You Talking to Me? A Corpus and Algorithm for Conversation Disentanglement

1 Jun 2008  ·  Micha Elsner, Eugene Charniak ·

When multiple conversations occur simultaneously, a listener must decide which conversation each utterance is part of in order to interpret and respond to it appropriately. We refer to this task as disentanglement. We present a corpus of Internet Relay Chat (IRC) dialogue in which the various conversations have been manually disentangled, and evaluate annotator reliability. This is, to our knowledge, the first such corpus for internet chat. We propose a graph-theoretic model for disentanglement, using discourse-based features which have not been previously applied to this task. The model’s predicted disentanglements are highly correlated with manual annotations.

PDF Abstract


Task Dataset Model Metric Name Metric Value Global Rank Benchmark
Conversation Disentanglement irc-disentanglement Linear VI 82.1 # 4
1-1 51.4 # 3
P 12.1 # 5
R 21.5 # 4
F 15.5 # 5
Conversation Disentanglement Linux IRC (Ch2 Elsner) Linear 1-1 53.1 # 1
Local 81.9 # 1
Shen F-1 55.1 # 1
Conversation Disentanglement Linux IRC (Ch2 Kummerfeld) Linear 1-1 59.7 # 1
Local 80.8 # 1
Shen F-1 63.0 # 1


No methods listed for this paper. Add relevant methods here