We present a simple but very effective approach to identifying high-quality data in noisy data sets for structured problems like parsing, by greedily exploiting partial structures. We analyze our approach in an annotation projection framework for dependency trees, and show how dependency parsers from two different paradigms (graph-based and transition-based) can be trained on the resulting tree fragments. We train parsers for Dutch to evaluate our method and to investigate to which degree graph-based and transitionbased parsers can benefit from incomplete training data. We find that partial correspondence projection gives rise to parsers that outperform parsers trained on aggressively filtered data sets, and achieve unlabeled attachment scores that are only 5 % behind the average UAS for Dutch in the CoNLL-X Shared Task on supervised parsing (Buchholz and

Various methods aim at overcoming the shortage of NLP resources, especially for resource-poor languages. We present a cross-lingual projection account that aims at inducing an annotated treebank to be used for parser induction for Polish. Our approach builds on Hwa et al.’s projection method [7] that we adapt to the LFG framework. The goal of the experiment is the induction of an LFG f-structure bank for Polish. The projection yields competitive results. The resulting f-structure bank may be used to train a dependency parser for Polish, or for automatic induction of a probabilistic LFG grammar. 1

We present here a method for automatically projecting structural information across translations, including canonical citation structure (such as chapters and sections), speaker information, quotations, markup for people and places, and any other element in TEI-compliant XML that delimits spans of text that are linguistically symmetrical in two languages. We evaluate this technique on two datasets, one containing perfectly transcribed texts and one containing errorful OCR, and achieve an accuracy rate of 88.2 % projecting 13,023 XML tags from source documents to their transcribed translations, with an 83.6 % accuracy rate when projecting to texts containing uncorrected OCR. This approach has the potential to allow a highly granular multilingual digital library to be bootstrapped by applying the knowledge contained in a small, heavily curated collection to a much larger but unstructured one.

(TLT8) was held at the Catholic University of the Sacred Heart in Milan