File Information
File: 05-lr/acl_arc_1_sum/cleansed_text/xml_by_section/concl/04/p04-2010_concl.xml
Size: 2,033 bytes
Last Modified: 2025-10-06 13:54:09
<?xml version="1.0" standalone="yes"?> <Paper uid="P04-2010"> <Title>A Machine Learning Approach to German Pronoun Resolution</Title> <Section position="7" start_page="0" end_page="0" type="concl"> <SectionTitle> 6 Conclusion and Future Work </SectionTitle> <Paragraph position="0"> The pronoun resolution system presented in this paper performs well for unannotated text of a limited domain. While the results are encouraging considering the knowledge-poor approach, experiments with a more complex textual domain show that the system is unsuitable for wide-coverage tasks such as question answering and summarisation. null To examine whether the system would yield comparable results in unrestricted text, it needs to be trained on a more diverse and possibly larger corpus. For this purpose, T&quot;uba-D/Z, a treebank consisting of German newswire text, is presently being annotated with coreference information. As the syntactic annotation of the treebank is richer than that of the HTC corpus, additional features may be derived from it. Experiments with T&quot;uba-D/Z will show whether the performance achieved for the HTC test set is scalable.</Paragraph> <Paragraph position="1"> For future versions of the system, it might also HTC-Gold HTC-Test Spiegel Baseline accuracy 46.7% 30.9% 31.1% Classification Fa0a2a1a4a3 score 77.9 62.8 30.4 Best-first, ignoring pronominal ant. 82.0% 67.2% 28.3% Best-first, resolving pronominal ant. 72.2% 49.1% 21.7% Closest-first, ignoring pronominal ant. 82.0% 57.3% 34.4% Closest-first, resolving pronominal ant. 72.2% 49.1% 22.8% F-score. HTC-Gold and HTC-Test stand for manually and automatically annotated test sets, respectively. be beneficial to use full parses instead of chunks.</Paragraph> <Paragraph position="2"> As most German verbs are morphologically unambiguous, an analysis of them could help disambiguate pronouns. However, due to the relatively free word order of the German language, this approach requires extensive reseach.</Paragraph> </Section> class="xml-element"></Paper>