Large language models show improved downstream task performance when prompted to generate step-by-step reasoning to justify their final answers. These reasoning steps greatly improve model interpretability and verification, but objectively studying their correctness (independent of the final answer) is difficult without reliable methods for automatic evaluation. We simply do not know how often the stated reasoning steps actually support the final end task predictions. In this work, we present ROSCOE, a suite of interpretable, unsupervised automatic scores that improve and extend previous text generation evaluation metrics. To evaluate ROSCOE against baseline metrics, we design a typology of reasoning errors and collect synthetic and human evaluation scores on commonly used reasoning datasets. In contrast with existing metrics, ROSCOE can measure semantic consistency, logicality, informativeness, fluency, and factuality - among other traits - by leveraging properties of step-by-step rationales. We empirically verify the strength of our metrics on five human annotated and six programmatically perturbed diagnostics datasets - covering a diverse set of tasks that require reasoning skills and show that ROSCOE can consistently outperform baseline metrics.
The rapid progress of Natural Language Processing (NLP) technologies has led to the widespread availability and effectiveness of text generation tools such as ChatGPT and Claude. While highly useful, these technologies also pose significant risks to the credibility of various media forms if they are employed for paraphrased plagiarism -- one of the most subtle forms of content misuse in scientific literature and general text media. Although automated methods for paraphrase identification have been developed, detecting this type of plagiarism remains challenging due to the inconsistent nature of the datasets used to train these methods. In this article, we examine traditional and contemporary approaches to paraphrase identification, investigating how the under-representation of certain paraphrase types in popular datasets, including those used to train Large Language Models (LLMs), affects the ability to detect plagiarism. We introduce and validate a new refined typology for paraphrases (ReParaphrased, REfined PARAPHRASE typology definitions) to better understand the disparities in paraphrase type representation. Lastly, we propose new directions for future research and dataset development to enhance AI-based paraphrase detection.
Software engineers working with the same programming language (PL) may speak different natural languages (NLs) and vice versa, erecting huge barriers to communication and working efficiency. Recent studies have demonstrated the effectiveness of generative pre-training in computer programs, yet they are always English-centric. In this work, we step towards bridging the gap between multilingual NLs and multilingual PLs for large language models (LLMs). We release ERNIE-Code, a unified pre-trained language model for 116 NLs and 6 PLs. We employ two methods for universal cross-lingual pre-training: span-corruption language modeling that learns patterns from monolingual NL or PL; and pivot-based translation language modeling that relies on parallel data of many NLs and PLs. Extensive results show that ERNIE-Code outperforms previous multilingual LLMs for PL or NL across a wide range of end tasks of code intelligence, including multilingual code-to-text, text-to-code, code-to-code, and text-to-text generation. We further show its advantage of zero-shot prompting on multilingual code summarization and text-to-text translation. We release our code and pre-trained checkpoints.
The two popular datasets ScanRefer [16] and ReferIt3D [3] connect natural language to real-world 3D data. In this paper, we curate a large-scale and complementary dataset extending both the aforementioned ones by associating all objects mentioned in a referential sentence to their underlying instances inside a 3D scene. Specifically, our Scan Entities in 3D (ScanEnts3D) dataset provides explicit correspondences between 369k objects across 84k natural referential sentences, covering 705 real-world scenes. Crucially, we show that by incorporating intuitive losses that enable learning from this novel dataset, we can significantly improve the performance of several recently introduced neural listening architectures, including improving the SoTA in both the Nr3D and ScanRefer benchmarks by 4.3% and 5.0%, respectively. Moreover, we experiment with competitive baselines and recent methods for the task of language generation and show that, as with neural listeners, 3D neural speakers can also noticeably benefit by training with ScanEnts3D, including improving the SoTA by 13.2 CIDEr points on the Nr3D benchmark. Overall, our carefully conducted experimental studies strongly support the conclusion that, by learning on ScanEnts3D, commonly used visio-linguistic 3D architectures can become more efficient and interpretable in their generalization without needing to provide these newly collected annotations at test time. The project's webpage is https://scanents3d.github.io/ .
This research revisits the classic Turing test and compares recent large language models such as ChatGPT for their abilities to reproduce human-level comprehension and compelling text generation. Two task challenges -- summarization, and question answering -- prompt ChatGPT to produce original content (98-99%) from a single text entry and also sequential questions originally posed by Turing in 1950. We score the original and generated content against the OpenAI GPT-2 Output Detector from 2019, and establish multiple cases where the generated content proves original and undetectable (98%). The question of a machine fooling a human judge recedes in this work relative to the question of "how would one prove it?" The original contribution of the work presents a metric and simple grammatical set for understanding the writing mechanics of chatbots in evaluating their readability and statistical clarity, engagement, delivery, and overall quality. While Turing's original prose scores at least 14% below the machine-generated output, the question of whether an algorithm displays hints of Turing's truly original thoughts (the "Lovelace 2.0" test) remains unanswered and potentially unanswerable for now.
The increasing privacy concerns on personal private text data promote the development of federated learning (FL) in recent years. However, the existing studies on applying FL in NLP are not suitable to coordinate participants with heterogeneous or private learning objectives. In this study, we further broaden the application scope of FL in NLP by proposing an Assign-Then-Contrast (denoted as ATC) framework, which enables clients with heterogeneous NLP tasks to construct an FL course and learn useful knowledge from each other. Specifically, the clients are suggested to first perform local training with the unified tasks assigned by the server rather than using their own learning objectives, which is called the Assign training stage. After that, in the Contrast training stage, clients train with different local learning objectives and exchange knowledge with other clients who contribute consistent and useful model updates. We conduct extensive experiments on six widely-used datasets covering both Natural Language Understanding (NLU) and Natural Language Generation (NLG) tasks, and the proposed ATC framework achieves significant improvements compared with various baseline methods. The source code is available at \url{https://github.com/alibaba/FederatedScope/tree/master/federatedscope/nlp/hetero_tasks}.
Recent pre-trained language models have shown promising capabilities in generating fluent and realistic natural language text. However, generating multi-sentence text with global content planning has been a long-existing research question. Current approaches for controlled text generation can hardly address this issue, as they usually condition on single known control attributes. In this study, we propose a low-cost yet effective framework which explicitly models the global content plan of the generated text. Specifically, it optimizes the joint distribution of the natural language sequence and the global content plan in a plug-and-play manner. We conduct extensive experiments on the well-established Recipe1M+ benchmark. Both automatic and human evaluations verify that our model achieves the state-of-the-art performance on the task of recipe generation
The input and output of most text generation tasks can be transformed to two sequences of tokens and they can be modeled using sequence-to-sequence learning modeling tools such as Transformers. These models are usually trained by maximizing the likelihood the output text sequence and assumes the input sequence and all gold preceding tokens are given during training, while during inference the model suffers from the exposure bias problem (i.e., it only has access to its previously predicted tokens rather gold tokens during beam search). In this paper, we propose MoCa ({\bf Mo}mentum {\bf Ca}libration) for text generation. MoCa is an online method that dynamically generates slowly evolving (but consistent) samples using a momentum moving average generator with beam search and MoCa learns to align its model scores of these samples with their actual qualities. Experiments on four text generation datasets (i.e., CNN/DailyMail, XSum, SAMSum and Gigaword) show MoCa consistently improves strong pre-trained transformers using vanilla fine-tuning and we achieve the state-of-the-art results on CNN/DailyMail and SAMSum datasets.
Privacy preserving deep learning is an emerging field in machine learning that aims to mitigate the privacy risks in the use of deep neural networks. One such risk is training data extraction from language models that have been trained on datasets, which contain personal and privacy sensitive information. In our study, we investigate the extent of named entity memorization in fine-tuned BERT models. We use single-label text classification as representative downstream task and employ three different fine-tuning setups in our experiments, including one with Differential Privacy (DP). We create a large number of text samples from the fine-tuned BERT models utilizing a custom sequential sampling strategy with two prompting strategies. We search in these samples for named entities and check if they are also present in the fine-tuning datasets. We experiment with two benchmark datasets in the domains of emails and blogs. We show that the application of DP has a detrimental effect on the text generation capabilities of BERT. Furthermore, we show that a fine-tuned BERT does not generate more named entities specific to the fine-tuning dataset than a BERT model that is pre-trained only. This suggests that BERT is unlikely to emit personal or privacy sensitive named entities. Overall, our results are important to understand to what extent BERT-based services are prone to training data extraction attacks.
The rapid development and application of natural language generation (NLG) techniques has revolutionized the field of automatic text production. However, these techniques are still limited in their ability to produce human-like text that is truly reasonable and informative. In this paper, we explore the importance of NLG being guided by knowledge, in order to convey human-like reasoning through language generation. We propose ten goals for intelligent NLG systems to pursue, and briefly review the achievement of NLG techniques guided by knowledge and reasoning. We also conclude by envisioning future directions and challenges in the pursuit of these goals.
We present a novel neural model for modern poetry generation in French. The model consists of two pretrained neural models that are fine-tuned for the poem generation task. The encoder of the model is a RoBERTa based one while the decoder is based on GPT-2. This way the model can benefit from the superior natural language understanding performance of RoBERTa and the good natural language generation performance of GPT-2. Our evaluation shows that the model can create French poetry successfully. On a 5 point scale, the lowest score of 3.57 was given by human judges to typicality and emotionality of the output poetry while the best score of 3.79 was given to understandability.
Text generation has long been a popular research topic in NLP. However, the task of generating recruitment emails from recruiters to candidates in the job recommendation scenario has received little attention by the research community. This work aims at defining the topic of automatic email generation for job recommendation, identifying the challenges, and providing a baseline template-based solution for Danish jobs. Evaluation by human experts shows that our method is effective. We wrap up by discussing the future research directions for better solving this task.
Hebrew is a Morphological rich language, making its modeling harder than simpler language. Recent developments such as Transformers in general and Bert in particular opened a path for Hebrew models that reach SOTA results, not falling short from other non-MRL languages. We explore the cutting edge in this field performing style transfer, text generation and classification over news articles collected from online archives. Furthermore, the news portals that feed our collective consciousness are an interesting corpus to study, as their analysis and tracing might reveal insights about our society and discourse.
Controlled text generation is a very important task in the arena of natural language processing due to its promising applications. In order to achieve this task we mainly introduce the novel soft prompt tuning method of using soft prompts at both encoder and decoder levels together in a T5 model and investigate the performance as the behaviour of an additional soft prompt related to the decoder of a T5 model in controlled text generation remained unexplored. Then we also investigate the feasibility of steering the output of this extended soft prompted T5 model at decoder level and finally analyse the utility of generated text to be used in AI related tasks such as training AI models with an interpretability analysis of the classifier trained with synthetic text, as there is a lack of proper analysis of methodologies in generating properly labelled data to be utilized in AI tasks. Through the performed in-depth intrinsic and extrinsic evaluations of this generation model along with the artificially generated data, we found that this model produced better results compared to the T5 model with a single soft prompt at encoder level and the sentiment classifier trained using this artificially generated data can produce comparable classification results to the results of a classifier trained with real labelled data and also the classifier decision is interpretable with respect to the input text content.
Advancement in large pretrained language models has significantly improved their performance for conditional language generation tasks including summarization albeit with hallucinations. To reduce hallucinations, conventional methods proposed improving beam search or using a fact checker as a postprocessing step. In this paper, we investigate the use of the Natural Language Inference (NLI) entailment metric to detect and prevent hallucinations in summary generation. We propose an NLI-assisted beam re-ranking mechanism by computing entailment probability scores between the input context and summarization model-generated beams during saliency-enhanced greedy decoding. Moreover, a diversity metric is introduced to compare its effectiveness against vanilla beam search. Our proposed algorithm significantly outperforms vanilla beam decoding on XSum and CNN/DM datasets.
Open-ended text generation with autoregressive language models (LMs) is one of the core tasks in natural language processing. However, maximization-based decoding methods (e.g., greedy/beam search) often lead to the degeneration problem, i.e., the generated text is unnatural and contains undesirable repetitions. Existing solutions to this problem either introduce randomness prone to incoherence or require a look-ahead mechanism that demands extra computational overhead. In this study, we formulate open-ended text generation from a new perspective, i.e., we view it as an exploration process within a directed graph. Thereby, we understand the phenomenon of degeneration as circular loops within the directed graph. Based on our formulation, we propose a novel decoding method -- \textit{momentum decoding} -- which encourages the LM to \textit{greedily} explore new nodes outside the current graph. Meanwhile, it also allows the LM to return to the existing nodes with a momentum downgraded by a pre-defined resistance function. We extensively test our approach on three benchmarks from different domains through automatic and human evaluations. The results show that momentum decoding performs comparably with the current state of the art while enjoying notably improved inference speed and computation FLOPs. Furthermore, we conduct a detailed analysis to reveal the merits and inner workings of our approach. Our codes and other related resources are publicly available at https://github.com/gmftbyGMFTBY/MomentumDecoding.
Large pre-trained language models have recently enabled open-ended generation frameworks (e.g., prompt-to-text NLG) to tackle a variety of tasks going beyond the traditional data-to-text generation. While this framework is more general, it is under-specified and often leads to a lack of controllability restricting their real-world usage. We propose a new grounded keys-to-text generation task: the task is to generate a factual description about an entity given a set of guiding keys, and grounding passages. To address this task, we introduce a new dataset, called EntDeGen. Inspired by recent QA-based evaluation measures, we propose an automatic metric, MAFE, for factual correctness of generated descriptions. Our EntDescriptor model is equipped with strong rankers to fetch helpful passages and generate entity descriptions. Experimental result shows a good correlation (60.14) between our proposed metric and human judgments of factuality. Our rankers significantly improved the factual correctness of generated descriptions (15.95% and 34.51% relative gains in recall and precision). Finally, our ablation study highlights the benefit of combining keys and groundings.
We present a robust methodology for evaluating biases in natural language generation(NLG) systems. Previous works use fixed hand-crafted prefix templates with mentions of various demographic groups to prompt models to generate continuations for bias analysis. These fixed prefix templates could themselves be specific in terms of styles or linguistic structures, which may lead to unreliable fairness conclusions that are not representative of the general trends from tone varying prompts. To study this problem, we paraphrase the prompts with different syntactic structures and use these to evaluate demographic bias in NLG systems. Our results suggest similar overall bias trends but some syntactic structures lead to contradictory conclusions compared to past works. We show that our methodology is more robust and that some syntactic structures prompt more toxic content while others could prompt less biased generation. This suggests the importance of not relying on a fixed syntactic structure and using tone-invariant prompts. Introducing syntactically-diverse prompts can achieve more robust NLG (bias) evaluation.
Memes are powerful means for effective communication on social media. Their effortless amalgamation of viral visuals and compelling messages can have far-reaching implications with proper marketing. Previous research on memes has primarily focused on characterizing their affective spectrum and detecting whether the meme's message insinuates any intended harm, such as hate, offense, racism, etc. However, memes often use abstraction, which can be elusive. Here, we introduce a novel task - EXCLAIM, generating explanations for visual semantic role labeling in memes. To this end, we curate ExHVV, a novel dataset that offers natural language explanations of connotative roles for three types of entities - heroes, villains, and victims, encompassing 4,680 entities present in 3K memes. We also benchmark ExHVV with several strong unimodal and multimodal baselines. Moreover, we posit LUMEN, a novel multimodal, multi-task learning framework that endeavors to address EXCLAIM optimally by jointly learning to predict the correct semantic roles and correspondingly to generate suitable natural language explanations. LUMEN distinctly outperforms the best baseline across 18 standard natural language generation evaluation metrics. Our systematic evaluation and analyses demonstrate that characteristic multimodal cues required for adjudicating semantic roles are also helpful for generating suitable explanations.
Climate change is threatening human health in unprecedented orders and many ways. These threats are expected to grow unless effective and evidence-based policies are developed and acted upon to minimize or eliminate them. Attaining such a task requires the highest degree of the flow of knowledge from science into policy. The multidisciplinary, location-specific, and vastness of published science makes it challenging to keep track of novel work in this area, as well as making the traditional knowledge synthesis methods inefficient in infusing science into policy. To this end, we consider developing multiple domain-specific language models (LMs) with different variations from Climate- and Health-related information, which can serve as a foundational step toward capturing available knowledge to enable solving different tasks, such as detecting similarities between climate- and health-related concepts, fact-checking, relation extraction, evidence of health effects to policy text generation, and more. To our knowledge, this is the first work that proposes developing multiple domain-specific language models for the considered domains. We will make the developed models, resources, and codebase available for the researchers.