaluation using LASER, human evaluation, and case studies reveals that our proposed methods significantly outperform pre-training methods without injected linguistic knowledge and they have a larger positive impact on the adequacy as compared to the fluency.

We propose word-region alignment-guided multimodal neural machine translation (MNMT), a novel model for MNMT that links the semantic correlation between textual and visual modalities using word-region alignment (WRA). Existing studies on MNMT have mainly focused on the effect of integrating visual and textual modalities. However, they do not leverage the semantic relevance between the two modalities. We advance the semantic correlation between textual and visual modalities in MNMT by incorporating WRA as a bridge. This proposal has been implemented on two mainstream architectures of neural machine translation (NMT): the recurrent neural network (RNN) and the transformer. Experiments on two public benchmarks, English–German and English–French translation tasks using the Multi30k dataset and English–Japanese translation tasks using the Flickr30kEnt-JP dataset prove that our model has a significant improvement with respect to the competitive baselines across different evaluation metrics and outperforms most of the existing MNMT models. For example, 1.0 BLEU scores are improved for the English–German task and 1.1 BLEU scores are improved for the English–French task on the Multi30k test2016 set; and 0.7 BLEU scores are improved for the English–Japanese task on the Flickr30kEnt-JP test set. Further analysis demonstrates that our model can achieve better translation performance by integrating WRA, leading to better visual information use.

Zhuoyuan Mao, Prakhar Gupta, Chenhui Chu, Martin Jaggi, Sadao Kurohashi. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

The rise of digitization of cultural documents offers large-scale contents, opening the road for development of AI systems in order to preserve, search, and deliver cultural heritage. To organize such cultural content also means to classify them, a task that is very familiar to modern computer science. Contextual information is often the key to structure such real world data, and we propose to use it in form of a knowledge graph. Such a knowledge graph, combined with content analysis, enhances the notion of proximity between artworks so it improves the performances in classification tasks. In this paper, we propose a novel use of a knowledge graph, that is constructed on annotated data and pseudo-labeled data. With label propagation, we boost artwork classification by training a model using a graph convolutional network, relying on the relationships between entities of the knowledge graph. Following a transductive learning framework, our experiments show that relying on a knowledge graph modeling the relations between labeled data and unlabeled data allows to achieve state-of-the-art results on multiple classification tasks on a dataset of paintings, and on a dataset of Buddha statues. Additionally, we show state-of-the-art results for the difficult case of dealing with unbalanced data, with the limitation of disregarding classes with extremely low degrees in the knowledge graph.

Tomoyuki Kajiwara, Chenhui Chu, Noriko Takemura, Yuta Nakashima, Hajime Nagahara. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

Humor is a very important communication tool; yet, it is an open problem for machines to understand humor. In this paper, we build a new multimodal dataset for humor prediction that includes subtitles and video frames, as well as humor labels associated with video's timestamps. On top of it, we present a model to predict whether a subtitle causes laughter. Our model uses the visual modality through facial expression and character name recognition, together with the verbal modality, to explore how the visual modality helps. In addition, we use an attention mechanism to adjust the weight for each modality to facilitate humor prediction. Interestingly, our experimental results show that the performance boost by combinations of different modalities, and the attention mechanism and the model mostly relies on the verbal modality.

With the help of the detailed annotated question answering dataset HotpotQA, recent question answering models are trained to justify their predicted answers with supporting facts from context documents. Some related works train the same model to find supporting facts and answers jointly without having specialized models for each task. The others train separate models for each task, but do not use supporting facts effectively to find the answer; they either use only the predicted sentences and ignore the remaining context, or do not use them at all. Furthermore, while complex graph-based models consider the bridge/connection between documents in the multi-hop setting, simple BERT-based models usually drop it. We propose FlexibleFocusedReader (FFReader), a model that 1) Flexibly focuses on predicted supporting facts (SFs) without ignoring the important remaining context, 2) Focuses on the bridge between documents, despite not using graph architectures, and 3) Jointly learns predicting SFs and answering with two specialized models. Our model achieves consistent improvement over the baseline. In particular, we find that flexibly focusing on SFs is important, rather than ignoring remaining context or not using SFs at all for finding the answer. We also find that tagging the entity that links the documents at hand is very beneficial. Finally, we show that joint training is crucial for FFReader.

The difference in word orders between source and target languages is a serious hurdle for machine translation. Preordering methods, which reorder the words in a source sentence before translation to obtain a similar word ordering with a target language, significantly improve the quality in statistical machine translation. While the information on the preordering position improved the translation quality in recurrent neural network-based models, questions such as how to use preordering information and whether it is helpful for the Transformer model remain unaddressed. In this article, we successfully employed preordering techniques in the Transformer-based neural machine translation. Specifically, we proposed a novel preordering encoding that exploits the reordering information of the source and target sentences as positional encoding in the Transformer model. Experimental results on ASPEC Japanese-English and WMT 2015 English-German, English-Czech, and English-Russian translation tasks confirmed that the proposed method significantly improved the translation quality evaluated by the BLEU scores of the Transformer model by 1.34 points in the Japanese-to-English task, 2.19 points in the English-to-German task, 0.15 points in the Czech-to-English task, and 1.48 points in the English-to-Russian task.

We propose a novel video understanding task by fusing knowledge-based and video question answering. First, we introduce KnowIT VQA, a video dataset with 24,282 human-generated question-answer pairs about a popular sitcom. The dataset combines visual, textual and temporal coherence reasoning together with knowledge-based questions, which need of the experience obtained from the viewing of the series to be answered. Second, we propose a video understanding model by combining the visual and textual video content with specific knowledge about the show. Our main findings are: (i) the incorporation of knowledge produces outstanding improvements for VQA in video, and (ii) the performance on KnowIT VQA still lags well behind human accuracy, indicating its usefulness for studying current video modelling limitations.

Advanced pre-trained models for text representation have achieved state-of-the-art performance on various text classification tasks.However, the discrepancy between the semantic similarity of texts and labelling standards affects classifiers, i.e. leading to lower performance in cases where classifiers should assign different labels to semantically similar texts.To address this problem, we propose a simple multitask learning model that uses negative supervision.Specifically, our model encourages texts with different labels to have distinct representations.Comprehensive experiments show that our model outperforms the stateof-the-art pre-trained model on both singleand multi-label classifications, sentence and document classifications, and classifications in three different languages.

Existing studies on multimodal neural machine translation (MNMT) have mainly focused on the effect of combining visual and textual modalities to improve translations. However, it has been suggested that the visual modality is only marginally beneficial. Conventional visual attention mechanisms have been used to select the visual features from equally-sized grids generated by convolutional neural networks (CNNs), and may have had modest effects on aligning the visual concepts associated with textual objects, because the grid visual features do not capture semantic information. In contrast, we propose the application of semantic image regions for MNMT by integrating visual and textual features using two individual attention mechanisms (double attention). We conducted experiments on the Multi30k dataset and achieved an improvement of 0.5 and 0.9 BLEU points for English-German and English-French translation tasks, compared with the MNMT with grid visual features. We also demonstrated concrete improvements on translation performance benefited from semantic image regions.

Adverse drug reactions are a severe problem that significantly degrade quality of life, or even threaten the life of patients. Patient-generated texts available on the web have been gaining attention as a promising source of information in this regard. While previous studies annotated such patient-generated content, they only reported on limited information, such as whether a text described an adverse drug reaction or not. Further, they only annotated short texts of a few sentences crawled from online forums and social networking services. The dataset we present in this paper is unique for the richness of annotated information, including detailed descriptions of drug reactions with full context. We crawled patient’s weblog articles shared on an online patient-networking platform and annotated the effects of drugs therein reported. We identified spans describing drug reactions and assigned labels for related drug names, standard codes for the symptoms of the reactions, and types of effects. As a first dataset, we annotated 677 drug reactions with these detailed labels based on 169 weblog articles by Japanese lung cancer patients. Our annotation dataset is made publicly available at our web site (https://yukiar.github.io/adr-jp/) for further research on the detection of adverse drug reactions and more broadly, on patient-generated text processing.

In this paper, we propose a full pipeline of analysis of a large corpus about a century of public meeting in historical Australian news papers, from construction to visual exploration. The corpus construction method is based on image processing and OCR. We digitize and transcribe texts of the specific topic of public meeting. Experiments show that our proposed method achieves a F-score of 87.8% for corpus construction. As a result, we built a content search tool for temporal and semantic content analysis.

Visual question answering (VQA) aims at answering questions about the visual content of an image or a video. Currently, most work on VQA is focused on image-based question answering, and less attention has been paid into answering questions about videos. However, VQA in video presents some unique challenges that are worth studying: it not only requires to model a sequence of visual features over time, but often it also needs to reason about associated subtitles. In this work, we propose to use BERT, a sequential modelling technique based on Transformers, to encode the complex semantics from video clips. Our proposed model jointly captures the visual and language information of a video scene by encoding not only the subtitles but also a sequence of visual concepts with a pretrained language-based Transformer. In our experiments, we exhaustively study the performance of our model by taking different input arrangements, showing outstanding improvements when compared against previous work on two well-known video VQA datasets: TVQA and Pororo.

Raj Dabre, Atsushi Fujita, Chenhui Chu. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

A paraphrase is a restatement of the meaning of a text in other words. Paraphrases have been studied to enhance the performance of many natural language processing tasks. In this paper, we propose a novel task iParaphrasing to extract visually grounded paraphrases (VGPs), which are different phrasal expressions describing the same visual concept in an image. These extracted VGPs have the potential to improve language and image multimodal tasks such as visual question answering and image captioning. How to model the similarity between VGPs is the key of iParaphrasing. We apply various existing methods as well as propose a novel neural network-based method with image attention, and report the results of the first attempt toward iParaphrasing.

Neural machine translation (NMT) is a deep learning based approach for machine translation, which yields the state-of-the-art translation performance in scenarios where large-scale parallel corpora are available. Although the high-quality and domain-specific translation is crucial in the real world, domain-specific corpora are usually scarce or nonexistent, and thus vanilla NMT performs poorly in such scenarios. Domain adaptation that leverages both out-of-domain parallel corpora as well as monolingual corpora for in-domain translation, is very important for domain-specific translation. In this paper, we give a comprehensive survey of the state-of-the-art domain adaptation techniques for NMT.

In this paper, we propose a new annotation approach to Chinese word segmentation, part-of-speech (POS) tagging and dependency labelling that aims to overcome the two major issues in traditional morphology-based annotation: Inconsistency and data sparsity. We re-annotate the Penn Chinese Treebank 5.0 (CTB5) and demonstrate the advantages of this approach compared to the original CTB5 annotation through word segmentation, POS tagging and machine translation experiments.

Parallel corpora are crucial for statistical machine translation (SMT); however, they are quite scarce for most language pairs and domains. As comparable corpora are far more available, many studies have been conducted to extract either parallel sentences or fragments from them for SMT. In this article, we propose an integrated system to extract both parallel sentences and fragments from comparable corpora. We first apply parallel sentence extraction to identify parallel sentences from comparable sentences. We then extract parallel fragments from the comparable sentences. Parallel sentence extraction is based on a parallel sentence candidate filter and classifier for parallel sentence identification. We improve it by proposing a novel filtering strategy and three novel feature sets for classification. Previous studies have found it difficult to accurately extract parallel fragments from comparable sentences. We propose an accurate parallel fragment extraction method that uses an alignment model to locate the parallel fragment candidates and an accurate lexicon-based filter to identify the truly parallel fragments. A case study on the Chinese--Japanese Wikipedia indicates that our proposed methods outperform previously proposed methods, and the parallel data extracted by our system significantly improves SMT performance.

The Chinese and Japanese languages share Chinese characters. Since the Chinese characters in Japanese originated from ancient China, many common Chinese characters exist between these two languages. Since Chinese characters contain significant semantic information and common Chinese characters share the same meaning in the two languages, they can be quite useful in Chinese-Japanese machine translation (MT). We therefore propose a method for creating a Chinese character mapping table for Japanese, traditional Chinese, and simplified Chinese, with the aim of constructing a complete resource of common Chinese characters. Furthermore, we point out two main problems in Chinese word segmentation for Chinese-Japanese MT, namely, unknown words and word segmentation granularity, and propose an approach exploiting common Chinese characters to solve these problems. We also propose a statistical method for detecting other semantically equivalent Chinese characters other than the common ones and a method for exploiting shared Chinese characters in phrase alignment. Results of the experiments carried out on a state-of-the-art phrase-based statistical MT system and an example-based MT system show that our proposed approaches can improve MT performance significantly, thereby verifying the effectiveness of shared Chinese characters for Chinese-Japanese MT.