| { |
| "paper_id": "H05-1008", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T03:34:41.004959Z" |
| }, |
| "title": "Redundancy-based Correction of Automatically Extracted Facts", |
| "authors": [ |
| { |
| "first": "Roman", |
| "middle": [], |
| "last": "Yangarber", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "University of Helsinki", |
| "location": { |
| "country": "Finland" |
| } |
| }, |
| "email": "" |
| }, |
| { |
| "first": "Lauri", |
| "middle": [], |
| "last": "Jokipii", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "University of Helsinki", |
| "location": { |
| "country": "Finland" |
| } |
| }, |
| "email": "" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "The accuracy of event extraction is limited by a number of complicating factors, with errors compounded at all sages inside the Information Extraction pipeline. In this paper, we present methods for recovering automatically from errors committed in the pipeline processing. Recovery is achieved via post-processing facts aggregated over a large collection of documents, and suggesting corrections based on evidence external to the document. A further improvement is derived from propagating multiple, locally non-best slot fills through the pipeline. Evaluation shows that the global analysis is over 10 times more likely to suggest valid corrections to the local-only analysis than it is to suggest erroneous ones. This yields a substantial overall gain, with no supervised training.", |
| "pdf_parse": { |
| "paper_id": "H05-1008", |
| "_pdf_hash": "", |
| "abstract": [ |
| { |
| "text": "The accuracy of event extraction is limited by a number of complicating factors, with errors compounded at all sages inside the Information Extraction pipeline. In this paper, we present methods for recovering automatically from errors committed in the pipeline processing. Recovery is achieved via post-processing facts aggregated over a large collection of documents, and suggesting corrections based on evidence external to the document. A further improvement is derived from propagating multiple, locally non-best slot fills through the pipeline. Evaluation shows that the global analysis is over 10 times more likely to suggest valid corrections to the local-only analysis than it is to suggest erroneous ones. This yields a substantial overall gain, with no supervised training.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Abstract", |
| "sec_num": null |
| } |
| ], |
| "body_text": [ |
| { |
| "text": "Information Extraction (IE) is a technology for finding facts in plain text, and coding them in a logical representation, such as, e.g., a relational database. IE is typically viewed and implemented as a sequence of stages-a \"pipeline\": While accuracy at the lowest levels can reach high 90's, as the stages advance, complexity increases and performance degrades considerably.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "The problem of IE as a whole, as well each of the listed subproblems, has been studied intensively for well over a decade, in many flavors and varieties. Key observations about much state-of-the-art IE are: a. IE is typically performed by a pipeline process; b. Only one hypothesis is propagated through the pipeline for each fact-the \"best guess\" the system can make for each slot fill;", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "c. IE is performed in a document-by-document fashion, applying a priori knowledge locally to each document.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "The a priori knowledge may be encoded in a set of rules, an automatically trained model, or a hybrid thereof. Information extracted from documentswhich may be termed a posteriori knowledgeis usually not reused across document boundaries, because the extracted facts are imprecise, and are therefore not a reliable basis for future extraction. Furthermore, locally non-best slot fills are not propagated through the pipeline, and are consequently not available downstream, nor for any global analysis.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "In most systems, these stages are performed in sequence. The locally-best slot fills are passed from the \"lower-\" to the \"higher-level\" modules, without feedback. Improvements are usually sought (e.g., the ACE research programme, (ACE, 2004) ) by boosting performance at the lower levels, to reap benefits in the subsequent stages, where fewer errors are propagated.", |
| "cite_spans": [ |
| { |
| "start": 230, |
| "end": 241, |
| "text": "(ACE, 2004)", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "The point of departure for this paper is: the IE process is noisy and imprecise at the singledocument level; this has been the case for some time, and though there is much active research in the area, the situation is not likely to change radically in the immediate future-rather, we can expect slow, incremental improvements over some years.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "In our experiments, we approach the performance problem from the opposite end: start with the extracted results and see if the totality of a posteriori knowledge about the domain-knowledge generated by the same noisy process we are trying to improve-can help recover from errors that stem from locally insufficient a priori knowledge.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "The aim of the research presented in this paper is to improve performance by aggregating related facts, which were extracted from a large document collection, and to examine to what extent the correctly extracted facts can help correct those that were extracted erroneously.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "The rest of the paper is organized as follows. Section 2 contains a brief review of relevant prior work. Section 3 presents the experimental setup: the text corpus, the IE process, the extracted facts, and what aspects of the the extracted facts we try to improve in this paper. Section 4 presents the methods for improving the quality of the data using global analysis, starting with a naive, baseline method, and proceeding with several extensions. Each method is then evaluated, and the results are examined in section 5. In section 6, we present further extensions currently under research, followed by the conclusion.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "As we stated in the introduction, typical IE systems consist of modules arranged in a cascade, or a pipeline. The modules themselves are be based on heuristic rules or automatically trained, there is an abundance of approaches in both camps (and everywhere in between,) to each of the pipeline stages listed in the introduction.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Prior Work", |
| "sec_num": "2" |
| }, |
| { |
| "text": "It is our view that to improve performance we ought to depart from the traditional linear, pipelinestyle design. This view is shared by others in the research community; the potential benefits have previously been recognized in several contexts.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Prior Work", |
| "sec_num": "2" |
| }, |
| { |
| "text": "In (Nahm and Mooney, 2000a; Nahm and Mooney, 2000b) , it was shown that learning rules from a fact base, extracted from a corpus of job postings for computer programmers, improves future extraction, even though the originally extracted facts themselves are far from error-free. The idea is to mine the data base for association rules, and then to integrate these rules into the extraction process.", |
| "cite_spans": [ |
| { |
| "start": 3, |
| "end": 27, |
| "text": "(Nahm and Mooney, 2000a;", |
| "ref_id": "BIBREF7" |
| }, |
| { |
| "start": 28, |
| "end": 51, |
| "text": "Nahm and Mooney, 2000b)", |
| "ref_id": "BIBREF8" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Prior Work", |
| "sec_num": "2" |
| }, |
| { |
| "text": "The baseline system is obtained by supervised learning from a few hundred manually annotated examples. Then the IE system is applied to successively larger sets of unlabeled examples, and association rules are mined from the extracted facts. The resulting combined system (trained model plus association rules) showed an improvement in performance on a test set, which correlated with the size of the unlabeled corpus.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Prior Work", |
| "sec_num": "2" |
| }, |
| { |
| "text": "In work on improving (Chinese) named entity tagging, (Ji and Grishman, 2004; Ji and Grishman, 2005) , show benefits to this component from integrating decisions made in later stages, viz. coreference, and relation extraction. 1 Tighter coupling and integration between IE and KDD components for mutual benefit is advocated by (McCallum and Jensen, 2003) , which present models based on CRFs and supervised training.", |
| "cite_spans": [ |
| { |
| "start": 53, |
| "end": 76, |
| "text": "(Ji and Grishman, 2004;", |
| "ref_id": "BIBREF3" |
| }, |
| { |
| "start": 77, |
| "end": 99, |
| "text": "Ji and Grishman, 2005)", |
| "ref_id": "BIBREF4" |
| }, |
| { |
| "start": 226, |
| "end": 227, |
| "text": "1", |
| "ref_id": null |
| }, |
| { |
| "start": 326, |
| "end": 353, |
| "text": "(McCallum and Jensen, 2003)", |
| "ref_id": "BIBREF5" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Prior Work", |
| "sec_num": "2" |
| }, |
| { |
| "text": "This work is related in spirit to the work presented in this paper, in its focus on leveraging crossdocument information that information-though it is inherently noisy-to improve local decisions. We expect that the approach could be quite powerful when these ideas are used in combination, and our experiments seem to confirm this expectation.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Prior Work", |
| "sec_num": "2" |
| }, |
| { |
| "text": "In this section we describe the text corpus, the underlying IE process, the form of the extracted facts, and the specific problem under study-i.e., which aspects of these facts we first try to improve.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Experimental Setup", |
| "sec_num": "3" |
| }, |
| { |
| "text": "We conducted experiments with redundancy-based auto-correction over a large database of facts extracted from the texts in ProMED-Mail, a mailing list which carries reports about outbreaks of infectious epidemics around the world and the efforts to contain them. This domain has been explored earlier; see, e.g., (Grishman et al., 2003) for an overview.", |
| "cite_spans": [ |
| { |
| "start": 312, |
| "end": 335, |
| "text": "(Grishman et al., 2003)", |
| "ref_id": "BIBREF1" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Corpus", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "Our underlying IE system is described in (Yangarber et al., 2005) . The system is a hybrid automatically-and manually-built pattern base for finding facts, an HMM-based name tagger, automatically compiled and manually verified domainspecific ontology, based in part on MeSH, (MeS, 2004) , and a rule-based co-reference module, that uses the ontology.", |
| "cite_spans": [ |
| { |
| "start": 41, |
| "end": 65, |
| "text": "(Yangarber et al., 2005)", |
| "ref_id": "BIBREF9" |
| }, |
| { |
| "start": 275, |
| "end": 286, |
| "text": "(MeS, 2004)", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Corpus", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "The database is live on-line, and is continuously updated with new incoming reports; it can be accessed at doremi.cs.helsinki.fi/plus/.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Corpus", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "Text reports have been collected by ProMED-Mail for over 10 years. The quality of reporting (and editing) has been rising over time, which is easy to observe in the text data. The distribution of the data, aggregated by month is shown in Figure 1 , where one can see a steady increase in volume over time. 2", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 238, |
| "end": 246, |
| "text": "Figure 1", |
| "ref_id": null |
| } |
| ], |
| "eq_spans": [], |
| "section": "Corpus", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "We now describe the makeup of the data extracted from text by the IE process, with basic terminology.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Extracted Facts", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "Each document in the corpus, contains a single report, which may contain one or more stories. Story breaks are indicated by layout features, and are extracted by heuristic rules, tuned for this domain and corpus. When processing a multi-story report, the IE system treats each story as a separate document; no information is shared among stories, except that the text of the main headline of a multi-story report is available to each story. 3 Since outbreaks may be described in complex ways, it is not obvious how to represent a single fact in this context. To break down this problem, we use the notion of an incident. Each story may contain 1995 1996 1997 1998 1999 2000 2001 2002 2004 Stories (30,015) Documents (22, 560) Figure 1: Distribution of data in ProMED-Mail multiple outbreak-related incidents/facts. 4 We analyze an outbreak as a series of incidents. The incidents may give \"redundant\" information about an outbreak, e.g., by covering overlapping time intervals or geographic areas. For example, a report may first state the number of cases within the last month, and then give the total for the entire year. We treat each of these statements as a separate incident; the containment relations among them are beyond the scope of our current goals. 5 Thus each incident corresponds to a partial description of an outbreak, over a period of time and geographic area. This makes it easy to represent each incident/fact as a separate row in the table.", |
| "cite_spans": [ |
| { |
| "start": 441, |
| "end": 442, |
| "text": "3", |
| "ref_id": null |
| }, |
| { |
| "start": 644, |
| "end": 648, |
| "text": "1995", |
| "ref_id": null |
| }, |
| { |
| "start": 649, |
| "end": 653, |
| "text": "1996", |
| "ref_id": null |
| }, |
| { |
| "start": 654, |
| "end": 658, |
| "text": "1997", |
| "ref_id": null |
| }, |
| { |
| "start": 659, |
| "end": 663, |
| "text": "1998", |
| "ref_id": "BIBREF0" |
| }, |
| { |
| "start": 664, |
| "end": 668, |
| "text": "1999", |
| "ref_id": null |
| }, |
| { |
| "start": 669, |
| "end": 673, |
| "text": "2000", |
| "ref_id": null |
| }, |
| { |
| "start": 674, |
| "end": 678, |
| "text": "2001", |
| "ref_id": null |
| }, |
| { |
| "start": 679, |
| "end": 683, |
| "text": "2002", |
| "ref_id": "BIBREF2" |
| }, |
| { |
| "start": 684, |
| "end": 688, |
| "text": "2004", |
| "ref_id": "BIBREF3" |
| }, |
| { |
| "start": 716, |
| "end": 720, |
| "text": "(22,", |
| "ref_id": null |
| }, |
| { |
| "start": 721, |
| "end": 725, |
| "text": "560)", |
| "ref_id": null |
| }, |
| { |
| "start": 815, |
| "end": 816, |
| "text": "4", |
| "ref_id": null |
| }, |
| { |
| "start": 1262, |
| "end": 1263, |
| "text": "5", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Extracted Facts", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "The key fields of the incident table are: Disease Name Location Date (start and end) Where possible, the system also extracts information about the victims affected in the incident-their count, severity (affected or dead), and a descriptor (people, animals, etc.). The system also extracts bookkeeping information about each incident: locations of mentions of the key fields in the text, etc.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Extracted Facts", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "The system's performance is currently at 71.16 Fmeasure: 67% recall, 74% precision. This score is obtained by a MUC scorer (Douthat, 1998 ) on a 50document test corpus, which was manually tagged with correct incidents with these slots. We have no blind-test corpus at present, but prior experience suggests that we ought to expect about a 10% reduction in F-measure on unseen data; this is approximately borne out by our informal evaluations. Further, the system attempts to \"normalize\" the key fields. An alias for a disease name (e.g., \"bird flu\") is mapped to a canonical name (\"avian influenza.\") 6 Date expressions are normalized to a standard format yyyy.mm.dd-yyyy.mm.dd. 7 Note that the system may not be able to normalize some entities, which then remain un-normalized.", |
| "cite_spans": [ |
| { |
| "start": 123, |
| "end": 137, |
| "text": "(Douthat, 1998", |
| "ref_id": "BIBREF0" |
| }, |
| { |
| "start": 679, |
| "end": 680, |
| "text": "7", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Extracted Facts", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "Such normalization is clearly helpful for searching, but it is not only a user-interface issue. Normalizing reduces sparseness of data; and since our intent is to aggregate related facts across a large fact base, excessive variation in the database fields would reduce the effectiveness of the proposed methods.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Extracted Facts", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "A more complex problem arises out of the need to normalize location names. For each record, we normalize the location field-which may be a name of a small village or a larger area-by relating it to the name of the containing country; we also decided to map locations in the United States to the name of the containing state, (rather than the name of the country, \"USA\"). 8 This mapping will be henceforth referred to as \"location-state,\" for short. The ideas presented in the introduction are explored in the remainder of this paper in the context of correcting the location-state mapping. Section 6 will touch upon our current work on extending the methodology to slots other than state. (Please see Section 5 for further justification of this choice for our initial experiments.)", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Experimental Focus: Location Normalization", |
| "sec_num": "3.3" |
| }, |
| { |
| "text": "To make the experiments interesting and fair, we kept the size of the gazetteer small. The a priori geographic knowledge base contains names of countries of the world (270), with aliases for several of them; a list of capitals and other selected major cities (300); a list of states in the USA and acronyms (50); major US cities (100); names of the (sub)continents (10), and oceans. In our current implementation, continents are treated semantically as \"states\" as well. 9 The IE system operates in a local, document-bydocument fashion. Upon encountering a location name that is not in its dictionaries, the system has two ways to map it to the state name. One way is by matching patterns over the immediate local context, (\"Milan, Italy\"). Failing that, it tries to find the corresponding state by positing an \"underspecified\" state name (as if referred to by a kind of special \"pronoun\") and mapping the location name to that. The reference resolution module then finds the most likely antecedent entity, of the semantic type \"state/country,\" where likelihood is determined by its proximity to the mention of the location name.", |
| "cite_spans": [ |
| { |
| "start": 471, |
| "end": 472, |
| "text": "9", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Experimental Focus: Location Normalization", |
| "sec_num": "3.3" |
| }, |
| { |
| "text": "Note that the IE system outputs only a single, best hypothesis for the state fill for each record.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Experimental Focus: Location Normalization", |
| "sec_num": "3.3" |
| }, |
| { |
| "text": "The database currently contains about Fig. 1 ). Each incident has a location and a state filler. We say a location name is \"ambiguous\" if it appears in the location slot of at least two records, which have different names in the state slot. The number of distinct \"ambiguous\" location names is \u00a7 \u00a3 . Note, this terminology is a bit sloppy: the fillers to which we refer as \"ambiguous location names\", may not be valid location names at all; they may simply be errors in the IE process. E.g., at the name classification stage, a disease name (especially if not in the disease dictionary) may be misclassified, and used as a filler for the location slot.", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 38, |
| "end": 44, |
| "text": "Fig. 1", |
| "ref_id": null |
| } |
| ], |
| "eq_spans": [], |
| "section": "The Data", |
| "sec_num": "3.4" |
| }, |
| { |
| "text": "\u00a2 \u00a1 \u00a4 \u00a3 \u00a6 \u00a5 \u00a7 \u00a9 in- dividual facts/incidents, extracted from \u00a5 \u00a3 \u00a7 sto- ries, from \u00a3 \u00a6 \u00a1 reports (cf.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "The Data", |
| "sec_num": "3.4" |
| }, |
| { |
| "text": "We further group together the location fills by stripping lower-case words that are not part of the proper name, from the front and the end of the fill. E.g., we group together \"southern Mumbai\" and \"the Mumbai area,\" as referring to the same name.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "The Data", |
| "sec_num": "3.4" |
| }, |
| { |
| "text": "After grouping and trimming insignificant words, the number of distinct names appearing in location fills is \u00a1 , which covers a total of \u00a1 \u00a5 records, or \u00a7 ! # \" of all extracted facts. As an estimate of the potential for erroneous mapping from locations to states, this is quite high, about \u00a7 in \u00a9 records. 10", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "The Data", |
| "sec_num": "3.4" |
| }, |
| { |
| "text": "We now present the methods of correcting possible errors in the location-state relation. A method tries to suggest a new value for the state fill for every incident I that contains an ambiguous location fill:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Experiments and Results", |
| "sec_num": "4" |
| }, |
| { |
| "text": "\u00a1 \u00a3 \u00a2 \u00a5 \u00a4 \u00a7 \u00a6 \u00a9 \u00a2 \" ! $ # & % ( ' ) ! 0 1 2 3 $ 4 6 5 8 7 @ 9 \u00a6 B A D C F E G \u00a2 F H \u00a3 P I \u00a2 \u00a3 & Q (1) where R H Q", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Experiments and Results", |
| "sec_num": "4" |
| }, |
| { |
| "text": "is a set of all candidate states considered by for I;", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Experiments and Results", |
| "sec_num": "4" |
| }, |
| { |
| "text": "\u00a6 S A T C E G \u00a2H P I \u00a2 \u00a3 & Q", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Experiments and Results", |
| "sec_num": "4" |
| }, |
| { |
| "text": "is the scoring function specific to . The method chooses the candidate state which maximizes the score.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Experiments and Results", |
| "sec_num": "4" |
| }, |
| { |
| "text": "For each method below, we discuss how", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Experiments and Results", |
| "sec_num": "4" |
| }, |
| { |
| "text": "R H and \u00a6 B A D C E G \u00a2H", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Experiments and Results", |
| "sec_num": "4" |
| }, |
| { |
| "text": "are constructed.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Experiments and Results", |
| "sec_num": "4" |
| }, |
| { |
| "text": "We begin with a simple recovery approach. We simply assume that the correct state for an ambiguous location name is the state most frequently associated with it in the database. We denote by is one of the \u00a1 ambiguous location names, we define:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Baseline: Raw Majority", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "R e d Q g f F I Q h p i r q r D h s V U u t X \u00a3 Y I Q h v \u1e81 x D h y \u00a6 B A D C E \u00a2 d P I Q h\u00a3 & Q S t f Q D h s V U u t X \u00a3 Y I Q h v \u1e81 x D h y t i.e.,", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Baseline: Raw Majority", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "is a candidate if it is a state fill in some incident whose location fill is also X ; the score is the number of times the pair X", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Ih", |
| "sec_num": null |
| }, |
| { |
| "text": "\u00a3 Y I h v", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Ih", |
| "sec_num": null |
| }, |
| { |
| "text": "appear together in some incident in", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Ih", |
| "sec_num": null |
| }, |
| { |
| "text": ". The majority winner is then suggested as the \"correct\" state, for every record containing X . By \"majority\" winner we mean the candidate with the maximal count, rather than a state with more than half of the votes. When the candidates tie for first place, no suggestions are madealthough it is quite likely that some of the records carrying X w ill have incorrect state fills. A manual evaluation of the performance of this method is shown in Table 1 , the Baseline column.", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 445, |
| "end": 452, |
| "text": "Table 1", |
| "ref_id": "TABREF0" |
| } |
| ], |
| "eq_spans": [], |
| "section": "U", |
| "sec_num": null |
| }, |
| { |
| "text": "The first row shows for how many records this method suggested a change from the original, IEfilled state. The baseline changed 858 incidents. sistently always maps some location name to the same wrong state; these cases are below the radar of our scheme, in which the starting point is the \"ambiguous\" locations.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "U", |
| "sec_num": null |
| }, |
| { |
| "text": "This constitutes about 13% out of the maximum number of changeable records, \u00a1 \u00a5 .", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "U", |
| "sec_num": null |
| }, |
| { |
| "text": "Thus, this number represents the volume of the potential gain or loss from the global analysis: the proportion of records that actually get modified.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "U", |
| "sec_num": null |
| }, |
| { |
| "text": "The remaining records were unchanged, either because the majority winner coincides with the original IE-extracted state, or because there was a tie for the top score, so no decision could be made.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "U", |
| "sec_num": null |
| }, |
| { |
| "text": "We manually verified a substantial sample of the modified records. When verifying the changes, we referred back to the text of the incident, and, when necessary, consulted further geographical sources to determine exactly whether the change was correct in each case.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "U", |
| "sec_num": null |
| }, |
| { |
| "text": "For the baseline we had manually verified 27.7% of the changes. Of these, 68.5% were a clear gain: an incorrect state was changed to a correct state. 6.3% were a clear loss, a correct state lost to an incorrect one. This produces quite a high baseline, surprisingly difficult to beat.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "U", |
| "sec_num": null |
| }, |
| { |
| "text": "The next two rows represent the \"grey\" areas. These are records which were difficult to judge, for one of two technical reasons. A: the \"location\" name was itself erroneous, in which case these redundancy-based approaches are not meaningful; or, B: the suggestion replaces an area by its subregion or super-region, e.g., changing \"Connecticut\" to \"USA\", or \"Europe\" to \"France.\" 11", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "U", |
| "sec_num": null |
| }, |
| { |
| "text": "Although it is not strictly meaningful to judge whether these changes constitute a gain or a loss, we nonetheless tried to assess whether changing the state hurt the accuracy of the incident, since the incident may have a correct state even though its location is erroneous (case A); likewise, it may be correct to say that a given location is indeed a part of Connecticut, in which case changing it to USA loses information, and is a kind of loss.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "U", |
| "sec_num": null |
| }, |
| { |
| "text": "That is the interpretation of the grey gain and loss instances. The final row, no loss, indicates the proportion of cases where an originally incorrect state name was changed to a new one, also incorrect.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "U", |
| "sec_num": null |
| }, |
| { |
| "text": "DB-filtered Confidence Multi-candidate ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Baseline", |
| "sec_num": null |
| }, |
| { |
| "text": "Changed \u00a7 \u00a5 \u00a4 ! \" \u00a1 \u00a1 \u00a5 \u00a4 !\u00a9 \" # \u00a9 \u00a9 \u00a2 \u00a1 \u00a5 \u00a3 \u00a4 !\u00a9 \" \u00a1 \u00a1 \u00a5 \u00a4 \u00a6 \u00a5 ! \u00a7 \" \u00a7 \u00a2 \u00a9 \u00a1 \u00a5 Verified \u00a9 !\u00a9 \" \u00a5 \u00a1 \u00a5 \u00a4 !\u00a1 \" \u00a5 \u00a1 # \u00a9 \u00a9 \u00a5 # \u00a9 ! \u00a7 \" \u00a5 \u00a1 \u00a1 \u00a1 \u00a4 ! \" \u00a6 \u00a4 \u00a7 \u00a2 \u00a9 Gain \u00a1 \u00a4 ! \" \u00a7 \u00a1 \u00a5 \u00a1 \u00a5 \u00a9 \u00a4 \u00a7 !\u00a5 \" \u00a7 \u00a2 \u00a3 \u00a1 \u00a5 !\u00a5 \" \u00a7 \u00a9 \u00a3 \u00a7 \u00a5 \u00a5 ! \" \u00a7 \u00a9 \u00a2 Loss \u00a1 \u00a4 !\u00a5 \" \u00a7 \u00a1 \u00a5 ! # \" \u00a1 \u00a5 \u00a7 !\u00a5 \" \u00a5 \u00a1 \u00a5 ! \" \u00a7 Grey gain \u00a7 ! \u00a3 \" \u00a1 \u00a1 \u00a5 \u00a7 !\u00a1 \" \u00a1 \u00a5 \u00a7 \u00a7 ! \" \u00a1 \u00a5 \u00a7 \u00a5 \u00a4 !\u00a9 \" \u00a5 \u00a2 \u00a3 \u00a1 Grey loss \u00a1 \u00a4 !\u00a9 \" \u00a7 \u00a1 \u00a1 \u00a5 \u00a4 ! \" \u00a7 \u00a5 ! \" \u00a5 ! \u00a7 \" \u00a1 \u00a1 No loss \u00a9 !\u00a1 \" \u00a7 \u00a1 \u00a5 \u00a4 ! \" \u00a7 \u00a9 \u00a3 \u00a1 \u00a5 \u00a1 \u00a4 !\u00a9 \" \u00a7 \u00a1 \u00a1 \u00a5 ! # \" \u00a7", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Baseline", |
| "sec_num": null |
| }, |
| { |
| "text": "Next we examined a variant of baseline raw majority vote, noting that simply choosing the state most frequently associated with a location name is a bit naive: the location-state relation is not functionali.e., some location names map to more than one state in reality. There are many locations which share the same name. 12 To approach this more intelligently, we define:", |
| "cite_spans": [ |
| { |
| "start": 322, |
| "end": 324, |
| "text": "12", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Database Filtering", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "R Q R e d Q ! \u00a6 \u00a9 \u00a8 \u00a2 F I \" # \u00a6 \u00a9C E % $ Q \u00a6 B A D C F E G \u00a2 P Ih\u00a3 & Q S \u00a6 B A D C E G \u00a2 d P Ih\u00a3 & Q", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Database Filtering", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "The baseline vote counting across the data base (DB) produced a ranked list of candidate states Ih for the location", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Database Filtering", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": ". We then filtered this list through", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "X", |
| "sec_num": null |
| }, |
| { |
| "text": "\u00a6 \u00a9 \u00a2 Q I & \" \u00a1 # \u00a6 BC E % $ Q", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "X", |
| "sec_num": null |
| }, |
| { |
| "text": ", the list of states mentioned in the story containing the incident . The filtered majority winner was selected as the suggested change.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "X", |
| "sec_num": null |
| }, |
| { |
| "text": "For example, the name \"Athens\" may refer to the city in Greece, or to the city in Georgia (USA). Suppose that Greece is the raw majority winner. The baseline method will always tag all instances of Athens as being in Greece. However, in a story about Georgia, Greece will likely not be mentioned at all, so it is safe to rule it out. This helps a minority winner, when the majority is not present in the story.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "X", |
| "sec_num": null |
| }, |
| { |
| "text": "Surprisingly, this method did not yield a substantial improvement over the baseline, (though it was more careful by changing fewer records). This may indicate that NWP is not an important source of errors here: though many truly ambiguous locations do exist, they do not account for many instances in this DB.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "X", |
| "sec_num": null |
| }, |
| { |
| "text": "A more clear improvement over the baseline is obtained by taking the local confidence of the statelocation association into account. For each record, we extend the IE analysis to produce a confidence value for the state. Confidence is computed by simple, document-local heuristics, as follows:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Confidence-Based Ranking", |
| "sec_num": "4.3" |
| }, |
| { |
| "text": "If the location and state are both within the span of text covered by the incident-text which was actually matched by a rule in the IE system,-or if the state is the unique state mentioned in the story, it gets a score of 2-the incident has high confidence in the state. Otherwise, if the state is outside the incident's span, but is inside the same sentence as the incident, and is also the unique state mentioned in that sentence, it gets a score of 1. Otherwise it receives a score of zero.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Confidence-Based Ranking", |
| "sec_num": "4.3" |
| }, |
| { |
| "text": "Given the confidence score for each (location X , state I ) pair, the majority counting is based on the cumulative confidence,", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Confidence-Based Ranking", |
| "sec_num": "4.3" |
| }, |
| { |
| "text": "A D C ' # ) ( 1 1 0 3 2 4 0 3 5 X \u00a3 Y I", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Confidence-Based Ranking", |
| "sec_num": "4.3" |
| }, |
| { |
| "text": "in the DB, rather than on the cumulative count of occurrences of this pair in the DB:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Confidence-Based Ranking", |
| "sec_num": "4.3" |
| }, |
| { |
| "text": "R ) 6 Q R Q \u00a6 B A D C E \u00a2 6 P Ih\u00a3 & Q S 7 7 9 8 2 & @ B A5 D C F E1 1 89 H G 7 9 8 A D C ' # ) ( 1 1 0 3 2 4 0 3 5 h", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Confidence-Based Ranking", |
| "sec_num": "4.3" |
| }, |
| { |
| "text": "Filtering through the story is also applied, as in the previous method. The resulting method favors more correct decisions, and fewer erroneous ones.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Confidence-Based Ranking", |
| "sec_num": "4.3" |
| }, |
| { |
| "text": "We should note here, that the notion of confidence of a fill (here, the state fill) is naturally extended to the notion of confidence of a record: For each of the three key fills-location, date, disease namecompute a confidence based on the same heuristics. Then we say that a record has high confidence, if it has non-zero confidence in all three of the key fills. The notion of record confidence is used in Section 6.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Confidence-Based Ranking", |
| "sec_num": "4.3" |
| }, |
| { |
| "text": "Finally, we tried propagating multiple candidate state hypotheses for each instance of an ambiguous location name", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Multi-Candidate Propagation", |
| "sec_num": "4.4" |
| }, |
| { |
| "text": "X : R \u00a1 Q S \u00a2 7 9 8 2 & @ ACG 7 9 8 \u00a6 \u00a9 \u00a8 \u00a2 F I \" # \u00a6 \u00a9C E % $ h \u00a6 B A D C E G \u00a2 P Ih\u00a3 & Q 7 7 9 8 2 & @ B AC 3 G 7 9 8 \u00a4 \u00a3 E C \u00a6 \u00a5 B P Ih\u00a3 & h", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Multi-Candidate Propagation", |
| "sec_num": "4.4" |
| }, |
| { |
| "text": "where the proximity is inversely proportional to the distance of", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Multi-Candidate Propagation", |
| "sec_num": "4.4" |
| }, |
| { |
| "text": "I Q h from incident \u00a5 h", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Multi-Candidate Propagation", |
| "sec_num": "4.4" |
| }, |
| { |
| "text": ", in the story of The resulting performance is substantially better than the baseline, while the number of changed records is substantially higher than in the competing methods. This is due to the fact that this method allows for a much larger pool of candidates than the others, and assigns to them much smoother weights, virtually eliminating ties in the ranking among hypotheses.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Multi-Candidate Propagation", |
| "sec_num": "4.4" |
| }, |
| { |
| "text": "h : \u00a3 E $ C \u00a7 \u00a5 P I \u00a2 \u00a3 & Q \u00a9 \u00a9 \u00a7 a Q \u00a7 P I \u00a2 \u00a3 & Q \u00a7 ( W \u00cc b C \u00a2 \u00a5 E \u00a4 I \u00a2", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Multi-Candidate Propagation", |
| "sec_num": "4.4" |
| }, |
| { |
| "text": "Among the four competing approaches presented above, the baseline performs surprisingly well. We should note that this research is not aimed specifically at improving geographic Named Entity resolution. It is the first in a series of experiments aiming to leverage redundancy across a large fact base extracted from text, to improve the quality of extracted data. We chose to experiment with this relation first because of its simplicity, and because the state field is a key field in our application.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Discussion", |
| "sec_num": "5" |
| }, |
| { |
| "text": "For this reason, the a priori geographic knowledge base was intentionally not as extensive as it might have been, had we tried in earnest to match locations with corresponding states (e.g., by incorporating the CIA Factbook, or other gazetteer).", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Discussion", |
| "sec_num": "5" |
| }, |
| { |
| "text": "The intent here is to investigate how a relation can be improved by leveraging redundancy across a large body of records. The support we used for geographic name resolution was therefore deliberately modest, cf. Section 3.3.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Discussion", |
| "sec_num": "5" |
| }, |
| { |
| "text": "It is quite feasible to enumerate the countries and the larger regions, since they number in the low hundreds, whereas there are many tens of thousands of cities, towns, villages, regions, districts, etc.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Discussion", |
| "sec_num": "5" |
| }, |
| { |
| "text": "Three parallel lines of current research are:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Current Work", |
| "sec_num": "6" |
| }, |
| { |
| "text": "1. combining evidence from multiple features 2. applying redundancy-based correction to other fields in the database 3. back-propagation of corrected results, to repair components that induced incorrect information.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Current Work", |
| "sec_num": "6" |
| }, |
| { |
| "text": "The results so far presented show that even a naive, intuitive approach can help correct local errors via global analysis. We are currently working on more complex extensions of these methods.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Current Work", |
| "sec_num": "6" |
| }, |
| { |
| "text": "Each method exploits one main feature of the underlying data: the distance from candidate state to the mention of the location name. In the multicandidate hypothesis method, this distance is exploited explicitly in the scoring function. In the other methods, it is used inside the co-reference module of the IE pipeline, to find the (single) locally-best state.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Current Work", |
| "sec_num": "6" |
| }, |
| { |
| "text": "However, other textual features of the state candidate should contribute to establishing the relations to a location mention, besides the raw distance. For example, at a given distance, it is very important whether the state is mentioned before the location (more likely to be related) vs. after the location (less likely). Another important feature: is the state mentioned in the main story/report headline? If so, its score should be raised. It is quite common for documents to declaim the focal state only once in the headline, and never mention it again, instead mentioning other states, neighboring, or otherwise relevant to the story. The distance measure used alone may be insufficient in such cases.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Current Work", |
| "sec_num": "6" |
| }, |
| { |
| "text": "How are these features to be combined? One path is to use some combination of features, such as a weighted sum, with parameters trained on a manually tagged data set. As we already have a reasonably sized set tagged for evaluation, we can split it into two, train the parameter on a larger portion, evaluate on a smaller one, and cross-validate.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Current Work", |
| "sec_num": "6" |
| }, |
| { |
| "text": "We will be using this approach as a baseline. However, we aim to use a much larger set of data to train the parameters, without manually tagging large training sets.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Current Work", |
| "sec_num": "6" |
| }, |
| { |
| "text": "The idea is to treat the set of incidents with high record confidence, Sec. 4.3, rather than manually tagged data, as ground truth. Again, there \"confident\" truth will not be completely error-free, but because error rates are lower among the confident records, we may be able to leverage global analysis to produce the desired effect: training parameters for more complex models-involving multiple features-for global re-ranking of decisions.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Current Work", |
| "sec_num": "6" |
| }, |
| { |
| "text": "Our approach rests on the idea that evidence aggregated across documents should help resolve difficult problems at the level of a given document.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Conclusion", |
| "sec_num": null |
| }, |
| { |
| "text": "Our experiments confirm that aggregating global information about related facts, and propagating locally non-best analyses through the pipeline, provide powerful sources of additional evidence, which are able to reverse incorrect decisions, based only on local and a priori information.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Conclusion", |
| "sec_num": null |
| }, |
| { |
| "text": "The proposed approach requires no supervision or training of any kind. It does, however require a substantial collection of evidence across a large body of extracted records; this approach needs a \"critical mass\" of data to be effective. Although large volume of facts is usually not reported in classic IE experiments, obtaining high volume should be natural in principle.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Conclusion", |
| "sec_num": null |
| }, |
| { |
| "text": "Performance on English named entity tasks reaches mid to high 90's in many domains.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "This is beneficial to the IE process, which operates better with formulaic, well-edited text.3 The format of the documents in the archive can be examined by browsing the source site www.promedmail.org.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "In this paper, we use the terms fact, incident, and event interchangeably.5 This problem is addressed in, e.g.,(Huttunen et al., 2002).", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "This is done by means of a set of scenario-specific patterns and a dictionary of about 2500 disease names with aliases.7 Some date intervals may not have a starting date, e.g., if the text states \"As of last Tuesday, the victim count is N...\"8 This decision was made because otherwise records with state = USA strongly skew the data, and complicate learning.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "By the same token, both Connecticut and USA are \"states.\" 10 Of course, it can be higher as well, if the IE system con-", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "Note, that for some locations, which are not within any one state's boundary, a continent is a \"correct state\", for example, \"the Amazon Region,\" or \"Serengeti.\"", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "We refer to this as the \"New-World phenomenon\" (NWP), due to its prevalence in the Americas: \"Santa Cruz\" occurs in several Latin American countries; locations named after saints are common. In the USA, city and county names often appear in multiple states-Winnebago County, Springfield; many cities are named after older European cities.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [ |
| { |
| "text": "We'd like to thank Winston Lin of New York University, who worked on an early version of these experiments.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Acknowledgement", |
| "sec_num": null |
| } |
| ], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "The Message Understanding Conference scoring software user's manual", |
| "authors": [ |
| { |
| "first": "A", |
| "middle": [], |
| "last": "Douthat", |
| "suffix": "" |
| } |
| ], |
| "year": 1998, |
| "venue": "Proc. 7th Message Understanding Conf. (MUC-7)", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "A. Douthat. 1998. The Message Understanding Con- ference scoring software user's manual. In Proc. 7th Message Understanding Conf. (MUC-7), Fairfax, VA.", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "Information extraction for enhanced access to disease outbreak reports", |
| "authors": [ |
| { |
| "first": "R", |
| "middle": [], |
| "last": "Grishman", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Huttunen", |
| "suffix": "" |
| }, |
| { |
| "first": "R", |
| "middle": [], |
| "last": "Yangarber", |
| "suffix": "" |
| } |
| ], |
| "year": 2003, |
| "venue": "J. of Biomed. Informatics", |
| "volume": "", |
| "issue": "4", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "R. Grishman, S. Huttunen, and R. Yangarber. 2003. In- formation extraction for enhanced access to disease outbreak reports. J. of Biomed. Informatics, 35(4).", |
| "links": null |
| }, |
| "BIBREF2": { |
| "ref_id": "b2", |
| "title": "Complexity of event structure in information extraction", |
| "authors": [ |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Huttunen", |
| "suffix": "" |
| }, |
| { |
| "first": "R", |
| "middle": [], |
| "last": "Yangarber", |
| "suffix": "" |
| }, |
| { |
| "first": "R", |
| "middle": [], |
| "last": "Grishman", |
| "suffix": "" |
| } |
| ], |
| "year": 2002, |
| "venue": "Proc. 19th Intl. Conf. Computational Linguistics (COLING 2002)", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "S. Huttunen, R. Yangarber, and R. Grishman. 2002. Complexity of event structure in information extrac- tion. In Proc. 19th Intl. Conf. Computational Linguis- tics (COLING 2002), Taipei.", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "Applying coreference to improve name recognition", |
| "authors": [ |
| { |
| "first": "H", |
| "middle": [], |
| "last": "Ji", |
| "suffix": "" |
| }, |
| { |
| "first": "R", |
| "middle": [], |
| "last": "Grishman", |
| "suffix": "" |
| } |
| ], |
| "year": 2004, |
| "venue": "Proc. Reference Resolution Wkshop", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "H. Ji and R. Grishman. 2004. Applying coreference to improve name recognition. In Proc. Reference Reso- lution Wkshop, (ACL-2004), Barcelona, Spain.", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "Improving name tagging by reference resolution and relation detection", |
| "authors": [ |
| { |
| "first": "H", |
| "middle": [], |
| "last": "Ji", |
| "suffix": "" |
| }, |
| { |
| "first": "R", |
| "middle": [], |
| "last": "Grishman", |
| "suffix": "" |
| } |
| ], |
| "year": 2005, |
| "venue": "Proc. ACL-2005", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "H. Ji and R. Grishman. 2005. Improving name tagging by reference resolution and relation detection. In Proc. ACL-2005, Amherst, Mass.", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "A note on the unification of information extraction and data mining using conditional-probability, relational models", |
| "authors": [ |
| { |
| "first": "A", |
| "middle": [], |
| "last": "Mccallum", |
| "suffix": "" |
| }, |
| { |
| "first": "D", |
| "middle": [], |
| "last": "Jensen", |
| "suffix": "" |
| } |
| ], |
| "year": 2003, |
| "venue": "IJ-CAI'03 Workshop on Learning Statistical Models from Relational Data", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "A. McCallum and D. Jensen. 2003. A note on the uni- fication of information extraction and data mining us- ing conditional-probability, relational models. In IJ- CAI'03 Workshop on Learning Statistical Models from Relational Data.", |
| "links": null |
| }, |
| "BIBREF6": { |
| "ref_id": "b6", |
| "title": "Medical subject headings", |
| "authors": [], |
| "year": null, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Medical subject headings.", |
| "links": null |
| }, |
| "BIBREF7": { |
| "ref_id": "b7", |
| "title": "A mutually beneficial integration of data mining and information extraction", |
| "authors": [ |
| { |
| "first": "U", |
| "middle": [ |
| "Y" |
| ], |
| "last": "Nahm", |
| "suffix": "" |
| }, |
| { |
| "first": "R", |
| "middle": [], |
| "last": "Mooney", |
| "suffix": "" |
| } |
| ], |
| "year": 2000, |
| "venue": "AAAI-2000", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "U. Y. Nahm and R. Mooney. 2000a. A mutually benefi- cial integration of data mining and information extrac- tion. In AAAI-2000, Austin, TX.", |
| "links": null |
| }, |
| "BIBREF8": { |
| "ref_id": "b8", |
| "title": "Using information extraction to aid the discovery of prediction rules from text", |
| "authors": [ |
| { |
| "first": "U", |
| "middle": [ |
| "Y" |
| ], |
| "last": "Nahm", |
| "suffix": "" |
| }, |
| { |
| "first": "R", |
| "middle": [], |
| "last": "Mooney", |
| "suffix": "" |
| } |
| ], |
| "year": 2000, |
| "venue": "KDD-2000 Text Mining Wkshop", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "U. Y. Nahm and R. Mooney. 2000b. Using information extraction to aid the discovery of prediction rules from text. In KDD-2000 Text Mining Wkshop, Boston, MA.", |
| "links": null |
| }, |
| "BIBREF9": { |
| "ref_id": "b9", |
| "title": "Extracting information about outbreaks of infectious epidemics", |
| "authors": [ |
| { |
| "first": "R", |
| "middle": [], |
| "last": "Yangarber", |
| "suffix": "" |
| }, |
| { |
| "first": "L", |
| "middle": [], |
| "last": "Jokipii", |
| "suffix": "" |
| }, |
| { |
| "first": "A", |
| "middle": [], |
| "last": "Rauramo", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Huttunen", |
| "suffix": "" |
| } |
| ], |
| "year": 2005, |
| "venue": "Proc. HLT-EMNLP 2005", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "R. Yangarber, L. Jokipii, A. Rauramo, and S. Hut- tunen. 2005. Extracting information about outbreaks of infectious epidemics. In Proc. HLT-EMNLP 2005", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "FIGREF0": { |
| "text": "1. Layout, tokenization, lexical analysis 2. Name recognition and classification 3. Shallow (commonly,) syntactic parsing 4. Resolution of co-reference among entities 5. Pattern-based event matching and role mapping 6. Normalization and output generation", |
| "type_str": "figure", |
| "num": null, |
| "uris": null |
| }, |
| "FIGREF2": { |
| "text": "i.e., are extracted as fills in I. In the baseline method,", |
| "type_str": "figure", |
| "num": null, |
| "uris": null |
| }, |
| "FIGREF3": { |
| "text": "giving a full point to a single, locally-best guess among the I 's, this point is shared proportionately among all competingI", |
| "type_str": "figure", |
| "num": null, |
| "uris": null |
| }, |
| "TABREF0": { |
| "type_str": "table", |
| "content": "<table/>", |
| "num": null, |
| "html": null, |
| "text": "Performance of Correction Methods" |
| } |
| } |
| } |
| } |
