Legal professionals frequently encounter long legal judgments that hold critical insights for their work. While recent advances have led to automated summarization solutions for legal documents, they typically provide generic summaries, which may not meet the diverse information needs of users. To address this gap, we introduce LexAbSumm, a novel dataset designed for aspect-based summarization of legal case decisions, sourced from the European Court of Human Rights jurisdiction. We evaluate several abstractive summarization models tailored for longer documents on LexAbSumm, revealing a challenge in conditioning these models to produce aspect-specific summaries. We release LexAbSum to facilitate research in aspect-based summarization for legal domain.
As human society transitions into the information age, reduction in our attention span is a contingency, and people who spend time reading lengthy news articles are decreasing rapidly and the need for succinct information is higher than ever before. Therefore, it is essential to provide a quick overview of important news by concisely summarizing the top news article and the most intuitive headline. When humans try to make summaries, they extract the essential information from the source and add useful phrases and grammatical annotations from the original extract. Humans have a unique ability to create abstractions. However, automatic summarization is a complicated problem to solve. The use of sequence-to-sequence (seq2seq) models for neural abstractive text summarization has been ascending as far as prevalence. Numerous innovative strategies have been proposed to develop the current seq2seq models further, permitting them to handle different issues like saliency, familiarity, and human lucidness and create excellent synopses. In this article, we aimed toward enhancing the present architectures and models for abstractive text summarization. The modifications have been aimed at fine-tuning hyper-parameters, attempting specific encoder-decoder combinations. We examined many experiments on an extensively used CNN/DailyMail dataset to check the effectiveness of various models.
In this paper, we present Paramanu-Ayn, a collection of legal language models trained exclusively on Indian legal case documents. This 97-million-parameter Auto-Regressive (AR) decoder-only model was pretrained from scratch with a context size of 8192 on a single GPU for just 185 hours, achieving an efficient MFU of 41.35. We also developed a legal domain specialized BPE tokenizer. We evaluated our model using perplexity and zero-shot tasks: case judgment prediction with explanation and abstractive case summarization. Paramanu-Ayn outperformed Llama-2 7B and Gemini-Pro in case judgment prediction with explanation task on test accuracy by nearly 2 percentage points, despite being 72 times smaller. In zero-shot abstractive summarization, it surpassed decoder-only LLMs generating fixed-length summaries (5000 tokens) by over 10 percentage points in BLEU and METEOR metrics, and by nearly 4 percentage points in BERTScore. Further evaluations on zero-shot commonsense and mathematical benchmarks showed that Paramanu-Ayn excelled despite being trained exclusively on legal documents, outperforming Llama-1, Llama-2, and Falcon on AGIEVAL-AQuA-RAT and AGIEVAL-SAT-Math tasks. We also instruction-tuned our model on 10,763 diverse legal tasks, including legal clause generation, legal drafting, case summarization, etc. The Paramanu-Ayn-instruct model scored above 8 out of 10 in clarity, relevance, completeness, and legal reasoning metrics by GPT-3.5-Turbo. We found that our models, were able to learn drafting knowledge and generalize to draft legal contracts and legal clauses with limited instruction-tuning. Hence, we conclude that for a strong domain-specialized generative language model (such as legal), domain specialized pretraining from scratch is more cost effective, environmentally friendly, and remains competitive with larger models or even better than adapting LLMs for legal domain tasks.
Large pre-trained language models (PLMs) are at the forefront of advances in Natural Language Processing. One widespread use case of PLMs is "prompting" - or in-context learning - where a user provides a description of a task and some completed examples of the task to a PLM as context before prompting the PLM to perform the task on a new example. Only the largest, most capable PLMs are able to perform in-context learning effectively, and these models are typically trained with a predominantly English corpus, leaving all other languages behind. The data limitations in most languages preclude the training of language-specific PLMs capable of prompting. Albeit the surge in work of prompting settings, it is still unclear how PLMs should be adapted cross-lingually specifically for prompting. We evaluate the possible methods to adapt LLaMa, a 7B parameter open-source PLM mainly trained in English, for prompting in low-resource languages, namely for Kinyarwanda, Hausa, and Luganda. We consider three methods: few-shot prompting (prompt), language-adaptive fine-tuning (LAFT), and neural machine translation (translate), and evaluate on abstractive summarization, multi-class topic classification, and named-entity recognition. Although LAFT carries the greatest compute cost and intuitively should lead to the best results, our experiments exhibit that LAFT is only occasionally the optimal choice for adapting PLMs for prompting. Rather, the translate and prompt settings are a compute-efficient and cost-effective method of few-shot prompting for the selected low-resource languages. We find that the results are task and language dependent but find that the prompting method is the best on average across all tasks and languages. Results show that the prompt setting performs better than both translating and LAFT with statistical significance for all shots when aggregated across all tasks and languages.
User alignment is crucial for adapting general-purpose language models (LMs) to downstream tasks, but human annotations are often not available for all types of instructions, especially those with customized constraints. We observe that user instructions typically contain constraints. While assessing response quality in terms of the whole instruction is often costly, efficiently evaluating the satisfaction rate of constraints is feasible. We investigate common constraints in NLP tasks, categorize them into three classes based on the types of their arguments, and propose a unified framework, ACT (Aligning to ConsTraints), to automatically produce supervision signals for user alignment with constraints. Specifically, ACT uses constraint verifiers, which are typically easy to implement in practice, to compute constraint satisfaction rate (CSR) of each response. It samples multiple responses for each prompt and collect preference labels based on their CSR automatically. Subsequently, ACT adapts the LM to the target task through a ranking-based learning process. Experiments on fine-grained entity typing, abstractive summarization, and temporal question answering show that ACT is able to enhance LMs' capability to adhere to different classes of constraints, thereby improving task performance. Further experiments show that the constraint-following capabilities are transferable.
Extensive efforts in the past have been directed toward the development of summarization datasets. However, a predominant number of these resources have been (semi)-automatically generated, typically through web data crawling, resulting in subpar resources for training and evaluating summarization systems, a quality compromise that is arguably due to the substantial costs associated with generating ground-truth summaries, particularly for diverse languages and specialized domains. To address this issue, we present ACLSum, a novel summarization dataset carefully crafted and evaluated by domain experts. In contrast to previous datasets, ACLSum facilitates multi-aspect summarization of scientific papers, covering challenges, approaches, and outcomes in depth. Through extensive experiments, we evaluate the quality of our resource and the performance of models based on pretrained language models and state-of-the-art large language models (LLMs). Additionally, we explore the effectiveness of extractive versus abstractive summarization within the scholarly domain on the basis of automatically discovered aspects. Our results corroborate previous findings in the general domain and indicate the general superiority of end-to-end aspect-based summarization. Our data is released at https://github.com/sobamchan/aclsum.
Semi-supervised dialogue summarization (SSDS) leverages model-generated summaries to reduce reliance on human-labeled data and improve the performance of summarization models. While addressing label noise, previous works on semi-supervised learning primarily focus on natural language understanding tasks, assuming each sample has a unique label. However, these methods are not directly applicable to SSDS, as it is a generative task, and each dialogue can be summarized in different ways. In this work, we propose a novel scoring approach, SiCF, which encapsulates three primary dimensions of summarization model quality: Semantic invariance (indicative of model confidence), Coverage (factual recall), and Faithfulness (factual precision). Using the SiCF score, we select unlabeled dialogues with high-quality generated summaries to train summarization models. Comprehensive experiments on three public datasets demonstrate the effectiveness of SiCF scores in uncertainty estimation and semi-supervised learning for dialogue summarization tasks. Our code is available at \url{https://github.com/amazon-science/summarization-sicf-score}.
We present a detailed replication study of the BASS framework, an abstractive summarization system based on the notion of Unified Semantic Graphs. Our investigation includes challenges in replicating key components and an ablation study to systematically isolate error sources rooted in replicating novel components. Our findings reveal discrepancies in performance compared to the original work. We highlight the significance of paying careful attention even to reasonably omitted details for replicating advanced frameworks like BASS, and emphasize key practices for writing replicable papers.
Privacy policy documents have a crucial role in educating individuals about the collection, usage, and protection of users' personal data by organizations. However, they are notorious for their lengthy, complex, and convoluted language especially involving privacy-related entities. Hence, they pose a significant challenge to users who attempt to comprehend organization's data usage policy. In this paper, we propose to enhance the interpretability and readability of policy documents by using controlled abstractive summarization -- we enforce the generated summaries to include critical privacy-related entities (e.g., data and medium) and organization's rationale (e.g.,target and reason) in collecting those entities. To achieve this, we develop PD-Sum, a policy-document summarization dataset with marked privacy-related entity labels. Our proposed model, EROS, identifies critical entities through a span-based entity extraction model and employs them to control the information content of the summaries using proximal policy optimization (PPO). Comparison shows encouraging improvement over various baselines. Furthermore, we furnish qualitative and human evaluations to establish the efficacy of EROS.
We present VBART, the first Turkish sequence-to-sequence Large Language Models (LLMs) pre-trained on a large corpus from scratch. VBART are compact LLMs based on good ideas leveraged from BART and mBART models and come in two sizes, Large and XLarge. Fine-tuned VBART models surpass the prior state-of-the-art results in abstractive text summarization, title generation, text paraphrasing, question answering and question generation tasks. They allow fine-tuning for future text generation tasks and datasets, carving a new path for Turkish Natural Language Processing (NLP) research. Our work shows that having a pre-trained LLM for Turkish outperforms up to 3x multilingual models, improving existing results and providing efficient models for training and inference. Moreover, we show that our monolingual tokenizer is up to 11x more efficient than multilingual tokenizers. Last but not least, we introduce a method to enlarge an existing pre-trained LLM and question the relevancy of Chinchilla Scaling Law to sequence-to-sequence masked language models. Our fine-tuned models, tokenizer and cleaned vngrs-web-corpus of 135 GB are publicly available at huggingface.co/vngrs-ai.
Among the various pre-trained neural language models that are popular today, dropout is already an indispensable regularization technique. To solve the inconsistency between training and inference caused by the randomness of dropout, some studies use consistency training to regularize dropout at the output layer. In this paper, we propose a novel Layer-wise Regularized Dropout (LR-Drop), which is specially designed for Transformer-based Language models. Specifically, LR-Drop layer-wise regularizes each Transformer layer using the consistency training strategy. Each training sample passes through the two siamese sub-models sampled by dropout, and then LR-Drop forces the hidden states, multi-head attention matrices, and output distribution of the two siamese sub-models to be consistent. The proposed LR-Drop can be regarded as a "self-distillation" framework, in which each sub-model generated by dropout is the other's "teacher" model and "student" model. Through extensive experiments on 8 natural language understanding datasets, 6 neural machine translation datasets, and 1 abstractive summarization dataset (a total of 15 datasets), we show that LR-Drop achieves superior performances, including state-of-the-art results.
Abstractive summarization models often generate factually inconsistent content particularly when the parametric knowledge of the model conflicts with the knowledge in the input document. In this paper, we analyze the robustness of fine-tuning based summarization models to the knowledge conflict, which we call factual adaptiveness. We utilize pre-trained language models to construct evaluation sets and find that factual adaptiveness is not strongly correlated with factual consistency on original datasets. Furthermore, we introduce a controllable counterfactual data augmentation method where the degree of knowledge conflict within the augmented data can be adjustable. Our experimental results on two pre-trained language models (PEGASUS and BART) and two fine-tuning datasets (XSum and CNN/DailyMail) demonstrate that our method enhances factual adaptiveness while achieving factual consistency on original datasets on par with the contrastive learning baseline.
This paper explores the realm of abstractive text summarization through the lens of the SEASON (Salience Allocation as Guidance for Abstractive SummarizatiON) technique, a model designed to enhance summarization by leveraging salience allocation techniques. The study evaluates SEASON's efficacy by comparing it with prominent models like BART, PEGASUS, and ProphetNet, all fine-tuned for various text summarization tasks. The assessment is conducted using diverse datasets including CNN/Dailymail, SAMSum, and Financial-news based Event-Driven Trading (EDT), with a specific focus on a financial dataset containing a substantial volume of news articles from 2020/03/01 to 2021/05/06. This paper employs various evaluation metrics such as ROUGE, METEOR, BERTScore, and MoverScore to evaluate the performance of these models fine-tuned for generating abstractive summaries. The analysis of these metrics offers a thorough insight into the strengths and weaknesses demonstrated by each model in summarizing news dataset, dialogue dataset and financial text dataset. The results presented in this paper not only contribute to the evaluation of the SEASON model's effectiveness but also illuminate the intricacies of salience allocation techniques across various types of datasets.
Improving factual consistency in abstractive summarization has been a focus of current research. One promising approach is the post-editing method. However, previous works have yet to make sufficient use of factual factors in summaries and suffers from the negative effect of the training datasets. In this paper, we first propose a novel factual error correction model FactCloze based on a conditional-generation cloze task. FactCloze can construct the causality among factual factors while being able to determine whether the blank can be answered or not. Then, we propose a data distillation method to generate a more faithful summarization dataset SummDSC via multiple-dimensional evaluation. We experimentally validate the effectiveness of our approach, which leads to an improvement in multiple factual consistency metrics compared to baselines.
This research addresses the limitations of pretrained models (PTMs) in generating accurate and comprehensive abstractive summaries for scientific articles, with a specific focus on the challenges posed by medical research. The proposed solution named medical text simplification and summarization (MedTSS) introduces a dedicated module designed to enrich source text for PTMs. MedTSS addresses issues related to token limits, reinforces multiple concepts, and mitigates entity hallucination problems without necessitating additional training. Furthermore, the module conducts linguistic analysis to simplify generated summaries, particularly tailored for the complex nature of medical research articles. The results demonstrate a significant enhancement, with MedTSS improving the Rouge-1 score from 16.46 to 35.17 without requiring additional training. By emphasizing knowledge-driven components, this framework offers a distinct perspective, challenging the common narrative of ’more data’ or ’more parameters.’ This alternative approach, especially applicable in health-related domains, signifies a broader contribution to the field of NLP. MedTSS serves as an innovative model that not only addresses the intricacies of medical research summarization but also presents a paradigm shift with implications for diverse domains beyond its initial scope.
Neural abstractive summarization models make summaries in an end-to-end manner, and little is known about how the source information is actually converted into summaries. In this paper, we define input sentences that contain essential information in the generated summary as $\textit{source sentences}$ and study how abstractive summaries are made by analyzing the source sentences. To this end, we annotate source sentences for reference summaries and system summaries generated by PEGASUS on document-summary pairs sampled from the CNN/DailyMail and XSum datasets. We also formulate automatic source sentence detection and compare multiple methods to establish a strong baseline for the task. Experimental results show that the perplexity-based method performs well in highly abstractive settings, while similarity-based methods perform robustly in relatively extractive settings. Our code and data are available at https://github.com/suhara/sourcesum.
Text summarization is the process of condensing a piece of text to fewer sentences, while still preserving its content. Chat transcript, in this context, is a textual copy of a digital or online conversation between a customer (caller) and agent(s). This paper presents an indigenously (locally) developed hybrid method that first combines extractive and abstractive summarization techniques in compressing ill-punctuated or un-punctuated chat transcripts to produce more readable punctuated summaries and then optimizes the overall quality of summarization through reinforcement learning. Extensive testing, evaluations, comparisons, and validation have demonstrated the efficacy of this approach for large-scale deployment of chat transcript summarization, in the absence of manually generated reference (annotated) summaries.
As NLP models become increasingly capable of understanding documents in terms of coherent entities rather than strings, obtaining the most salient entities for each document is not only an important end task in itself but also vital for Information Retrieval (IR) and other downstream applications such as controllable summarization. In this paper, we present and evaluate GUMsley, the first entity salience dataset covering all named and non-named salient entities for 12 genres of English text, aligned with entity types, Wikification links and full coreference resolution annotations. We promote a strict definition of salience using human summaries and demonstrate high inter-annotator agreement for salience based on whether a source entity is mentioned in the summary. Our evaluation shows poor performance by pre-trained SOTA summarization models and zero-shot LLM prompting in capturing salient entities in generated summaries. We also show that predicting or providing salient entities to several model architectures enhances performance and helps derive higher-quality summaries by alleviating the entity hallucination problem in existing abstractive summarization.
State-space models are a low-complexity alternative to transformers for encoding long sequences and capturing long-term dependencies. We propose LOCOST: an encoder-decoder architecture based on state-space models for conditional text generation with long context inputs. With a computational complexity of $O(L \log L)$, this architecture can handle significantly longer sequences than state-of-the-art models that are based on sparse attention patterns. We evaluate our model on a series of long document abstractive summarization tasks. The model reaches a performance level that is 93-96% comparable to the top-performing sparse transformers of the same size while saving up to 50% memory during training and up to 87% during inference. Additionally, LOCOST effectively handles input texts exceeding 600K tokens at inference time, setting new state-of-the-art results on full-book summarization and opening new perspectives for long input processing.
Constitutions are foundational legal documents that underpin the governmental and societal structures. As such, they are a reflection of a nation's cultural and social uniqueness, but also contribute to establish topics of universal importance, like citizens' rights and duties (RD). In this work, using the renowned GPT-3.5, we leverage generative large language models to understand constitutional passages that transcend national boundaries. A key contribution of our study is the introduction of a novel application of abstractive summarization on a multi-source collection of constitutional texts, with a focus on European countries' constitution passages related to RD topics. Our results show the meaningfulness of GPT-3.5 to produce informative, coherent and faithful summaries capturing RD topics across European countries.