With recent advancements in the area of Natural Language Processing, the focus is slowly shifting from a purely English-centric view towards more language-specific solutions, including German. Especially practical for businesses to analyze their growing amount of textual data are text summarization systems, which transform long input documents into compressed and more digestible summary texts. In this work, we assess the particular landscape of German abstractive text summarization and investigate the reasons why practically useful solutions for abstractive text summarization are still absent in industry. Our focus is two-fold, analyzing a) training resources, and b) publicly available summarization systems. We are able to show that popular existing datasets exhibit crucial flaws in their assumptions about the original sources, which frequently leads to detrimental effects on system generalization and evaluation biases. We confirm that for the most popular training dataset, MLSUM, over 50% of the training set is unsuitable for abstractive summarization purposes. Furthermore, available systems frequently fail to compare to simple baselines, and ignore more effective and efficient extractive summarization approaches. We attribute poor evaluation quality to a variety of different factors, which are investigated in more detail in this work: A lack of qualitative (and diverse) gold data considered for training, understudied (and untreated) positional biases in some of the existing datasets, and the lack of easily accessible and streamlined pre-processing strategies or analysis tools. We provide a comprehensive assessment of available models on the cleaned datasets, and find that this can lead to a reduction of more than 20 ROUGE-1 points during evaluation. The code for dataset filtering and reproducing results can be found online at https://github.com/dennlinger/summaries
We introduce SPEAR-TTS, a multi-speaker text-to-speech (TTS) system that can be trained with minimal supervision. By combining two types of discrete speech representations, we cast TTS as a composition of two sequence-to-sequence tasks: from text to high-level semantic tokens (akin to "reading") and from semantic tokens to low-level acoustic tokens ("speaking"). Decoupling these two tasks enables training of the "speaking" module using abundant audio-only data, and unlocks the highly efficient combination of pretraining and backtranslation to reduce the need for parallel data when training the "reading" component. To control the speaker identity, we adopt example prompting, which allows SPEAR-TTS to generalize to unseen speakers using only a short sample of 3 seconds, without any explicit speaker representation or speaker-id labels. Our experiments demonstrate that SPEAR-TTS achieves a character error rate that is competitive with state-of-the-art methods using only 15 minutes of parallel data, while matching ground-truth speech in terms of naturalness and acoustic quality, as measured in subjective tests.
A prominent weakness of modern language models (LMs) is their tendency to generate factually incorrect text, which hinders their usability. A natural question is whether such factual errors can be detected automatically. Inspired by truth-seeking mechanisms in law, we propose a factuality evaluation framework for LMs that is based on cross-examination. Our key idea is that an incorrect claim is likely to result in inconsistency with other claims that the model generates. To discover such inconsistencies, we facilitate a multi-turn interaction between the LM that generated the claim and another LM (acting as an examiner) which introduces questions to discover inconsistencies. We empirically evaluate our method on factual claims made by multiple recent LMs on four benchmarks, finding that it outperforms existing methods and baselines, often by a large gap. Our results demonstrate the potential of using interacting LMs for capturing factual errors.
The increasing adoption of text-to-speech technologies has led to a growing demand for natural and emotive voices that adapt to a conversation's context and emotional tone. This need is particularly relevant for interactive narrative-driven systems such as video games, TV shows, and graphic novels. To address this need, we present the Emotive Narrative Storytelling (EMNS) corpus, a dataset of high-quality British English speech with labelled utterances designed to enhance interactive experiences with dynamic and expressive language. We provide high-quality clean audio recordings and natural language description pairs with transcripts and user-reviewed and self-reported labels for features such as word emphasis, expressiveness, and emotion labels. EMNS improves on existing datasets by providing higher quality and clean recording to aid more natural and expressive speech synthesis techniques for interactive narrative-driven experiences. Additionally, we release our remote and scalable data collection system to the research community.
Non-autoregressive Transformers (NATs) reduce the inference latency of Autoregressive Transformers (ATs) by predicting words all at once rather than in sequential order. They have achieved remarkable progress in machine translation as well as many other applications. However, a long-standing challenge for NATs is the learning of multi-modality data distribution, which is the main cause of the performance gap between NATs and ATs. In this paper, we propose to ease the difficulty of modality learning via sampling from the model distribution instead of the data distribution. We derive contrastive constraints to stabilize the training process and integrate this resulting objective with the state-of-the-art NAT architecture DA-Transformer. Our model \method is examined on 3 different tasks, including machine translation, text summarization, and paraphrasing with 5 benchmarks. Results show that our approach outperforms previous non-autoregressive baselines by a significant margin and establishes new state-of-the-art results for non-autoregressive transformers on all the benchmarks.
Diffusion models are a class of flexible generative models trained with an approximation to the log-likelihood objective. However, most use cases of diffusion models are not concerned with likelihoods, but instead with downstream objectives such as human-perceived image quality or drug effectiveness. In this paper, we investigate reinforcement learning methods for directly optimizing diffusion models for such objectives. We describe how posing denoising as a multi-step decision-making problem enables a class of policy gradient algorithms, which we refer to as denoising diffusion policy optimization (DDPO), that are more effective than alternative reward-weighted likelihood approaches. Empirically, DDPO is able to adapt text-to-image diffusion models to objectives that are difficult to express via prompting, such as image compressibility, and those derived from human feedback, such as aesthetic quality. Finally, we show that DDPO can improve prompt-image alignment using feedback from a vision-language model without the need for additional data collection or human annotation.
Scientific literature understanding tasks have gained significant attention due to their potential to accelerate scientific discovery. Pre-trained language models (LMs) have shown effectiveness in these tasks, especially when tuned via contrastive learning. However, jointly utilizing pre-training data across multiple heterogeneous tasks (e.g., extreme classification, citation prediction, and literature search) remains largely unexplored. To bridge this gap, we propose a multi-task contrastive learning framework, SciMult, with a focus on facilitating common knowledge sharing across different scientific literature understanding tasks while preventing task-specific skills from interfering with each other. To be specific, we explore two techniques -- task-aware specialization and instruction tuning. The former adopts a Mixture-of-Experts Transformer architecture with task-aware sub-layers; the latter prepends task-specific instructions to the input text so as to produce task-aware outputs. Extensive experiments on a comprehensive collection of benchmark datasets verify the effectiveness of our task-aware specialization strategy in various tasks, where we outperform state-of-the-art scientific LMs.
Knowing the language of an input text/audio is a necessary first step for using almost every natural language processing (NLP) tool such as taggers, parsers, or translation systems. Language identification is a well-studied problem, sometimes even considered solved; in reality, most of the world's 7000 languages are not supported by current systems. This lack of representation affects large-scale data mining efforts and further exacerbates data shortage for low-resource languages. We take a step towards tackling the data bottleneck by compiling a corpus of over 50K parallel children's stories in 350+ languages and dialects, and the computation bottleneck by building lightweight hierarchical models for language identification. Our data can serve as benchmark data for language identification of short texts and for understudied translation directions such as those between Indian or African languages. Our proposed method, Hierarchical LIMIT, uses limited computation to expand coverage into excluded languages while maintaining prediction quality.
Debiasing methods that seek to mitigate the tendency of Language Models (LMs) to occasionally output toxic or inappropriate text have recently gained traction. In this paper, we propose a standardized protocol which distinguishes methods that yield not only desirable results, but are also consistent with their mechanisms and specifications. For example, we ask, given a debiasing method that is developed to reduce toxicity in LMs, if the definition of toxicity used by the debiasing method is reversed, would the debiasing results also be reversed? We used such considerations to devise three criteria for our new protocol: Specification Polarity, Specification Importance, and Domain Transferability. As a case study, we apply our protocol to a popular debiasing method, Self-Debiasing, and compare it to one we propose, called Instructive Debiasing, and demonstrate that consistency is as important an aspect to debiasing viability as is simply a desirable result. We show that our protocol provides essential insights into the generalizability and interpretability of debiasing methods that may otherwise go overlooked.
Are language models culturally biased? It is important that language models conform to the cultural aspects of the communities they serve. However, we show in this paper that language models suffer from a significant bias towards Western culture when handling and generating text in Arabic, often preferring, and producing Western-fitting content as opposed to the relevant Arab content. We quantify this bias through a likelihood scoring-based metric using naturally occurring contexts that we collect from online social media. Our experiments reveal that both Arabic monolingual and multilingual models exhibit bias towards Western culture in eight different cultural aspects: person names, food, clothing, location, literature, beverage, religion, and sports. Models also tend to exhibit more bias when prompted with Arabic sentences that are more linguistically aligned with English. These findings raise concerns about the cultural relevance of current language models. Our analyses show that providing culture-indicating tokens or culturally-relevant demonstrations to the model can help in debiasing.