{ "paper_id": "M92-1039", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T03:13:24.369809Z" }, "title": "Description of the LINK System Used for MUC-4", "authors": [ { "first": "Steven", "middle": [ "L" ], "last": "Lytinen", "suffix": "", "affiliation": {}, "email": "" }, { "first": "Sayan", "middle": [], "last": "Bhattacharyya", "suffix": "", "affiliation": {}, "email": "" }, { "first": "Robert", "middle": [ "R" ], "last": "Burridge", "suffix": "", "affiliation": {}, "email": "" }, { "first": "Peter", "middle": [ "M" ], "last": "Hastings", "suffix": "", "affiliation": {}, "email": "" }, { "first": "Christian", "middle": [], "last": "Huyck", "suffix": "", "affiliation": {}, "email": "" }, { "first": "Karen", "middle": [ "A" ], "last": "Lipinsky", "suffix": "", "affiliation": {}, "email": "" }, { "first": "Eric", "middle": [ "S" ], "last": "Mcdaniel", "suffix": "", "affiliation": {}, "email": "" }, { "first": "Karenann", "middle": [ "K" ], "last": "Terrell", "suffix": "", "affiliation": {}, "email": "" } ], "year": "", "venue": null, "identifiers": {}, "abstract": "", "pdf_parse": { "paper_id": "M92-1039", "_pdf_hash": "", "abstract": [], "body_text": [ { "text": "The University of Michigan's natural language processing system, called LINK, is a unificationbased system which we have developed over the last four years . Prior to MUC-4, LINK ha d been used to extract information from free-form texts in two narrow application domains . One application corpus contained terse descriptions of symptoms displayed by malfunctioning automobiles, and the repairs which fixed them . The other corpus described sequences of activities to be performed on an assembly line . In empirical testing in these two domains, LINK correctly processed 70% of previously unseen descriptions . A template was counted as correct only if all of the fillers in the template were filled correctly . In addition, LINK generated incomplete (but not incorrect) templates for another 15% of the descriptions .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Background", "sec_num": null }, { "text": "These previous domains were much narrower than the MUC-4 terrorism domain . As a comparison, the lexicons for the previous domains contained only 300-500 words, compared wit h 6700 words in our MUC-4 test configuration . Previous grammar size ranged from 75-100 rules , compared with over 500 rules in the MUC-4 knowledge base . In addition, the previous application domains consisted only of single-sentence inputs . Thus, the integration of information from multiple sentences was not an issue in our previous work .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Background", "sec_num": null }, { "text": "The MUC-4 LINK system consists of the modules shown in figure 1 . One sentence at a time passes through the modules in the order shown in the figure . Each module's function i s described below . To help explain the role of each module, its performance on various parts o f message TST2-MUC4-0048 is shown .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Flow of contro l", "sec_num": null }, { "text": "The tokenizer produces LISP-readable files from a 100-article source file . It also performs a few simple editing tasks, such as separating the text into sentences, and removing text that i s in brackets .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "The tokenizer", "sec_num": null }, { "text": "The filter determines which sentences in an article should be passed to the remainder of th e system for processing . While we originally had in mind more sophisticated filtering techniques , the filter in the test configuration simply passed on any sentences containing one or more word s whose definitions were deemed interesting . Interesting definitions included any word meanin g any of the template actions (BOMBING, ATTACK, . . .) as well as a few other concepts likely t o appear as template fillers, such as EXPLOSIVE and HOSTAGE . Any sentence which did not contain any words whose definitions were deemed interesting were discarded and not processe d further.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "The filter", "sec_num": null }, { "text": "The preprocessor is responsible for initializing the environment in which the LINK parse r operates . Since LINK is a bottom-up chart parser, this means that the preprocessor must initialize the chart . The initialization is constructed by looking up each word in the sentence , and adding a link into the chart corresponding to each possible sense of a word . In addition, likely noun phrases are identified and grouped together, and NP links are entered into the char t for these groups of words .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "The preprocesso r", "sec_num": null }, { "text": "Noun phrases are preidentified for two reasons . First, undefined words often appear as parts of noun phrases. The grouping of these unidentified words eliminates the need to deal with the m in the parser itself . Second, preprocessing noun phrases enabled us to encode parsing heuristic s in the preprocessor which could not easily be encoded in the parser itself, such as preferring th e longest possible noun phrase . This improves the efficiency of the system .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "The preprocesso r", "sec_num": null }, { "text": "Each link in the initial chart contains both syntactic and semantic information about a word or noun phrase . For a single word, this semantic information is simply copied from th e definition of the word . For noun phrases, semantic information on a link is the result of unifying, or merging together, semantic information from all of the defined words in the noun phrase . Adjacent nouns whose definitions cannot unify are not grouped together into a single noun phrase by the preprocessor . For example, \"government headquarters\" is not initially grouped as a single NP, since the meanings of \"government\" and \"headquarters\" cannot be unified . Thus , it might be more accurate to say that the preprocessor identifies \"noun clusters \" rather than noun phrases .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "The preprocesso r", "sec_num": null }, { "text": "If all of the words of an NP are undefined, then a default semantic definition is assigned . For the test configuration of the system, the default definition was HUMAN-OR-PLACE, a definition which could be refined during processing to be any of the set fills for the HUM TGT , PHYS TGT, or LOCATION fields .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "The preprocesso r", "sec_num": null }, { "text": "The preprocessor is also responsible for identifying names of people . A list of names that appeared in the HUM TGT : NAME fields of the MUC-3 development answer keys is used to identify names, along with a few simple heuristics for identifying likely additional names . Fo r example, any undefined word ending in a `z ' is considered a potential name .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "The preprocesso r", "sec_num": null }, { "text": "Here are the results produced by the preprocessor for the first sentence in article TST2 -MUC4-0048 . The initial chart is displayed, with potential noun phrases already grouped together : The LINK parser LINK is a bottom-up, unification-based chart parser . Its grammar rules are quite similar i n form to those used in PATR-II (Shieber, 1986) . We have incorporated semantic information int o LINK's grammar, along the lines of HPSG (Pollard and Sag, 1987) . The integration of syntacti c and semantic knowledge into the same grammar formalism is crucial to our system's ability t o process large texts in a reasonable length of time, and to producing the semantic analysis use d to generate templates .", "cite_spans": [ { "start": 329, "end": 344, "text": "(Shieber, 1986)", "ref_id": "BIBREF1" }, { "start": 435, "end": 458, "text": "(Pollard and Sag, 1987)", "ref_id": "BIBREF0" } ], "ref_spans": [], "eq_spans": [], "section": "The preprocesso r", "sec_num": null }, { "text": "Here is a simplified example of a constraint rule : Each equation in this rule specifies a property which . any node labeled S must have . A property consists of a path, or a sequence of arcs with the appropriate labels starting from the node in question ; and a value, which is another node to be found at the end of the path. Equations specify the values of properties in one of two ways . They may specify the label of th e node to be found at the end of the path, as in equations 1 and 2 (i .e., the arc from an S node labeled 1 leads to a node labeled NP) . We will call these labeling equations. Or, they may specify that two paths must lead to the identical node, as in equations 3-5 . Identity here is defined by the unification operation ; i.e, if two paths must lead to the identical node, then the nodes a t the end of the two paths must unify. Unification merges the properties of two nodes ; thus, two paths can unify if their values have no properties which explicitly contradict each other . These equations will be called unifying equations.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Next sentence (1) : SALVADORAN PRESIDENT-ELECT ALFREDO CRISTIANI CONDEMNE D THE TERRORIST KILLING OF ATTORNEY GENERAL ROBERTO GARCIA ALVARADO AND ACCUSED THE FARABUNDO MARTI NATIONAL LIBERATION FRONT OF THE CRIME", "sec_num": null }, { "text": "Links are placed in the chart to represent potential constituents that the parser identifies . These links contain both syntactic and semantic information, represented in the form of a directed acyclic graph (DAG) . The DAGs correspond to the information in the set of gramma r rules used to build a constituent .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Next sentence (1) : SALVADORAN PRESIDENT-ELECT ALFREDO CRISTIANI CONDEMNE D THE TERRORIST KILLING OF ATTORNEY GENERAL ROBERTO GARCIA ALVARADO AND ACCUSED THE FARABUNDO MARTI NATIONAL LIBERATION FRONT OF THE CRIME", "sec_num": null }, { "text": "The core of the grammar is a set of domain-independent rules that handle all regular ver b tenses, and many of the simple english constructions . The rules encode both syntactic an d semantic constraints, which allows much of the work of finding the actor, object, location, etc . to be done during the parse . This rule handles all constructions of the form \"It has been said that [sentence]\" or \"It wa s reported by the government today that [sentence],\" etc .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Next sentence (1) : SALVADORAN PRESIDENT-ELECT ALFREDO CRISTIANI CONDEMNE D THE TERRORIST KILLING OF ATTORNEY GENERAL ROBERTO GARCIA ALVARADO AND ACCUSED THE FARABUNDO MARTI NATIONAL LIBERATION FRONT OF THE CRIME", "sec_num": null }, { "text": "Although the preprocessor is responsible for finding simple noun phrases, the correct interpretation of complex NPs relies on semantics and is handled by a set of grammar rules for NPs . This information is not merged with the template generated from sentence 1 because of th e mismatch between the OBJECTs of the two representations .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Next sentence (1) : SALVADORAN PRESIDENT-ELECT ALFREDO CRISTIANI CONDEMNE D THE TERRORIST KILLING OF ATTORNEY GENERAL ROBERTO GARCIA ALVARADO AND ACCUSED THE FARABUNDO MARTI NATIONAL LIBERATION FRONT OF THE CRIME", "sec_num": null }, { "text": "Sentence 22 illustrates the inability of our system to perform reference resolution : Because the referent of `them' cannot be resolved, it is dropped from the representation of th e sentence, and the result is that no information is added to the response templates for this article .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Next sentence (1) : SALVADORAN PRESIDENT-ELECT ALFREDO CRISTIANI CONDEMNE D THE TERRORIST KILLING OF ATTORNEY GENERAL ROBERTO GARCIA ALVARADO AND ACCUSED THE FARABUNDO MARTI NATIONAL LIBERATION FRONT OF THE CRIME", "sec_num": null }, { "text": "Sentence", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Next sentence (1) : SALVADORAN PRESIDENT-ELECT ALFREDO CRISTIANI CONDEMNE D THE TERRORIST KILLING OF ATTORNEY GENERAL ROBERTO GARCIA ALVARADO AND ACCUSED THE FARABUNDO MARTI NATIONAL LIBERATION FRONT OF THE CRIME", "sec_num": null } ], "back_matter": [ { "text": "These include past particples used as adjectives (e .g., \"the kidnapped priests\"); noun phrase complements (e .g., \"Noriega, the president of Panama\") and some noun-noun constructions (e.g., \"government headquarters\" or \"FMLN terrorists\") .If a sentence fails to parse completely, the chart can be inspected to see what constituent s have been constructed, and what their semantic content is . Thus, after a failed parse, the system examined the chart, identifying those links which contained information relevant to the construction of templates . Links which contained the most relevant information (i .e., th e greatest number of slots filled which could map to template fields) were selected and passed t o the postprocessor for incorporation into templates .An example parse of sentence 1 from article TST2-MUC4-0048 is shown below . ", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "annex", "sec_num": null }, { "text": "The LINK parser utilizes semantic/domain knowledge during processing . This information is organized in an inheritance hierarchy. Figure 2 presents the actions from the hierarchy used i n the MUC-4 domain, along with constraints on fillers of slots for actions . Slot-filling constraints on a concept may either be defined for that concept or inherited from the concept's ancestor s in the tree . For example, since ATTACK requires an OBJECT that is a HUMAN-OR-PLACE , this restriction also implicitly holds for actions like SHOOT and ROBBERY . KIDNAPPIN G is an example of a concept which makes a further restriction on a previously constrained slot . HUMAN-OR-OFFICIAL, the OBJECT of this action, must be a descendant of HUMAN-OR-PLACE.", "cite_spans": [], "ref_spans": [ { "start": 130, "end": 138, "text": "Figure 2", "ref_id": null } ], "eq_spans": [], "section": "The Inheritance Hierarchy", "sec_num": null }, { "text": "The postprocessor receives semantic representations from the LINK parser for each sentenc e in an article, and is responsible for producing response templates . It first checks to see if the representation of a sentence can be added to an existing template, or if it requires a ne w template . This decision is based on the compatibility of several template fields : the DATE , LOCATION, INCIDENT CATEGORY, and INDIVIDUAL ID fields . If all of these fields ar e compatible, then additional information is added to an existing template ; otherwise, a new template is constructed .To illustrate the strengths and weaknesses of the postprocessor, we will examine the processing of article TST2-MUC4-0048. Several sentences in this article generated templates . Here are the results produced by the LINK parser for sentences 1 and 2 : The templates for these sentences are merged, since the OBJECT of both actions appear t o be the same person .Later in the article, the following sentence appears :Sentence 11 : GUERRILLAS ATTACKED MERINO'S HOME IN SAN SALVADOR 5 DAYS AG O WITH EXPLOSIVES", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "The Postprocessor", "sec_num": null } ], "bib_entries": { "BIBREF0": { "ref_id": "b0", "title": "Information-based Syntax and Semantics", "authors": [ { "first": "C", "middle": [], "last": "Pollard", "suffix": "" }, { "first": "I", "middle": [], "last": "Sag", "suffix": "" } ], "year": 1987, "venue": "Center for the Study of Language and Information", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Pollard, C ., and Sag ., I . (1987) . Information-based Syntax and Semantics. Menlo Park, CA : Center for the Study of Language and Information .", "links": null }, "BIBREF1": { "ref_id": "b1", "title": "An Introduction to Unification-based Approaches to Grammar", "authors": [ { "first": "S", "middle": [], "last": "Shieber", "suffix": "" } ], "year": 1986, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Shieber, S . (1986) . An Introduction to Unification-based Approaches to Grammar . CSLI, Stan- ford CA .", "links": null } }, "ref_entries": { "FIGREF0": { "type_str": "figure", "text": "Modules of the MUC-4 LINK syste m", "uris": null, "num": null }, "FIGREF3": { "type_str": "figure", "text": "The core is augmented by a set of rules that handle common constructions from the 130 0 MUC-3 development articles . Typical examples of this are \"Meanwhile, [sentence] .\" or \" ...an attack on [place] ...\", and are handled in as general a rule as possible provided the correct semantics may be given the parent based on the semantics of the children . An example of a grammar rule for a specific type of construction is shown below .", "uris": null, "num": null } } } }