We propose a new multilingual content analysis method for estimating conflict between opposition and government from parliamentary speeeches. While the burgeoning field of quantitative text analysis has provided useful tools to extract policy positions from text or estimate sentiment of political speech, the application of such techniques is hitherto limited to monolingual contexts. We present a possible solution to this problem by presenting a reliable multilingual approach to estimating sentiment from speeches. The sentiment analysis approach can be easily combined with existing text similarity measures to produce estimates of political conflict. We present a range of validation exercises using debates from several parliaments and compare our approach to hand-coding and unsupervised scaling techniques. Finally, we discuss ways how the approach can be generalized to other comparative settings.
