Multi-Reward Reinforced Summarization with Saliency and Entailment

Pasunuru, Ramakanth; Bansal, Mohit

doi:10.18653/v1/n18-2102

Cited by 135 publications

(150 citation statements)

References 35 publications

(80 reference statements)

Supporting

Mentioning

140

Contrasting

Order By: Relevance

“…RL has been applied to both extractive and abstractive summarisation in recent years (Ryang and Abekawa 2012;Rioux et al 2014;Gkatzia et al 2014;Henß et al 2015;Paulus et al 2017;Pasunuru and Bansal 2018;Kryscinski et al 2018). Most existing RL-based document summarisation systems either use heuristic functions (e.g., Ryang and Abekawa 2012;Rioux et al 2014), which do not rely on reference summaries, or ROUGE scores requiring reference summaries as the rewards for RL (Paulus et al 2017;Pasunuru and Bansal 2018;Kryscinski et al 2018). However, neither ROUGE nor the heuristics-based rewards can precisely reflect real users' requirements on summaries (Chaganty et al 2018); hence, using these imprecise rewards can severely mislead the RL-based summariser.…”

Section: Introductionmentioning

confidence: 99%

Preference-based interactive multi-document summarisation

2019

View full text Add to dashboard Cite

Interactive NLP is a promising paradigm to close the gap between automatic NLP systems and the human upper bound. Preference-based interactive learning has been successfully applied, but the existing methods require several thousand interaction rounds even in simulations with perfect user feedback. In this paper, we study preference-based interactive summarisation. To reduce the number of interaction rounds, we propose the Active Preferencebased ReInforcement Learning (APRIL) framework. APRIL uses Active Learning to query the user, Preference Learning to learn a summary ranking function from the preferences, and neural Reinforcement Learning to efficiently search for the (near-)optimal summary. Our results show that users can easily provide reliable preferences over summaries and that APRIL outperforms the state-of-the-art preference-based interactive method in both simulation and real-user experiments. 1 We first introduced APRIL in (Gao et al. 2018). Towards the end of §1 we discuss how this article substantially extends our previous work.

show abstract

Section: Introductionmentioning

confidence: 99%

Preference-based interactive multi-document summarisation

2019

View full text Add to dashboard Cite

show abstract

“…RL is also gaining popularity as it can directly optimize non-differentiable metrics (Pasunuru and Bansal, 2018;Venkatraman et al, 2015;. proposed an intra-decoder model and combined RL and MLE to deal with summaries with bad qualities.…”

Section: Related Workmentioning

confidence: 99%

Clickbait? Sensational Headline Generation with Auto-tuned Reinforcement Learning

Madotto

et al. 2019

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conferen

View full text Add to dashboard Cite

Sensational headlines are headlines that capture people's attention and generate reader interest. Conventional abstractive headline generation methods, unlike human writers, do not optimize for maximal reader attention. In this paper, we propose a model that generates sensational headlines without labeled data. We first train a sensationalism scorer by classifying online headlines with many comments ("clickbait") against a baseline of headlines generated from a summarization model. The score from the sensationalism scorer is used as the reward for a reinforcement learner. However, maximizing the noisy sensationalism reward will generate unnatural phrases instead of sensational headlines. To effectively leverage this noisy reward, we propose a novel loss function, Auto-tuned Reinforcement Learning (ARL), to dynamically balance reinforcement learning (RL) with maximum likelihood estimation (MLE). Human evaluation shows that 60.8% of samples generated by our model are sensational, which is significantly better than the Pointer-Gen baseline (See et al., 2017) and other RL models.

show abstract

“…ext-oracle further shows that improved sentence selection would bring further performance gains to extractive approaches. Abstractive systems trained on these datasets often have a hard time beating the lead, let alone ext-oracle, or display a low degree of novelty in their summaries (See et al, 2017;Tan & Wan, 2017;Paulus et al, 2018;Pasunuru & Bansal, 2018;Celikyilmaz et al, 2018;Gehrmann et al, 2018). Interestingly, lead and ext-oracle perform poorly on XSum underlying the fact that it contains genuinely abstractive summaries.…”

Section: How Abstractive Is Xsum?mentioning

confidence: 99%

“…However, these datasets often favor extractive models which create a summary by identifying (and subsequently concatenating) the most important sentences in a document (Cheng & Lapata, 2016;Nallapati, Zhai, & Zhou, 2017;Narayan et al, 2018b). Abstractive approaches, despite being more faithful to the actual summarization task -professional editors employ various rewrite operations to transform article sentences into a summary including compression, aggregation, and paraphrasing (Jing, 2002) aside from writing sentences from scratch -they either lag behind extractive ones or are mostly extractive, exhibiting a small degree of abstraction (See et al, 2017;Paulus et al, 2018;Pasunuru & Bansal, 2018;Celikyilmaz et al, 2018;Gehrmann, Deng, & Rush, 2018).…”

Section: Introductionmentioning

confidence: 99%

“…. State-of-the-art abstractive systems on the CNN/Daily mail and New York Times datasets(Celikyilmaz et al, 2018;Pasunuru & Bansal, 2018;Kryściński et al, 2018) use reinforcement learning to directly optimize the evaluation metric relevant for the summarization task. Although our model could be optimized with reinforcement learning objectives, we leave this to future work and present comparisons with related models which are all trained with the maximum-likelihood objective.…”

mentioning

confidence: 99%

See 1 more Smart Citation

What is this Article about? Extreme Summarization with Topic-aware Convolutional Neural Networks

Narayan¹,

Cohen²,

Lapata³

2019

jair

View full text Add to dashboard Cite

We introduce extreme summarization, a new single-document summarization task which aims at creating a short, one-sentence news summary answering the question "What is the article about?". We argue that extreme summarization, by nature, is not amenable to extractive strategies and requires an abstractive modeling approach. In the hope of driving research on this task further: (a) we collect a real-world, large scale dataset by harvesting online articles from the British Broadcasting Corporation (BBC); and (b) propose a novel abstractive model which is conditioned on the article's topics and based entirely on convolutional neural networks. We demonstrate experimentally that this architecture captures long-range dependencies in a document and recognizes pertinent content, outperforming an oracle extractive system and state-of-the-art abstractive approaches when evaluated automatically and by humans on the extreme summarization dataset. 1 Summary: A man and a child have been killed after a light aircraft made an emergency landing on a beach in Portugal.Document: Authorities said the incident took place on Sao Joao beach in Caparica, south-west of Lisbon. The National Maritime Authority said a middle-aged man and a young girl died after they were unable to avoid the plane. The plane's only two occupants were unharmed, it added. The Diario de Noticias newspaper quoted an eyewitness who said the plane had been flying at a low altitude over the beach, although he did not realise anything was wrong until other beachgoers began running. One young witness told Reuters news agency: "I was near the water when I saw the plane. I called my parents, the plane fell on the sand and ran over two people, fatally hurting them and another was injured, I think, but I'm not sure, people were running away." "The plane is still there, but the ambulances arrived quickly. I think maybe the fuel ran out because I find it weird that it landed on the beach." Other reports said the victims had been sunbathing when the plane made its emergency landing. The Associated Press news agency said the girl who died had been with her parents, who were unhurt. The agency quoted witnesses from local television broadcasts. Joao Quadros, who was on the beach, tweeted photos of the aftermath, saying the plane had passed by his son by a matter of metres. There had been no noise, he said. Video footage from the scene carried by local broadcasters showed a small recreational plane parked on the sand, apparently intact and surrounded by beachgoers and emergency workers. One wing seemed to be misaligned in those photos. The cause of the emergency landing remains unclear.2. Experiments with Transformer architectures are outside the scope of this paper. Recent work (Perez-Beltrachini, Liu, & Lapata, 2019) on multiodocument summarization shows that Transformer-based models perform on par with their convolutional alternatives.

show abstract

Multi-Reward Reinforced Summarization with Saliency and Entailment

Cited by 135 publications

References 35 publications

Preference-based interactive multi-document summarisation

Preference-based interactive multi-document summarisation

Clickbait? Sensational Headline Generation with Auto-tuned Reinforcement Learning

What is this Article about? Extreme Summarization with Topic-aware Convolutional Neural Networks

Contact Info

Product

Resources

About