A Study on Causal Relations and its Automatic Identification in Tamil

The objective of the present work is to identify the cause-effect expressions in Tamil text. We have classified the causal markers into different types for computational purposes. The linear order of cause-effect markers and arguments are explained with examples. Tamil corpora consisting of 31,741 sentences were annotated manually for this task. To overcome the structural interdependencies existing in cause-effect relations, we came up with the separate set of features for each type and the type specific models were generated. We have introduced the Sliding window type specific testing approach. Post-processing using linguistic and heuristic rules improves the system performance. We have conducted performance distribution and ten-fold cross validation experiments and the results are encouraging.