TY - GEN
T1 - Spanning tree approaches for statistical sentence generation
AU - Wan, Stephen
AU - Dras, Mark
AU - Dale, Robert
AU - Paris, Cécile
PY - 2010
Y1 - 2010
N2 - In abstractive summarisation, summaries can include novel sentences that are generated automatically. In order to improve the grammaticality of the generated sentences, we model a global (sentence) level syntactic structure. We couch statistical sentence generation as a spanning tree problem in order to search for the best dependency tree spanning a set of chosen words. We also introduce a new search algorithm for this task that models argument satisfaction to improve the linguistic validity of the generated tree. We treat the allocation of modifiers to heads as a weighted bipartite graph matching problem (also known as the assignment problem), a well studied problem in graph theory. Using BLEU to measure performance on a string regeneration task, we demonstrate an improvement over standard language model baselines, illustrating the benefit of the spanning tree approach incorporating an argument satisfaction model.
AB - In abstractive summarisation, summaries can include novel sentences that are generated automatically. In order to improve the grammaticality of the generated sentences, we model a global (sentence) level syntactic structure. We couch statistical sentence generation as a spanning tree problem in order to search for the best dependency tree spanning a set of chosen words. We also introduce a new search algorithm for this task that models argument satisfaction to improve the linguistic validity of the generated tree. We treat the allocation of modifiers to heads as a weighted bipartite graph matching problem (also known as the assignment problem), a well studied problem in graph theory. Using BLEU to measure performance on a string regeneration task, we demonstrate an improvement over standard language model baselines, illustrating the benefit of the spanning tree approach incorporating an argument satisfaction model.
UR - http://www.scopus.com/inward/record.url?scp=77956325340&partnerID=8YFLogxK
U2 - 10.1007/978-3-642-15573-4_2
DO - 10.1007/978-3-642-15573-4_2
M3 - Conference proceeding contribution
AN - SCOPUS:77956325340
SN - 3642155723
SN - 9783642155727
VL - 5790 LNAI
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 13
EP - 44
BT - Empirical Methods in Natural Language Generation - Data-Oriented Methods and Empirical Evaluation
PB - Springer, Springer Nature
CY - Berlin; Heidelberg
T2 - 12th European Workshop on Natural Language Generation, ENLG 2009
Y2 - 30 March 2009 through 3 April 2009
ER -