This technical report explores the ability of ChatGPT in recognizing emotions from text, which can be the basis of various applications like interactive chatbots, data annotation, and mental health analysis. While prior research has shown ChatGPT's basic ability in sentiment analysis, its performance in more nuanced emotion recognition is not yet explored. Here, we conducted experiments to evaluate its performance of emotion recognition across different datasets and emotion labels. Our findings indicate a reasonable level of reproducibility in its performance, with noticeable improvement through fine-tuning. However, the performance varies with different emotion labels and datasets, highlighting an inherent instability and possible bias. The choice of dataset and emotion labels significantly impacts ChatGPT's emotion recognition performance. This paper sheds light on the importance of dataset and label selection, and the potential of fine-tuning in enhancing ChatGPT's emotion recognition capabilities, providing a groundwork for better integration of emotion analysis in applications using ChatGPT.
Query-specific article generation is the task of, given a search query, generate a single article that gives an overview of the topic. We envision such articles as an alternative to presenting a ranking of search results. While generative Large Language Models (LLMs) like chatGPT also address this task, they are known to hallucinate new information, their models are secret, hard to analyze and control. Some generative LLMs provide supporting references, yet these are often unrelated to the generated content. As an alternative, we propose to study article generation systems that integrate document retrieval, query-specific clustering, and summarization. By design, such models can provide actual citations as provenance for their generated text. In particular, we contribute an evaluation framework that allows to separately trains and evaluate each of these three components before combining them into one system. We experimentally demonstrate that a system comprised of the best-performing individual components also obtains the best F-1 overall system quality.
We call into question the recently popularized method of direct model editing as a means of correcting factual errors in LLM generations. We contrast model editing with three similar but distinct approaches that pursue better defined objectives: (1) retrieval-based architectures, which decouple factual memory from inference and linguistic capabilities embodied in LLMs; (2) concept erasure methods, which aim at preventing systemic bias in generated text; and (3) attribution methods, which aim at grounding generations into identified textual sources. We argue that direct model editing cannot be trusted as a systematic remedy for the disadvantages inherent to LLMs, and while it has proven potential in improving model explainability, it opens risks by reinforcing the notion that models can be trusted for factuality. We call for cautious promotion and application of model editing as part of the LLM deployment process, and for responsibly limiting the use cases of LLMs to those not relying on editing as a critical component.
In this work, we introduce a framework for cross-lingual speech synthesis, which involves an upstream Voice Conversion (VC) model and a downstream Text-To-Speech (TTS) model. The proposed framework consists of 4 stages. In the first two stages, we use a VC model to convert utterances in the target locale to the voice of the target speaker. In the third stage, the converted data is combined with the linguistic features and durations from recordings in the target language, which are then used to train a single-speaker acoustic model. Finally, the last stage entails the training of a locale-independent vocoder. Our evaluations show that the proposed paradigm outperforms state-of-the-art approaches which are based on training a large multilingual TTS model. In addition, our experiments demonstrate the robustness of our approach with different model architectures, languages, speakers and amounts of data. Moreover, our solution is especially beneficial in low-resource settings.
Recognizing text lines from images is a challenging problem, especially for handwritten documents due to large variations in writing styles. While text line recognition models are generally trained on large corpora of real and synthetic data, such models can still make frequent mistakes if the handwriting is inscrutable or the image acquisition process adds corruptions, such as noise, blur, compression, etc. Writing style is generally quite consistent for an individual, which can be leveraged to correct mistakes made by such models. Motivated by this, we introduce the problem of adapting text line recognition models during test time. We focus on a challenging and realistic setting where, given only a single test image consisting of multiple text lines, the task is to adapt the model such that it performs better on the image, without any labels. We propose an iterative self-training approach that uses feedback from the language model to update the optical model, with confident self-labels in each iteration. The confidence measure is based on an augmentation mechanism that evaluates the divergence of the prediction of the model in a local region. We perform rigorous evaluation of our method on several benchmark datasets as well as their corrupted versions. Experimental results on multiple datasets spanning multiple scripts show that the proposed adaptation method offers an absolute improvement of up to 8% in character error rate with just a few iterations of self-training at test time.
This paper present our work in the DSAA 2023 Challenge about Link Prediction for Wikipedia Articles. We use traditional machine learning models with POS tags (part-of-speech tags) features extracted from text to train the classification model for predicting whether two nodes has the link. Then, we use these tags to test on various machine learning models. We obtained the results by F1 score at 0.99999 and got 7th place in the competition. Our source code is publicly available at this link: https://github.com/Tam1032/DSAA2023-Challenge-Link-prediction-DS-UIT_SAT
Conditional natural language generation methods often require either expensive fine-tuning or training a large language model from scratch. Both are unlikely to lead to good results without a substantial amount of data and computational resources. Prompt learning without changing the parameters of a large language model presents a promising alternative. It is a cost-effective approach, while still achieving competitive results. While this procedure is now established for zero- and few-shot text classification and structured prediction, it has received limited attention in conditional text generation. We present the first automatic prompt optimization approach for emotion-conditioned text generation with instruction-fine-tuned models. Our method uses an iterative optimization procedure that changes the prompt by adding, removing, or replacing tokens. As objective function, we only require a text classifier that measures the realization of the conditional variable in the generated text. We evaluate the method on emotion-conditioned text generation with a focus on event reports and compare it to manually designed prompts that also act as the seed for the optimization procedure. The optimized prompts achieve 0.75 macro-average F1 to fulfill the emotion condition in contrast to manually designed seed prompts with only 0.22 macro-average F1.
Despite significant results achieved by Contrastive Language-Image Pretraining (CLIP) in zero-shot image recognition, limited effort has been made exploring its potential for zero-shot video recognition. This paper presents Open-VCLIP++, a simple yet effective framework that adapts CLIP to a strong zero-shot video classifier, capable of identifying novel actions and events during testing. Open-VCLIP++ minimally modifies CLIP to capture spatial-temporal relationships in videos, thereby creating a specialized video classifier while striving for generalization. We formally demonstrate that training Open-VCLIP++ is tantamount to continual learning with zero historical data. To address this problem, we introduce Interpolated Weight Optimization, a technique that leverages the advantages of weight interpolation during both training and testing. Furthermore, we build upon large language models to produce fine-grained video descriptions. These detailed descriptions are further aligned with video features, facilitating a better transfer of CLIP to the video domain. Our approach is evaluated on three widely used action recognition datasets, following a variety of zero-shot evaluation protocols. The results demonstrate that our method surpasses existing state-of-the-art techniques by significant margins. Specifically, we achieve zero-shot accuracy scores of 88.1%, 58.7%, and 81.2% on UCF, HMDB, and Kinetics-600 datasets respectively, outpacing the best-performing alternative methods by 8.5%, 8.2%, and 12.3%. We also evaluate our approach on the MSR-VTT video-text retrieval dataset, where it delivers competitive video-to-text and text-to-video retrieval performance, while utilizing substantially less fine-tuning data compared to other methods. Code is released at https://github.com/wengzejia1/Open-VCLIP.
Multi-modal learning has become increasingly popular due to its ability to leverage information from different data sources (e.g., text and images) to improve the model performance. Recently, CLIP has emerged as an effective approach that employs vision-language contrastive pretraining to learn joint image and text representations and exhibits remarkable performance in zero-shot learning and text-guided natural image generation. Despite the huge practical success of CLIP, its theoretical understanding remains elusive. In this paper, we formally study transferrable representation learning underlying CLIP and demonstrate how features from different modalities get aligned. We also analyze its zero-shot transfer performance on the downstream tasks. Inspired by our analysis, we propose a new CLIP-type approach, which achieves better performance than CLIP and other state-of-the-art methods on benchmark datasets.
Red-teaming has been a widely adopted way to evaluate the harmfulness of Large Language Models (LLMs). It aims to jailbreak a model's safety behavior to make it act as a helpful agent disregarding the harmfulness of the query. Existing methods are primarily based on input text-based red-teaming such as adversarial prompts, low-resource prompts, or contextualized prompts to condition the model in a way to bypass its safe behavior. Bypassing the guardrails uncovers hidden harmful information and biases in the model that are left untreated or newly introduced by its safety training. However, prompt-based attacks fail to provide such a diagnosis owing to their low attack success rate, and applicability to specific models. In this paper, we present a new perspective on LLM safety research i.e., parametric red-teaming through Unalignment. It simply (instruction) tunes the model parameters to break model guardrails that are not deeply rooted in the model's behavior. Unalignment using as few as 100 examples can significantly bypass commonly referred to as CHATGPT, to the point where it responds with an 88% success rate to harmful queries on two safety benchmark datasets. On open-source models such as VICUNA-7B and LLAMA-2-CHAT 7B AND 13B, it shows an attack success rate of more than 91%. On bias evaluations, Unalignment exposes inherent biases in safety-aligned models such as CHATGPT and LLAMA- 2-CHAT where the model's responses are strongly biased and opinionated 64% of the time.