### Human coherence judgements

The file humans.avg contains average human ratings on the coherence of
DUC summaries. The collection of the data is detailed in the
CL manuscript in Section 5 and also in the following paper (Section 4.2):

Regina Barzilay and Mirella Lapata. 2005.  Modeling Local Coherence:
An Entity-Based Approach In Proceedings of the 43rd Annual Meeting of
the Association for Computational Linguistics, 141-148. Ann Arbor.


The file reads as follows: 

- the first column describes the DUC document cluster from which the
  summary was generated 
- the second column describes the system or human that generated the summary 
- the third column is the average
  rating of the coherence of the summary

So, for example, the following line: 

D30003  16      5.07142857142857

means that system 16 generated a summary for cluster D30003 whose
rating was 5.07142857142857. In our study we used 5 summarization
systems: 13, 16, 18, 26, and 6. We also used human written summaries
(by DUC assessors). The latter are denoted with the letter H.

The summaries were taken from systems that participated in DUC
2003. You should register with DUC to obtain access to the original
summaries. 

### LSA scores for coherence experiments

data1-test-lsa.tar.gz: LSA scores for test data for ordering
		       experiments (Earthquakes domain, Experiment 1)
data2-test-lsa.tar.gz: LSA scores for test data for ordering
		       experiments (Accidents domin Experiment 1)
lsa_summaries_test:    LSA scores for test data for summarisation
	               experiments

### Readability experiments 

The data for the readability experiments is in the readability/
directory. We used 5-fold cross validation. The file names mean the following:


svm_perplex[1-5]_grid_ana.data:  training files for folds 1-5, for
				 model with coreference
svm_perplex[1-5]_grid_ana.test:  training files for folds 1-5 for
				 model with coreference


svm_perplex[1-5]_grid_ana.data:  training files for folds 1-5, for
				 model with coreference
svm_perplex[1-5]_grid_ana.test:  test files for folds 1-5 for
				 model with coreference


svm_perplex[1-5]_grid-no_ana.data: training files for folds 1-5, for
				   model with coreference

svm_perplex[1-5]_grid-no_ana.test: test files for folds 1-5, for
				   model without coreference

lsa[1-5].test:                     LSA scores for folds 1-5 (test data).