... variety of subjects occurring in the training data. In addition, reach-ing a precision of 92.0% for explicit subjects us-ing just 20% of the training data is far more ex-pensive in terms of the ... algorithms, including JRip in WEKA(Măuller, 2006), with precision of 74% and recall of 60%, and K-nearest neighbors in TiMBL: both in (Evans, 2001) with precision of 73% and recall of 69%, and in (Boyd ... Computational LinguisticsElliphant: Improved Automatic Detection of Zero Subjects and Impersonal Constructions in SpanishLuz Rello∗NLP and Web Research GroupsUniv. Pompeu FabraBarcelona, SpainRicardo...