Improving summaries by revising them

Abstract

This paper describes a program which revises a draft text by aggregating together descriptions of discourse entities, in addition to deleting extraneous information. In contrast to knowledgerich sentence aggregation approaches explored in the past, this approach exploits statistical parsing and robust coreference detection. In an evaluation involving revision of topic-related summaries using informativeness measures from the TIPSTER SUMMAC evaluation, the results show gains in informativeness without compromising readability.

Syntactic sentence compression in the biomedical domain: facilitating access to related articles

“…Topiary, the University of Maryland's system which integrates ''parse-and-trim'' techniques with topic term extraction, was among the highest scoring systems for all tasks on all measures-in some cases, even beating the performance of humans in terms of automatic metrics. Other systems that make use of similar techniques include (Mani et al 1999;Jing 2000), and more recently, (Blair-Goldensohn et al 2004;Conroy et al 2005). In our approach, sentence compression is achieved by removing elements-no attempt is made to reorder material within a sentence.…”

“…In our approach, sentence compression is achieved by removing elements-no attempt is made to reorder material within a sentence. Since our task does not involve multiple sentences, there is no opportunity to generate output that combines fragments from different sources, for example, including one sentence as a relative clause inside another (Mani et al 1999). Thus, we conceive of sentence compression solely as the task of selecting sentential elements (words, phrases, clauses, etc.)…”

Automatic Summarization

“…Sentence revision was historically the first language generation task attempted in the context of summarization [89,116,146,147,162]. Sentence revision involves re-using text collected from the input to the summarizer, but parts of the final summary are automatically modified by substituting some expressions with other more appropriate expressions, given the context of the new summary.…”

Automatic Summarization from Multiple Documents

“…Furthermore, the newer, five-step specification of the summarization process by Mani indicates the problem of element identification, which the previous definitions ([MB99] and [ENW04]) seem not to take into account, although it can be part of the analysis step. Also, in [MGB99] we find an elaboration of the condensing (reductive) transformation of text, which is reported to involve the following operations: selection of salient portions of text, aggregation of the information over various selected portions and abstraction of information to a more general level, as is also indicated in [Jon99]. The above indicate that, even though there is serious overlap between definitions, there is no consensus over the detailed description of the process.…”

“…Mani et al in [MGB99] speak of the usefulness of reformulation and implement a text revision mechanism using elimination of sentence constituents, aggregation of sentence constituents and smoothing (which is analysed into reference adjustment and reduction of text within sentence limits).…”