Annotating Bridging Anaphors in Italian: in Search of Reliability
Summary
Bridging anaphors are definite descriptions whose referent is new but is inferred from an earlier entity. They are hard to resolve and hard for annotators to agree on. Caselli and Prodanof propose an XML scheme for Italian bridging FDNPs of the type "definite article + N" and test how reliable it is. The design comes from a corpus study of 17 articles (about 10,000 words) from "il Sole-24 Ore", taken from the SI-TAL Treebank. The authors classified every full definite NP and analysed bridging anchors with Centering Theory. The standoff scheme defines markables for nominal expressions, verbal expressions and other elements, plus links annotated with relation types and salience measures. Annotators follow strict linguistic tests and preference rankings, both for choosing a class and for choosing an anchor. In a three-annotator experiment, agreement reached K = 0.73, and agreement on the bridging class was much higher than in earlier English work. This suggests that tests constrained by local focus make bridging annotation reliable enough to support automatic resolution.
Key points
- Corpus study: 1412 FDNPs were analysed.
- First Mention was the largest class at 58.61%.
- Bridging accounted for 299 cases, or 63.88% of all anaphoric FDNPs. This points to bridging being more productive in Italian than in English.
- Bridging subclasses: lexical relations made up 39.79% and inferential ones 36.45%. Rhetorical relation, discourse topic and event (anchored by a verb) formed the smaller groups.
- Anchors and local focus: 53.48% of anchors were Cb or Cp of earlier sentences. At least 70% appeared in the current sentence or the one just before it, which supports Sidner's local-focus hypothesis.
- Scheme tags:
<ne>,<ve>,<seg>,<link>and<ante>.- Boolean ANAPHORIC and BRIDGING attributes are included.
- The REL attribute takes one of eight values: underspec, title, event, frame, elem, poss, subset or ident.
- LOOKBACK and CENTERING record salience.
- Evaluation: 3 annotators labelled 267 FDNPs from 3 articles.
- Overall K was 0.73, computed on 257 items with the Doubt class excluded.
- All annotators agreed on 70.41% of items.
- Per-class agreement on Bridging was 82.08%, compared with 59% and 31% in Vieira's (1998) experiments.
Cited by
- Cascading Collective Classification for Bridging Anaphora Recognition using a Rich Linguistic Feature Set
- Global Inference for Bridging Anaphora Resolution
- Unrestricted Bridging Resolution
- Bridging Resolution: A Survey of the State of the Art
Cite
@inproceedings{caselli2006annotating,
title = {{Annotating Bridging Anaphors in Italian: in Search of Reliability}},
author = {Caselli, Tommaso and Prodanof, Irina},
booktitle = {Proceedings of the Fifth International Conference on Language Resources and Evaluation (LREC'06)},
year = {2006},
doi = {10.63317/3fkqf3cbny28},
url = {https://doi.org/10.63317/3fkqf3cbny28}
}References
- Asher, N and A. Lascarides.1998, Bridging, in Journal of Semantics vol. 15 (1): 83-113. find
- Caselli T. 2005 Lexical anaphora: a corpus-based model and an XML annotation scheme for bridging anaphora in Italian, M.A. Thesis, University of Pavia Chierchia, G.1995, Dynamics of Meaning: anaphora, presuppositions and the Theory of Grammar, University of Chicago Press, Chicago. find
- Fraurud, K.1990, Definiteness and the Processing of Noun Phrase in Natural Discourse in Journal of Semantics, vol. 7 (4): 395-433. find
- Gardent, C. et alii.2003, Which bridges for bridging FDNPs?, in Proceedings of the Fourth International Workshop on Linguistically Interpreted Corpora, Budapest, Hungary. find
- Grosz, B. et alii.1995, Centering: A framework for modelling the local coherence of discourse, in Computational Linguistics, vol. 21 (2): 202-225. find
- Kleiber, G.1999, Associative anaphora and part-whole relationship: the condition of alienation and the principle of ontological congruence, in Journal of Pragmatics, vol. 31: 339-362. find
- Korzen, I.2003, Anafora associativa: aspetti lessicali, testuali e contestuali, in Atti del XXXIV Congresso Internazionale di studi della Società di Linguistica Italiana (SLI). Firenze 19-21 ottobre 2000, Maraschino, N. and T. Poggi Salani (eds)Bulzoni, Roma. find
- Poesio, M.2003, Associative FDNPs and salience, in Proceedings of the EACL Workshop on Computational Treatments of Anaphora, Budapest. find
- Poesio, M., R. Stevenson, B Di Eugenio and J. Hitzeman.2004a, Centering: A Parametric theory and its instantiations in Computational Linguistics, vol. 30 (3). find
- Poesio M., R Metha, A. Maroudas and J. Hitzeman. 2004b, Learning to Resolve Bridging References, in Proceeding of ACL-04. find our card
- Prince, E.1981, Toward a Taxonomy of given-new information, in Radical Pragmatics, P. Cole (ed), Academic Press, New York. find
- Vieira R.1998 Definite Description Processing in Unrestricted Text, Ph.D. Thesis, University of Edinburgh. find
- Vieira, R. and M. Poesio.2000, An Empirically-Based System for Processing FDNPs, in Computational Linguistics, vol. 26 (4): 539-593. find
This page in Markdown → https://md.research.elamur.ai/free/papers/caselli-2006-annotating-bridging-anaphors-italian-search/