Lost in Translation? Predicting Party Group Affiliation from European Parliament Debates∗

Ten countries joined the European Union in 2004. This offered a rare opportunity for the existing party groups to substantively increase their share of the seats in the European Parliament by recruiting national party delegations from the new member states. As most of new member states have a relative short history of competitive multiparty system, there were weaker ties between parties in new and old member states when compared to previous rounds of enlargement. Since the allocation of office spoils in the EP is fairly proportional amongst party groups ,(Mamadouh and Raunio, 2003) it was assumed that national parties from the new member states – less ideologically committed to any of the belief systems held by the traditional Western European party families– would shift the allocation of some offices in the EP by opting to join certain party group who controlled a larger share of office spoils. In this paper we provide a novel approach to evaluate whether the party group choice of the national parties from the new member states may have been affected by such considerations. By employing a Support Vector Machine, a common classification tool from computational linguistics, we evaluate to what extent it is possible to correctly predict the party group affiliation of participants in European Parliament debates on the basis of the content of their speeches. Our results show that the differences in belief systems between party groups as expressed by Members of the European Parliament (MEPs) plenary session speeches in the 5 (1999-2004) European Parliament are helpful in predicting the party group affiliation of MEPs from older member states in the 6 European Parliament. However, these difference are not helpful when trying to predict the party group affiliation of MEPs from the new member states. This suggests that the national parties of these new member states opted to join the more established party groups for other reasons than similarity in belief systems. ∗We would like to thank Indraneel Sircar for assistance in collecting the speech data and Bei Wu for help with the initial analysis of the data. All errors remain our own. Bjorn Hoyland would like to acknowledge the generous financial support provided by his department. Legislatures aim to produce two essential tasks in democratic societies; legislate and provide a forum for the debate over policy alternatives. While the literature on the European Parliament (EP) suggests that it is a powerful legislature, most scholars agree that EP plenary sessions do not provide a forum for European public debates (Eriksen and Fossum, 2000). This can be explained by several factors. First, Members of the European Parliament (MEPs) are allowed to speak in any of the official languages of the Union. The speech is immediately translated and communicated to the rest of the chamber and to the audience with headphones. This small translation delay means that there is little room for spontaneity, unlike in other legislatures like the British Westminster Parliament or the U.S. Congress. Second, voting does not normally occur on the same day as the debate. And third, the debates are usually poorly attended. This probably explains why there is very little media attention focusing on the debate of the EU Parliament. There is little reason to expect that party groups exercise strong control over the content of individual speeches in the debates of the plenary as it is neither likely to change the outcome of the vote nor to be reported widely in the press. The debates do however relate directly to the policy issues facing the legislators. Furthermore, speeches allow for far more nuances than voting decisions. These debate could thus provide available source of information about politics in the European Parliament. Text is becoming an increasingly common source of data in political science. There has been several innovate projects proposing new methods for extracting actors ideal points on the basis of the text (Laver and Garry, 2000; Laver, Benoit and Garry, 2003; Monroe and Maeda, 2004; Slapin and Proksch, 2008). Our project is less ambitious, we are simply interested in correctly classifying the ideological party group affiliation of members in the European Parliament. Instead of developing custom software for the classification of political science text (Hopkins and King, 2007), we borrow an existing classification tool from computational linguistics, Support Vector Machine (Baayen, 2008). In this paper, we utilize debates to evaluate to what extent national parties from the new member states joined party groups with whom they share similar belief systems, as expressed through choice of words in plenary debates. We use a common classification algorithm, Support Vector Machine, to predict party group affiliation of EP plenary