Mirella Lapata · page 2
Papers
- Sentence Compression Beyond Word Deletion (2008 · 251 citations)
Cohn and Lapata extend sentence compression beyond word deletion. In their abstractive setting, a sentence may also be shortened by substituting, reordering, or inserting words. Existing compression corpora contain only deletions, and paraphrase corpora are not designed around information loss. So the authors built a… - Automatic Evaluation of Information Ordering: Kendall's Tau (2006 · 248 citations)
Information ordering means arranging preselected content units, such as sentences, propositions, database facts or discourse trees, into a coherent sequence. - Bayesian Word Sense Induction (2009 · 234 citations)
The paper addresses word sense induction: discovering a word's senses directly from raw text rather than from a fixed inventory such as a dictionary. - Neural Latent Extractive Document Summarization (2018 · 220 citations)
Extractive summarizers are usually trained on sentence-level True/False labels. Most datasets supply only document–summary pairs, so these labels must be derived heuristically, for example by choosing the sentence subset with the highest ROUGE against the reference, and they may be suboptimal. Zhang, Lapata, Wei and… - A Bayesian Model of Diachronic Meaning Change (2016 · 206 citations)
Frermann and Lapata address the automatic tracking of how word meanings evolve over history, which can support lexicography as well as information retrieval and question answering. - Learning Structured Text Representations (2018 · 203 citations)
Liu and Lapata look for a way to make neural document encoders aware of discourse structure without annotated discourse corpora or an external parser, whose mistakes spread through pipeline systems. - Machine Translation by Triangulation: Making Effective Use of Multi-Parallel Corpora (2007 · 190 citations)
Phrase-based SMT degrades sharply when bilingual training data is scarce: phrase translation probabilities become unreliable and many longer phrases are never seen. - Data-to-text Generation with Entity Modeling (2019 · 181 citations)
Neural encoder-decoder systems that turn structured tables into text usually handle entities such as players and teams as ordinary vocabulary items. - Unsupervised Opinion Summarization as Copycat-Review Generation (2020 · 173 citations)
The paper tackles abstractive opinion summarization without any reference summaries for training. Gold summaries are costly to create, and review data covers many domains. The authors present Copycat, a model trained end-to-end only on large collections of product and business reviews from Amazon and Yelp. It is a… - Graph Connectivity Measures for Unsupervised Word Sense Disambiguation (2007 · 168 citations)
Supervised word sense disambiguation (WSD) depends on costly sense-annotated data, so Navigli and Lapata study unsupervised graph-based WSD. - Multiple Instance Learning Networks for Fine-Grained Sentiment Analysis (2018 · 154 citations)
Document-level star ratings are plentiful, but labels for individual sentences or clauses are expensive to collect. - Extractive Opinion Summarization in Quantized Transformer Spaces (2021 · 151 citations)
Opinion summarization means condensing what many reviewers say about one entity. Training references barely exist, inputs can reach thousands of reviews, and users may want either an overview or a focus on one aspect. Abstractive systems that learn by reconstructing reviews tend to hallucinate and slip into… - Improving Language Model Negotiation with Self-Play and In-Context Learning from AI Feedback (2023 · 132 citations)
The paper asks whether large language models can make each other better negotiators with little human involvement, in the spirit of AlphaGo Zero's self-play. - Movie Script Summarization as Graph-based Scene Extraction (2015 · 122 citations)
Gorinski and Lapata introduce screenplay summarization as a new task. They argue it could help script readers, who must sift through thousands of submissions, by supporting browsing and giving a fast sense of a plot. They treat a summary as an extracted chain of consecutive scenes. A good chain should be important to… - Constructing Corpora for the Development and Evaluation of Paraphrase Systems (2008 · 118 citations)
Paraphrase acquisition research has had no shared benchmark. Systems were usually judged by people rating sample outputs, which is slow, hard to reproduce and says almost nothing about recall. Cohn, Callison-Burch and Lapata address this gap by building a monolingual parallel corpus in which paraphrases are made… - Aspect-Controllable Opinion Summarization (2021 · 114 citations)
Most opinion summarizers condense many reviews into one generic summary built around the most frequently voiced opinions. - Discourse Constraints for Document Compression (2010 · 105 citations)
Most sentence compression systems shorten each sentence on its own and ignore the surrounding text. Real applications such as summarization, however, need a shorter document that still reads coherently. Clarke and Lapata address this gap with a compression model that operates at the document level and accounts for… - Verb Class Disambiguation Using Informative Priors (2004 · 100 citations)
Levin's (1993) verb classes group English verbs by shared diathesis alternations, but the inventory only lists which classes a verb can belong to. - Data-to-text Generation with Macro Planning (2021 · 98 citations)
Neural encoder-decoder systems for data-to-text generation write fluent prose. However, they often hallucinate facts, select content poorly and order it incoherently, and these problems worsen in long, multi-paragraph outputs such as sports reports. Puduppully and Lapata revisit the classic pipeline idea of… - Inferring Sentence-internal Temporal Relations (2004 · 83 citations)
Much temporal information in text is implicit. A reader can tell which of two events came first even without an explicit cue. Lapata and Lascarides ask whether such ordering knowledge can be learned without hand-annotated temporal data. They use sentences in which a main clause and a subordinate clause are linked by… - Conditional Generation with a Question-Answering Blueprint (2023 · 70 citations)
Neural seq-to-seq generators frequently hallucinate, omit important details, and act as black boxes that are hard to inspect or steer. - Context-aware Frame-Semantic Role Labeling (2015 · 69 citations)
FrameNet-style semantic role labeling usually assigns roles one sentence at a time and learns from a small annotated dataset. Yet the correct role often depends on the surrounding discourse. For example, "Greece" in a delivery event could be a destination (GOAL) or a political actor (RECIPIENT). - Learning an Executable Neural Semantic Parser (2019 · 63 citations)
Semantic parsing converts a question into a machine-readable program that can be run against a knowledge base or database to get an answer. - QAmeleon: Multilingual QA with Only 5 Examples (2023 · 45 citations)
Extractive question answering datasets are large mainly for English, so QA systems in other languages usually depend on zero-shot transfer or machine-translated training data. - Learning Contextually Informed Representations for Linear-Time Discourse Parsing (2017 · 44 citations)
The paper addresses a trade-off in RST discourse parsing. Cubic-time CKY-style parsers reach top accuracy but are slow on long documents. Linear-time parsers are fast but usually depend on local, hand-engineered features. The authors present CIDER, a greedy, bottom-up, two-stage parser. It first builds a discourse…
This page in Markdown → https://md.research.elamur.ai/free/en/authors/mirella-lapata/page/2/