Text Detoxification as Style Transfer in English and Hindi

This paper focuses on text detoxification, i.e., automatically converting toxic text into non-toxic text. This task contributes to safer and more respectful online communication and can be considered a Text Style Transfer (TST) task, where the text style changes while its content is preserved. We present three approaches: knowledge transfer from a similar task, multi-task learning approach, combining sequence-to-sequence modeling with various toxicity classification tasks, and, delete and reconstruct approach. To support our research, we utilize a dataset provided by Dementieva et al.(2021), which contains multiple versions of detoxified texts corresponding to toxic texts. In our experiments, we selected the best variants through expert human annotators, creating a dataset where each toxic sentence is paired with a single, appropriate detoxified version. Additionally, we introduced a small Hindi parallel dataset, aligning with a part of the English dataset, suitable for evaluation purposes. Our results demonstrate that our approach effectively balances text detoxication while preserving the actual content and maintaining fluency.

Leak, Cheat, Repeat: Data Contamination and Evaluation Malpractices in Closed-Source LLMs

Natural Language Processing (NLP) research is increasingly focusing on the use of Large Language Models (LLMs), with some of the most popular ones being either fully or partially closed-source. The lack of access to model details, especially regarding training data, has repeatedly raised concerns about data contamination among researchers. Several attempts have been made to address this issue, but they are limited to anecdotal evidence and trial and error. Additionally, they overlook the problem of \emph{indirect} data leaking, where models are iteratively improved by using data coming from users. In this work, we conduct the first systematic analysis of work using OpenAI's GPT-3.5 and GPT-4, the most prominently used LLMs today, in the context of data contamination. By analysing 255 papers and considering OpenAI's data usage policy, we extensively document the amount of data leaked to these models during the first year after the model's release. We report that these models have been globally exposed to $\sim$4.7M samples from 263 benchmarks. At the same time, we document a number of evaluation malpractices emerging in the reviewed papers, such as unfair or missing baseline comparisons and reproducibility issues. We release our results as a collaborative project on https://leak-llm.github.io/, where other researchers can contribute to our efforts.

Beyond Reference-Based Metrics: Analyzing Behaviors of Open LLMs on Data-to-Text Generation

We investigate to which extent open large language models (LLMs) can generate coherent and relevant text from structured data. To prevent bias from benchmarks leaked into LLM training data, we collect Quintd-1: an ad-hoc benchmark for five data-to-text (D2T) generation tasks, consisting of structured data records in standard formats gathered from public APIs. We leverage reference-free evaluation metrics and LLMs' in-context learning capabilities, allowing us to test the models with no human-written references. Our evaluation focuses on annotating semantic accuracy errors on token-level, combining human annotators and a metric based on GPT-4. Our systematic examination of the models' behavior across domains and tasks suggests that state-of-the-art open LLMs with 7B parameters can generate fluent and coherent text from various standard data formats in zero-shot settings. However, we also show that semantic accuracy of the outputs remains a major issue: on our benchmark, 80% of outputs of open LLMs contain a semantic error according to human annotators (91% according to GPT-4). Our code, data, and model outputs are available at https://d2t-llm.github.io.

Balancing the Style-Content Trade-Off in Sentiment Transfer Using Polarity-Aware Denoising

Text sentiment transfer aims to flip the sentiment polarity of a sentence (positive to negative or vice versa) while preserving its sentiment-independent content. Although current models show good results at changing the sentiment, content preservation in transferred sentences is insufficient. In this paper, we present a sentiment transfer model based on polarity-aware denoising, which accurately controls the sentiment attributes in generated text, preserving the content to a great extent and helping to balance the style-content trade-off. Our proposed model is structured around two key stages in the sentiment transfer process: better representation learning using a shared encoder and sentiment-controlled generation using separate sentiment-specific decoders. Empirical results show that our methods outperforms state-of-the-art baselines in terms of content preservation while staying competitive in terms of style transfer accuracy and fluency.

LEEETs-Dial: Linguistic Entrainment in End-to-End Task-oriented Dialogue systems

Linguistic entrainment, or alignment, represents a phenomenon where linguistic patterns employed by conversational participants converge to one another. While alignment has been shown to produce a more natural user experience, most dialogue systems do not have any provisions for it. In this work, we introduce methods for achieving dialogue alignment in a GPT-2-based end-to-end dialogue system through the utilization of shared vocabulary. We experiment with training instance weighting, alignment-specific loss, and additional conditioning to generate responses that align with the user. By comparing different entrainment techniques on the MultiWOZ dataset, we demonstrate that all three approaches produce significantly better-aligned results than the baseline, as confirmed by both automated and manual evaluation metrics.

Critic-Driven Decoding for Mitigating Hallucinations in Data-to-text Generation

Hallucination of text ungrounded in the input is a well-known problem in neural data-to-text generation. Many methods have been proposed to mitigate it, but they typically require altering model architecture or collecting additional data, and thus cannot be easily applied to an existing model. In this paper, we explore a new way to mitigate hallucinations by combining the probabilistic output of a generator language model (LM) with the output of a special "text critic" classifier, which guides the generation by assessing the match between the input data and the text generated so far. Our method does not need any changes to the underlying LM's architecture or training procedure and can thus be combined with any model and decoding operating on word probabilities. The critic does not need any additional training data, using the base LM's training data and synthetic negative examples. Our experimental results show that our method improves over the baseline on the WebNLG and OpenDialKG benchmarks.

With a Little Help from the Authors: Reproducing Human Evaluation of an MT Error Detector

This work presents our efforts to reproduce the results of the human evaluation experiment presented in the paper of Vamvas and Sennrich (2022), which evaluated an automatic system detecting over- and undertranslations (translations containing more or less information than the original) in machine translation (MT) outputs. Despite the high quality of the documentation and code provided by the authors, we discuss some problems we found in reproducing the exact experimental setup and offer recommendations for improving reproducibility. Our replicated results generally confirm the conclusions of the original study, but in some cases, statistically significant differences were observed, suggesting a high variability of human annotation.

Three Ways of Using Large Language Models to Evaluate Chat

This paper describes the systems submitted by team6 for ChatEval, the DSTC 11 Track 4 competition. We present three different approaches to predicting turn-level qualities of chatbot responses based on large language models (LLMs). We report improvement over the baseline using dynamic few-shot examples from a vector store for the prompts for ChatGPT. We also analyze the performance of the other two approaches and report needed improvements for future work. We developed the three systems over just two weeks, showing the potential of LLMs for this task. An ablation study conducted after the challenge deadline shows that the new Llama 2 models are closing the performance gap between ChatGPT and open-source LLMs. However, we find that the Llama 2 models do not benefit from few-shot examples in the same way as ChatGPT.

Tackling Hallucinations in Neural Chart Summarization

Hallucinations in text generation occur when the system produces text that is not grounded in the input. In this work, we tackle the problem of hallucinations in neural chart summarization. Our analysis shows that the target side of chart summarization training datasets often contains additional information, leading to hallucinations. We propose a natural language inference (NLI) based method to preprocess the training data and show through human evaluation that our method significantly reduces hallucinations. We also found that shortening long-distance dependencies in the input sequence and adding chart-related information like title and legends improves the overall performance.

Missing Information, Unresponsive Authors, Experimental Flaws: The Impossibility of Assessing the Reproducibility of Previous Human Evaluations in NLP

We report our efforts in identifying a set of previous human evaluations in NLP that would be suitable for a coordinated study examining what makes human evaluations in NLP more/less reproducible. We present our results and findings, which include that just 13\% of papers had (i) sufficiently low barriers to reproduction, and (ii) enough obtainable information, to be considered for reproduction, and that all but one of the experiments we selected for reproduction was discovered to have flaws that made the meaningfulness of conducting a reproduction questionable. As a result, we had to change our coordinated study design from a reproduce approach to a standardise-then-reproduce-twice approach. Our overall (negative) finding that the great majority of human evaluations in NLP is not repeatable and/or not reproducible and/or too flawed to justify reproduction, paints a dire picture, but presents an opportunity for a rethink about how to design and report human evaluations in NLP.