This report provides a preliminary evaluation of ChatGPT for machine translation, including translation prompt, multilingual and robustness. We adopt the prompts advised by to trigger its ability find that candidate generally work well with minor performance differences. By evaluating on number benchmark test sets, we performs competitively commercial products (e.g., Google Translate) high-resource European languages but lags behind significantly low-resource or distant languages. As...

Abstract Large language models (LLMs) have demonstrated impressive capabilities in general scenarios, exhibiting a level of aptitude that approaches, some aspects even surpasses, human-level intelligence. Among their numerous skills, the translation abilities LLMs received considerable attention. Compared to typical machine focuses solely on source-to-target mapping, LLM-based can potentially mimic human process, which might take preparatory steps ensure high-quality translation. This work...

Real-time emotion recognition (RTER) in conversations is significant for developing emotionally intelligent chatting machines. Without the future context RTER, it becomes critical to build memory bank carefully capturing historical and summarize memories appropriately retrieve relevant information. We propose an Attention Gated Hierarchical Memory Network (AGHMN) address problems of prior work: (1) Commonly used convolutional neural networks (CNNs) utterance feature extraction are less...

ChatGPT is a cutting-edge artificial intelligence language model developed by OpenAI, which has attracted lot of attention due to its surprisingly strong ability in answering follow-up questions. In this report, we aim evaluate on the Grammatical Error Correction(GEC) task, and compare it with commercial GEC product (e.g., Grammarly) state-of-the-art models GECToR). By testing CoNLL2014 benchmark dataset, find that performs not as well those baselines terms automatic evaluation metrics...

In this paper, we address three challenges in utterance-level emotion recognition dialogue systems: (1) the same word can deliver different emotions contexts; (2) some are rarely seen general dialogues; (3) long-range contextual information is hard to be effectively captured. We therefore propose a hierarchical Gated Recurrent Unit (HiGRU) framework with lower-level GRU model word-level inputs and an upper-level capture contexts of embeddings. Moreover, promote two variants, HiGRU individual...

Wenxiang Jiao, Haiqin Yang, Irwin King, Michael R. Lyu. Proceedings of the 2019 Conference North American Chapter Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers). 2019.

Wenxuan Wang, Wenxiang Jiao, Yongchang Hao, Xing Shuming Shi, Zhaopeng Tu, Michael Lyu. Proceedings of the 60th Annual Meeting Association for Computational Linguistics (Volume 1: Long Papers). 2022.

Large language models (LLMs) have demonstrated impressive capabilities in general scenarios, exhibiting a level of aptitude that approaches, some aspects even surpasses, human-level intelligence. Among their numerous skills, the translation abilities LLMs received considerable attention. Compared to typical machine focuses solely on source-to-target mapping, LLM-based can potentially mimic human process which might take preparatory steps ensure high-quality translation. This work explores...

Modern large language models (LLMs) like ChatGPT have shown remarkable performance on general tasks but still struggle complex reasoning tasks, which drives the research cognitive behaviors of LLMs to explore human-like problem-solving strategies. Along this direction, one representative strategy is self-reflection, asks an LLM refine solution with feedback generated by itself iteratively. However, our study shows that such reflection-style methods suffer from Degeneration-of-Thought (DoT)...

Wenxiang Jiao, Xing Wang, Zhaopeng Tu, Shuming Shi, Michael Lyu, Irwin King. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

End-to-end sign language translation (SLT) aims to directly convert videos into spoken texts without intermediate representations. It has been challenging due the data scarcity of labeled and modality gap between texts. To tackle these challenges, we propose a novel Cross-modality Data Augmentation (XmDA) framework transfer powerful gloss-to-text capabilities end-to-end (i.e., video-to-text). Specifically, XmDA consists two key components: cross-modality mix-up knowledge distillation. The...

Large language models (LLMs) like ChatGPT have exhibited remarkable abilities on a wide range of natural processing (NLP) tasks, including various machine translation accomplished during chat. However, these are only accessible through restricted APIs, which creates barriers to new research and advancements in the field. Therefore, we propose ParroT, framework enhance regulate chat based open-source LLMs (e.g., LLaMA), human-written feedback data. Specifically, ParroT reformulates data into...

Yongchang Hao, Shilin He, Wenxiang Jiao, Zhaopeng Tu, Michael Lyu, Xing Wang. Proceedings of the 2021 Conference North American Chapter Association for Computational Linguistics: Human Language Technologies. 2021.

Decision-making, a complicated task requiring various types of abilities, presents an excellent framework for assessing Large Language Models (LLMs). Our research investigates LLMs' decision-making capabilities through the lens well-established field, Game Theory. We focus specifically on games that support participation more than two agents simultaneously. Subsequently, we introduce our framework, GAMA-Bench, including eight classical multi-agent games. design scoring scheme to assess...

In a ground-to-satellite communication system with preset EDFA, the EDFA's performance will be affected by space environment. With 250 Gy radiation, gain decreases 2 dB from 19.97 at 20 °C. The BER increases 2.5 orders of magnitude 10(-10), and more radiation. situation aggravates when temperature rises 73 laser's divergence-angle transmitter radius have optimal values to make lowest increasing receiver diameter makes lower BERs, so setting these parameters appropriate compensate degradation...

Large-scale training datasets lie at the core of recent success neural machine translation (NMT) models. However, complex patterns and potential noises in large-scale data make NMT models difficult. In this work, we explore to identify inactive examples which contribute less model performance, show that existence depends on distribution. We further introduce rejuvenation improve by exploiting examples. The proposed framework consists three phases. First, train an identification original...

In a space optical communication system with an amplification sub-system, the performance of erbium-doped fiber amplifier (EDFA) will worsen due to effect radiation. Consequently, EDFA not work under its optimal state which has been already designed on ground. To fix this problem, study basic characteristics radiation is conducted. simulation tests, gain and length both decrease dose dynamically adapt such effects, new self-adaptive established makes improvement 7 dB in when reaches 5000 Gy....

Emotion Recognition in Conversations (ERC) aims to predict the emotional state of speakers conversations, which is essentially a text classification task. Unlike sentence-level problem, available supervised data for ERC task limited, potentially prevents models from playing their maximum effect. In this paper, we propose novel approach leverage unsupervised conversation data, more accessible. Specifically, Conversation Completion (ConvCom) task, attempts select correct answer candidate...

We report a nano-optical conveyor belt containing an array of gold plasmonic non-concentric nanorings (PNNRs) for the realization trapping and unidirectional transportation nanoparticles through rotating polarization excitation beam. The location hot spots within asymmetric nanostructure is dependent, thus making it possible to manipulate trapped target by incident state. In case PNNR, two poles have highly unbalanced trap potential. This greatly enhances chance transferring particles...

Shudong Liu, Xuebo Derek F. Wong, Zhaocong Li, Wenxiang Jiao, Lidia S. Chao, Min Zhang. Proceedings of the 61st Annual Meeting Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Evaluating Large Language Models' (LLMs) anthropomorphic capabilities has become increasingly important in contemporary discourse. Utilizing the emotion appraisal theory from psychology, we propose to evaluate empathy ability of LLMs, i.e., how their feelings change when presented with specific situations. After a careful and comprehensive survey, collect dataset containing over 400 situations that have proven effective eliciting eight emotions central our study. Categorizing into 36...

Safety lies at the core of development Large Language Models (LLMs). There is ample work on aligning LLMs with human ethics and preferences, including data filtering in pretraining, supervised fine-tuning, reinforcement learning from feedback, red teaming, etc. In this study, we discover that chat cipher can bypass safety alignment techniques LLMs, which are mainly conducted natural languages. We propose a novel framework CipherChat to systematically examine generalizability non-natural...