{ "paper_id": "M93-1030", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T03:14:33.820470Z" }, "title": "Topic Session on DISCOURSE", "authors": [ { "first": "Damaris", "middle": [ "M" ], "last": "Ayuso", "suffix": "", "affiliation": {}, "email": "" } ], "year": "", "venue": null, "identifiers": {}, "abstract": "", "pdf_parse": { "paper_id": "M93-1030", "_pdf_hash": "", "abstract": [], "body_text": [ { "text": "Wendy Lehnert (UMass) presented a view of the coreference problem from the perspective of machine learning, i n particular, the idea of treating coreference as classification . To pursue this view, the overall coreference proble m needs to be broken up into sub-problems of the appropriate granularity that are amenable to machine-learnin g paradigms. Once a problem is identified, some of the issues that arise are : finding the right set of features to use i n the classifier's training vectors ; determining which features are domain dependent and which are domain independent ; and fording the right size for the feature set, as more features require more training data . Dr. Lehnert presented an example of this approach for learning rules which recognize appositive constructions. UMass created a trainin g interface where all possible appositive constructions are highlighted for a human-in-the-loop, who responds yes/no to each proposal . Some of the domain-independent features used in the training were syntactic in nature (\"does the NP start with an indefinite article?\"), and others were domain dependent (\"does it contain a corporate designator?\") .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Jin Wang (NMSU) contended that although achieving a domain-independent reference resolution module would be desireable, it is an impossible goal . As an example, he cited the case of company descriptions versus equipment descriptions, where semantic coreferential constraints for the two classes are very different. In addition, being able to ex[ress coreference constraints in a declarative way is also an unrealistic goal. MUC-5 had an example justifyin g this assertion : recognizing name aliases--different naming conventions of company names and device names require d treatments which performed surgical operations on strings . A more realistic goal, it was argued, is to design a kind of pre-compiler which provides a user with a library of routines from which to draw . The porting job becomes defming the domain-specific routines specifying the merge methods and conditions which are then combined with th e domain-independent components to produce a reference resolution program for the new domain .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Chinatsu Aone (SRA) described the new multilingual discourse module of the Solomon text processing system . This module is data-driven, with the goal of achieving a module with core algorithms and data-structures that ar e domain and language independent . It uses information in three knowledge bases (KBs): the discourse domain KB , which identifies the discourse phenomena which will be handled in each domain ; the knowledge source KB whic h contains a hierarchy of anaphora resolution strategies, such as filtering functions and possible-referent generatin g functions ; and the discourse phenomena KB, which identifies the knowledge sources to use for each discours e phenomenon. Porting of the discourse module involves extending core information in the three knowledge bases .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Dan Moldovan (USC) introduced a proposal for a computational model of reference resolution based on paralle l marker-passing in a conceptual network . A semantic instantiation of the interpreted text is created in the network , and activators are propagated outward following relational links . The level of activation of nodes is used as an indication of discourse focus. In performing reference resolution, the most active concept which is acceptabl e syntactically and semantically is taken to be the referent. A major outstanding problem being studied is how far t o propagate markers in the network . A parallel processing approach is proposed to handle computational complexity .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Kazunori Muraki (NEC) described the discourse processor of VENIEX . It combines fragments generated by th e parser into microelectronics-capability frames . The relationships among the resulting frames are determined b y resolving coreference for entities and capabilities . This is done by applying a set of heuristic rules that achieve good performance in the Japanese microelectronics texts . Reference resolution of a wide variety of expressions is treated , including anaphoric expressions, cleft sentences, and ellipsis .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Marc Vilain (MITRE) described how discourse processing in the Alembic system makes use of a propositiona l database, simplifying the tasks of reference resolution and event merging . The propositional database is a ne w component in Alembic and forms the substrate for semantic processing and data extraction . The equality system provided by this propositional substrate is utilized to seamlessly integrate facts from different places in the text whe n a reference is resolved . Another valuable task supported by a propositional substrate is inference . Instead of havin g complicated rules for determining the compatibility of event or template objects, and then specifying how the y should be merged, Alembic implements an inferential strategy for the merging process .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null } ], "back_matter": [], "bib_entries": {}, "ref_entries": {} } }