| { |
| "paper_id": "R13-1044", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T14:55:25.708879Z" |
| }, |
| "title": "Semantic relation recognition within Polish noun phrase: A rule-based approach", |
| "authors": [ |
| { |
| "first": "Pawe\u0142", |
| "middle": [], |
| "last": "K\u0119dzia", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "University of Technology", |
| "location": {} |
| }, |
| "email": "pawel.kedzia@pwr.wroc.pl" |
| }, |
| { |
| "first": "Marek", |
| "middle": [], |
| "last": "Maziarz", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "University of Technology", |
| "location": {} |
| }, |
| "email": "marek.maziarz@pwr.wroc.pl" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "The paper 1 presents a rule-based approach to semantic relation recognition within the Polish noun phrase. A set of semantic relations, including some thematic relations, has been determined for the need of experiments. The method consists in two steps: first the system recognizes word pairs and triples, and then it classifies the relations. Evaluation was performed on random samples from two balanced Polish corpora.", |
| "pdf_parse": { |
| "paper_id": "R13-1044", |
| "_pdf_hash": "", |
| "abstract": [ |
| { |
| "text": "The paper 1 presents a rule-based approach to semantic relation recognition within the Polish noun phrase. A set of semantic relations, including some thematic relations, has been determined for the need of experiments. The method consists in two steps: first the system recognizes word pairs and triples, and then it classifies the relations. Evaluation was performed on random samples from two balanced Polish corpora.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Abstract", |
| "sec_num": null |
| } |
| ], |
| "body_text": [ |
| { |
| "text": "Semantic relation recognition is a well-known task in natural language processing. Although the relation recognition within noun phrase and between nominals was studied intensely, the task is still challenge for semantic analysis of Polish. We are aware of few papers and projects dealing with Semantic Role Labelling between predicates and their arguments, cf. (Go\u0142uchowski and Przepi\u00f3rkowski, 2012) or (Lun, 2009) , but of none concerning semantic relation recognition inside Polish noun phrase.", |
| "cite_spans": [ |
| { |
| "start": 362, |
| "end": 400, |
| "text": "(Go\u0142uchowski and Przepi\u00f3rkowski, 2012)", |
| "ref_id": "BIBREF8" |
| }, |
| { |
| "start": 404, |
| "end": 415, |
| "text": "(Lun, 2009)", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "In (Nastase et al., 2006) authors classify semantic relations between a head and a modifier of a noun phrase. Number of all relation types was equal to 30. These relations were grouped into 5 more general groups. The authors experimented with decision trees, instance-based learning and Support Vector Machines. For each relation they learnt the binary classifier; as the baseline for F-measure they used the model with all of examples classified as positive and recall being equal to 100%. With regard to the semantic relation the baseline ranged between 17.78% and 60.35%.", |
| "cite_spans": [ |
| { |
| "start": 3, |
| "end": 25, |
| "text": "(Nastase et al., 2006)", |
| "ref_id": "BIBREF20" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Related work", |
| "sec_num": "2" |
| }, |
| { |
| "text": "Identifying the semantic relations inside compound nouns was presented in (Uchiyama et al., 2008) . The authors used SVM classifier and in the best configuration of features, they achieved accuracy of about 84%.", |
| "cite_spans": [ |
| { |
| "start": 74, |
| "end": 97, |
| "text": "(Uchiyama et al., 2008)", |
| "ref_id": "BIBREF32" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Related work", |
| "sec_num": "2" |
| }, |
| { |
| "text": "In (Rosario and Hearst, 2001 ) authors used neural networks to determine 20 semantic relationssimilarily to (Nastase et al., 2006) -between a head and a modifier of noun phrase. They used a domain-specific lexical hierarchy of medicine. The authors achieved accuracy of about 60%.", |
| "cite_spans": [ |
| { |
| "start": 3, |
| "end": 28, |
| "text": "(Rosario and Hearst, 2001", |
| "ref_id": "BIBREF29" |
| }, |
| { |
| "start": 108, |
| "end": 130, |
| "text": "(Nastase et al., 2006)", |
| "ref_id": "BIBREF20" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Related work", |
| "sec_num": "2" |
| }, |
| { |
| "text": "The workshop SemEval-2010 (task 8) concerned the recognition of semantic relations between nominals. In (Tratz and Hovy, 2010) the authors developed a system based on the Maximum Entropy classifier, able to detect 10 bidirectional semantic relations Achieved F-measures depended on the system configuration and lay between 66, 68% and 77, 75%. The same set of semantic relations was used in (Rink and Harabagiu, 2010) . The authors used Support Vector Machines classifier and a very rich set of features (i.e., part of speech for all constituents of a semantic relation pair, number of words between the nominals, features based on paths in the dependency tree from Stanford dependency parser). F-measure of this approach was 82.19%.", |
| "cite_spans": [ |
| { |
| "start": 104, |
| "end": 126, |
| "text": "(Tratz and Hovy, 2010)", |
| "ref_id": "BIBREF30" |
| }, |
| { |
| "start": 391, |
| "end": 417, |
| "text": "(Rink and Harabagiu, 2010)", |
| "ref_id": "BIBREF28" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Related work", |
| "sec_num": "2" |
| }, |
| { |
| "text": "Authors in (Tymoshenko and Giuliano, 2010 ) used shallow syntactic parsing and semantic information from ResearchCyc (Lenat, 1995) in the same task of recognizing semantic relations. They used liner combination of kernels (semantic and syntactic) using Support Vector Machines classifier. For the best combination of kernels, they obtained F-measure equal to 77.62%.", |
| "cite_spans": [ |
| { |
| "start": 11, |
| "end": 41, |
| "text": "(Tymoshenko and Giuliano, 2010", |
| "ref_id": "BIBREF31" |
| }, |
| { |
| "start": 105, |
| "end": 130, |
| "text": "ResearchCyc (Lenat, 1995)", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Related work", |
| "sec_num": "2" |
| }, |
| { |
| "text": "There are some works, where rule-based approaches were used. In (Huang, 2009) there has been proposed an approach for automatic construction of rules identifying ten types of seman-tic relations, using five types of input informations. The relation instances were extracted from Modern Chinese Standard Dictionary. The authors achieved very high precision (range from 0, 81 to 0.99), but recall was low -about 0, 2. In (Hearst, 1992) authors used set of manually written rules for identification of hyperonymy relations. (Ben Abacha and Zweigenbaum, 2011) used linguistic patterns (built semi-automatically from corpora) to identify semantic relatios in medical texts. In this domain-specific task they achieved 75.72% precision and 60, 46% recall.", |
| "cite_spans": [ |
| { |
| "start": 64, |
| "end": 77, |
| "text": "(Huang, 2009)", |
| "ref_id": "BIBREF10" |
| }, |
| { |
| "start": 419, |
| "end": 433, |
| "text": "(Hearst, 1992)", |
| "ref_id": "BIBREF9" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Related work", |
| "sec_num": "2" |
| }, |
| { |
| "text": "We seek for semantic relations within nominal phrases. The relation set consists of 12 semantic relations, of which 5 are thematic (semantic) roles 2 . Definitions of our semantic relations are based on works of (Kearns, 2011) , (Palmer et al., 2010) , (Van Valin, 2004) , (Larson, 1996) , (Dowty, 1991) , (J\u0119drzejko, 1993) , (Laskowski and Wr\u00f3bel, 1997) . We tried to select relations that are very frequent or frequent in Polish texts. 3 The relation set is following (thematic roles are marked with theta, other relations -with rho):", |
| "cite_spans": [ |
| { |
| "start": 212, |
| "end": 226, |
| "text": "(Kearns, 2011)", |
| "ref_id": null |
| }, |
| { |
| "start": 229, |
| "end": 250, |
| "text": "(Palmer et al., 2010)", |
| "ref_id": "BIBREF21" |
| }, |
| { |
| "start": 253, |
| "end": 270, |
| "text": "(Van Valin, 2004)", |
| "ref_id": "BIBREF33" |
| }, |
| { |
| "start": 273, |
| "end": 287, |
| "text": "(Larson, 1996)", |
| "ref_id": "BIBREF14" |
| }, |
| { |
| "start": 290, |
| "end": 303, |
| "text": "(Dowty, 1991)", |
| "ref_id": "BIBREF4" |
| }, |
| { |
| "start": 306, |
| "end": 323, |
| "text": "(J\u0119drzejko, 1993)", |
| "ref_id": "BIBREF11" |
| }, |
| { |
| "start": 326, |
| "end": 354, |
| "text": "(Laskowski and Wr\u00f3bel, 1997)", |
| "ref_id": null |
| }, |
| { |
| "start": 438, |
| "end": 439, |
| "text": "3", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognized semantic relation types", |
| "sec_num": "3" |
| }, |
| { |
| "text": "Proto-Agent \u03b8 -it is an instigator of an action or an entity that is in a particular state, it may undergoe change of state not caused by another participant; for predicates denoting relations -it is the first element of the relation: (cz\u0142owiek) wykszta\u0142cony przez Jana \u03b8 '(man) educated by John \u03b8 ', wyj\u0105cy wilk \u03b8 'howling wolf \u03b8 '. The Proto-Agent macrorole covers subroles of Agent, Causer and nonagentive non-causative Actor (cf. Actor macrorole in (Kearns, 2011) ).", |
| "cite_spans": [ |
| { |
| "start": 453, |
| "end": 467, |
| "text": "(Kearns, 2011)", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognized semantic relation types", |
| "sec_num": "3" |
| }, |
| { |
| "text": "Proto-Patient \u03b8 is the second macrorole -it is an entity undergoing action, event or change of state caused by another participant; for predicates denoting relations -it is the second element of a given relation: wykszta\u0142cenie kogo\u015b \u03b8 'educating someone \u03b8 ', (Jan) posiadaj\u0105cy maj\u0105tek \u03b8 '(John) possessing an estate \u03b8 '. According to (Dowty, 1991) 2 In Polish, as in other Indo-European languages, verbs could be nominalized during a process of syntactic transformation (J\u0119drzejko, 1993) , (Kolln, 1990) . Such nominalized predicates could be linked with nouns by thematic relations.", |
| "cite_spans": [ |
| { |
| "start": 334, |
| "end": 347, |
| "text": "(Dowty, 1991)", |
| "ref_id": "BIBREF4" |
| }, |
| { |
| "start": 470, |
| "end": 487, |
| "text": "(J\u0119drzejko, 1993)", |
| "ref_id": "BIBREF11" |
| }, |
| { |
| "start": 490, |
| "end": 503, |
| "text": "(Kolln, 1990)", |
| "ref_id": "BIBREF13" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognized semantic relation types", |
| "sec_num": "3" |
| }, |
| { |
| "text": "3 Rationale for selection of the presented semantic relation types was their frequencies in a four-text sample taken from a Polish corpus KPWr. Together chosen relations account for ca 80% of all semantic relation occurrences in these texts. Most of our relation types could be found on the list of the most frequent relation types in the English noun phrase (Moldovan et al., 2004, Tab. 1 ). many thematic roles come down to the macroroles of Proto-Patient and Proto-Agent.", |
| "cite_spans": [ |
| { |
| "start": 359, |
| "end": 389, |
| "text": "(Moldovan et al., 2004, Tab. 1", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognized semantic relation types", |
| "sec_num": "3" |
| }, |
| { |
| "text": "Instrument \u03b8 is a tool, a device or means used by someone in order to cause something, it is sometimes regarded as a secondary cause of situation or change of state: przeszyty w\u0142\u00f3czni\u0105 'speared with a spear', lina \u03b8 cumownicza adjective 'a hawser, lit.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognized semantic relation types", |
| "sec_num": "3" |
| }, |
| { |
| "text": "mooring rope \u03b8 '.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognized semantic relation types", |
| "sec_num": "3" |
| }, |
| { |
| "text": "Material \u03b8 is an entity that is used by someone to produce something from it, material undergoes change of state resulting in its disappearance and emerging of a result: zrobiony z mosi\u0105dzu \u03b8 'made out of brass \u03b8 ', mosi\u0119\u017cna \u03b8 figurka 'brass \u03b8 statuette'.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognized semantic relation types", |
| "sec_num": "3" |
| }, |
| { |
| "text": "Purpose \u03b8 -an entity or a situation toward which the event is directed or an individual which benefits from the event (purpose combines goal, beneficiary and recipient roles): wr\u0119czenie (medali) olimpijczykom 'giving (medals) to Olympians \u03b8 , sala koncertowa \u03b8 'a concert \u03b8 hall'.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognized semantic relation types", |
| "sec_num": "3" |
| }, |
| { |
| "text": "Location is a physical place at which a given event is localised, a place being destination of an event, a path or a source of motion, or simply a place at which a particular individual is situated: wr\u0119czenie (medali) w auli 'giving (medals) at the lecture theatre ', przedzieranie si\u0119 przez moczary 'struggling through the swamp '.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognized semantic relation types", |
| "sec_num": "3" |
| }, |
| { |
| "text": "Time is a particular moment or a duration of an event -it localises a situation within the flow of events or gives its duration: przedzieranie si\u0119 przez godzin\u0119 /w\u015brod\u0119 'struggling for an hour /on Wednesday '.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognized semantic relation types", |
| "sec_num": "3" |
| }, |
| { |
| "text": "Temporal/spatial meronymy -these relations point onto a spatial or temporal part of a place/location/time/period): poniedzia\u0142kowy poranek 'Monday morning ',\u015brodek zimy 'middle of the winter', koniec drogi 'end of the road', stolica kraju 'capital of the country'.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognized semantic relation types", |
| "sec_num": "3" |
| }, |
| { |
| "text": "Attribute is a property of an individual or an event, such as colour, size, weigth, intensity, duration etc., which might be expressed with a qualitative adjective: czerwony samoch\u00f3d 'red car', g\u0142o\u015bna muzyka 'loud music'.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognized semantic relation types", |
| "sec_num": "3" |
| }, |
| { |
| "text": "Family (member) is a relative or an in-law to someone, the relation is bidirectional and reflexive: syn kr\u00f3la 'king's son ', moja \u017cona 'my wife ' (I am a relative to my wife).", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognized semantic relation types", |
| "sec_num": "3" |
| }, |
| { |
| "text": "Order gives a position of an entity or an event in an ordered sequence/chain: druga odpowied\u017a '2nd answer', lata 80 . 'eighties, lit. eightieth years'.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognized semantic relation types", |
| "sec_num": "3" |
| }, |
| { |
| "text": "Quantity is an amount of something or a cardinality of a given set: pi\u0119ciu pan\u00f3w 'five men', kieliszek wina 'glass of wine'.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognized semantic relation types", |
| "sec_num": "3" |
| }, |
| { |
| "text": "Our rule-based system proceeds in two steps 4 : first it recognizes word pairs and triples, then operators classifying relations enter.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Semantic relation recognition rule-based algorithm", |
| "sec_num": "4" |
| }, |
| { |
| "text": "Since we consider relations within noun phrases, we must identify them correctly. We made use of a CRF shallow parser (Radziszewski and Pawlaczek, 2012) trained on an annotated corpus of Polish (KPWr) (Broda et al., 2012) which comprises shallow syntactic annotation level . KPWr contains 326 annotated text samples representing different genres and styles: blogs, press articles, official and legal texts and Polish Wikipedia articles, it comprises 106358 annotations (phrases and phrase heads, and predicateargument relations).", |
| "cite_spans": [ |
| { |
| "start": 118, |
| "end": 152, |
| "text": "(Radziszewski and Pawlaczek, 2012)", |
| "ref_id": "BIBREF24" |
| }, |
| { |
| "start": 201, |
| "end": 221, |
| "text": "(Broda et al., 2012)", |
| "ref_id": "BIBREF1" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognizing word pairs and triples", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "Noun and preposition phrases (NPs/PPs) from the corpus correspond to arguments of predicateargument structure. Each such NP/PP constists of one or several smaller phrases based on agreement (AgPs, for details, please look at cited works).", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognizing word pairs and triples", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "Here is an example NP from the corpus (a head of the phrase is boldfaced, AgP heads are underlined):", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognizing word pairs and triples", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "[[samolot wyprodukowany] AgP [przez PZL] AgP [w roku 1938] AgP [w \u0141odzi] AgP ] N P", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognizing word pairs and triples", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "'aircraft made by PZL in (year) 1938 in \u0141\u00f3d\u017a (city)'", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognizing word pairs and triples", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "There is no reliable deep parser for Polish (Go\u0142uchowski and Przepi\u00f3rkowski, 2012 ), thus we decided to construct a simple rule-based algorithm for deepened shallow parsing of Polish NPs/PPs. The algorithm works on tagged texts -we used (Radziszewski, 2013) tagger. Parsing rules make use of an output from the CRF shallow parser (Radziszewski and Pawlaczek, 2012) , in particular: borders of whole NPs/PPs, and of their constituents (i.e., phrases based on agreement, AgPs). Found pairs and triples are directly connected within a syntactic structure.", |
| "cite_spans": [ |
| { |
| "start": 44, |
| "end": 81, |
| "text": "(Go\u0142uchowski and Przepi\u00f3rkowski, 2012", |
| "ref_id": "BIBREF8" |
| }, |
| { |
| "start": 237, |
| "end": 257, |
| "text": "(Radziszewski, 2013)", |
| "ref_id": "BIBREF27" |
| }, |
| { |
| "start": 330, |
| "end": 364, |
| "text": "(Radziszewski and Pawlaczek, 2012)", |
| "ref_id": "BIBREF24" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognizing word pairs and triples", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "Hand-written rules act like a partial dependency parser. The pairs consist of one subordinate and one superordinate token, the triples comprise one superordinate token and a subordinate preposition phrase (preposition + governed nominal head of a subordinate noun phrase).", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognizing word pairs and triples", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "The whole algorithm runs in a main loop which iterates AgP i heads. We start from the first AgP 0 head to the left, then we proceed to the right, jumping from AgP i head to the closest AgP i+1 head to the right. For every AgP i head we run a cascadelike chain of rules (numbered from 1 to 7) for genetives, nominatives, small preposition phrases (being a part of larger NPs or PPs), coordination, other known to the tagger tokens, other unknown to the tagger tokens and for modifiers. The algorithm in pseudocode was shown in Algorithm 1", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognizing word pairs and triples", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "The algorithm gives following description for just analysed phrase, \"R + number\" denotes the number of a rule in the Algorithm 1 activated on the word pair or triple (for instance, R3 means that the rule number 3 was activated): R7: samolot \u2190 wyprodukowany 'plane made', R3: wyprodukowany \u2190 przez PZL 'by PZL',R3: wyprodukowany \u2190 w roku 'in year', R3: wyprodukowany \u2190 w \u0141odzi 'in \u0141\u00f3d\u017a' .", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognizing word pairs and triples", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "Such simple shallow parsing algorithm operates quite well on an annotated part of KPWr with Fmeasure equal to 84%, P = 88%, R = 80%. 5", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Recognizing word pairs and triples", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "Having identified pairs and triples we run on them operators written in a constraint language WCCL . The operators are language-specific and utilize morphosyntactic features (POS, case, number and gender), domains of Polish WordNet lexical units (word-sense pairs ), thousands of derivational relation instances between nouns, adjectives and verbs from the wordnet 6 and information about syntactic frames of nominalized predicates, taken from Polish valence dictionary (D\u0119bowski and Woli\u0144ski, 2007) .", |
| "cite_spans": [ |
| { |
| "start": 470, |
| "end": 499, |
| "text": "(D\u0119bowski and Woli\u0144ski, 2007)", |
| "ref_id": "BIBREF5" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Applying WCCL operators", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "Each of written operators refers to one semantic relation. In other words, each semantic relation is described by one or by many WCCL operators. If an operator is successfully applied to a pair (or a Algorithm 1 Rule-based algorithm for the recognition of word pairs and triples 1. genetive attachment -link AgP i head in genetive to the closest AgP i\u22121 head to the left or to the closest nominalized predicate to the left:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Applying WCCL operators", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "\u2022 if there is none -link it to the closest predicate to the right; \u2022 if there is none -link the considered AgP i head to the head of the whole NP/PP; 2. nominative attachment -link AgP i head in nominative to the closest AgP i\u22121 head to the left or to the closest nominalized predicate to the left:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Applying WCCL operators", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "\u2022 if there is none -link it to the closest AgP i+1 head to the right or to the closest nominalized predicate to the right; \u2022 if there is none -link the considered AgP i head to the head of the whole NP/PP; 3. small PP attachment -link a head of AgP i containing a small PP to the closest nominalized predicate to the left:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Applying WCCL operators", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "\u2022 if there is none -to the closest nominalized predicate to the right such that it is not an element of AgP j>i containing a preposition; \u2022 if there is none -to the closest AgP i\u22121 head to the left; \u2022 if there is none -link AgP i with our whole NP/PP head; 4. coordinated syntactic groups -look for such AgP i that is preceded by a coordination conjunction (i.e., i 'and', oraz 'and', lub 'or') or by coordinating comma ('coordinating comma' is such a comma that is placed between two AgPs whose heads are agreed on case), such coordination marker cannot be an element of any AgP:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Applying WCCL operators", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "\u2022 if there is such a marker, look to the left in order to find such AgP j<i head which is agreed on case with our AgP i head -then create a new relation instance by copying the link AgP j \u2192 X and replacing AgP j head by the AgP i head in that copied linkage, i.e., create the relation instance AgP i \u2192 X; \u2022 if it is not possible -do not introduce any relation;", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Applying WCCL operators", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "5. head token provided with POS known to the CRF tagger -link the AgP i head to the closest nominalized predicate to the left:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Applying WCCL operators", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "\u2022 if there is none -to the closest nominalized predicate to the right such that it is not an element of AgP j>i containing a preposition; \u2022 if there is none -link AgP i head to the closest AgP j<i head to the left such that AgP j<i does not contain any preposition; \u2022 if there is none such AgP j<i -connect AgP i to the whole NP/PP head; 6. other cases (the AgP i head was not provided any known POS by the CRF tagger) -in such cases link AgP i head to the closest AgP j<i head to the left; if there is none -do not make any decision; 7. relations inside AgPs -link adjectival and participial modifiers to the head of AgP i .", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Applying WCCL operators", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "triple), then we know what semantic relation between the pair (or triple) occurs. Otherwise, we assume that the semantic relation does not occur. For example, our Proto-Patient relation was described by the 6 WCCL operators. One of them is presented in Listing 1. This operator uses two dictionaries with valence frames (acc -a list of verbs possessing any accusative frame, frames -a list of verbs described in the Polish valence dictionary (D\u0119bowski, 2013) ) and morphosyntactic information about part of speech (class) and case.", |
| "cite_spans": [ |
| { |
| "start": 442, |
| "end": 458, |
| "text": "(D\u0119bowski, 2013)", |
| "ref_id": "BIBREF6" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Applying WCCL operators", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "This operator PROTO-PATIENT-acc captures pairs like dr\u0119cz\u0105cy pact Janka noun.acc\u2212\u03b8 'tormenting John \u03b8 ' with a noun playing a Proto-Patient role of the predicate dr\u0119cz\u0105cy. The operator first checks whether a predicate (active participle) has an accusative frame or is outside the dictionary of D\u0119bowski (\"frames\"). Since dr\u0119czy\u0107 'to torment' is in acc dictionary and since Janek 'John' has subst class and acc case -the boolean operator returns 'true'.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Applying WCCL operators", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "Let us present another example: the Proto-Agent macrorole is recognized by 5 operators, in Listing 2 was shown one of them. The PROTO-AGENT-ger-przez-acc operator is written for triples, i.e., for a triple wydanie pact przez pron wydawc\u0119 noun.acc\u2212\u03b8 'publishing by the publisher \u03b8 '. The first element in the triple is a gerund form of verb wyda\u0107 'to publish'. The operator checks whether the verb wyda\u0107 has in its frame accusative/genetive or whether it cannot be found in D\u0119bowski's dictionary (position 0 in the triple, frames).", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Applying WCCL operators", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "Listing 1: One of the WCCL operators describing Proto-Patient relation. Language details has been described in , abbreviations for grammatical categories has been explained in (Przepi\u00f3rkowski et al., 2012) Next the operator seeks for the preposition przez 'by' at position 1. Then it tests if the first meaning of the lemma wydawca 'publisher' does not belong to the domain 'time' (= Polish czas) in Polish WordNet (position 2). Indeed, the first meaning of wydawca is in the domain 'person' (that iformation is avaiable in the dictionary noun_domain). At the end, we check whether the last token of our triple is in accusative. Because all of these conditions are fulfilled, the operator returns 'true', and we may assume that the last token takes the role of Proto-Agent. In Listing 3 one operator for family ralation was shown. FAMILY-agpp used to recognize this relation for word pairs. The operator, inter alia, uses semantic dictionary of kinship names built on the basis of Polish WordNet (the dictionary kinship), lammas of possessive pronouns (e.g., m\u00f3j 'my', tw\u00f3j 'yours').", |
| "cite_spans": [ |
| { |
| "start": 176, |
| "end": 205, |
| "text": "(Przepi\u00f3rkowski et al., 2012)", |
| "ref_id": "BIBREF8" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Applying WCCL operators", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "Listing 3: Two WCCL operators describing Family relation ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Applying WCCL operators", |
| "sec_num": "4.2" |
| }, |
| { |
| "text": "Evaluation of the presented semantic relation recognition algorithm was performed in three steps. First experiment (labelled kpwr) was performed on a random sample of the KPWr corpus (26 out of 326 texts, aproximately one thirteenth of the corpus). In this experiment we made use of syntactic annotations from KPWr (cf. Tab. 1). Second experiment was performed on a random sample of 100 texts taken from yet another Polish corpus, called NKJP (Przepi\u00f3rkowski et al., 2012 , nkjp, approximately one tenth of the corpus) 7 . Since NKJP lacked syntactic annotations of KPWr style, we were forced to run on it the CRF shallow parser (described in Sec. 4.1). This experiment gave us information about performance of our algorithm on a 'bare' text (see Tab. 2). Evaluation in the experiments was done by a professional linguist.", |
| "cite_spans": [ |
| { |
| "start": 443, |
| "end": 471, |
| "text": "(Przepi\u00f3rkowski et al., 2012", |
| "ref_id": "BIBREF8" |
| } |
| ], |
| "ref_spans": [ |
| { |
| "start": 747, |
| "end": 751, |
| "text": "Tab.", |
| "ref_id": null |
| } |
| ], |
| "eq_spans": [], |
| "section": "Results and conclusions", |
| "sec_num": "5" |
| }, |
| { |
| "text": "At last, four baseline models were constructed and evaluated on the two corpora (Tab. 3). We created baselines similar to that presented in (Uchiyama et al., 2008) , which was majority model. We chose the most frequent relation, which in the sample from KPWr was Proto-Patient (with the number of 113 instances out of 268 relation instances), this relation type was also the most frequent in the sample of NKJP (411 out of 1950 relation instances). For each corpora two baselines were calculated: in Baseline #1 we assumed that we had perfectly recognized all occurences of semantic relations (of any type), in Baseline #2 we simply signed with 'Proto-Patient' label every recognized by our system semantic relation instance. Baseline #2 is realistic, while #1 is idealistic, since to obtain #1 we should be able to recognize every single relation instance within a corpus. Baselines #1 are upper limits for all majority models (including #2). Our two idealistic baselines are higher than the realistic baselines (see Tab. 3). Percentile bootstrap methods (DiCiccio and Efron, 1996) , (DiCiccio and Romano, 1988) were applied to statistical significance and confidence interval (CI) analysis of the data. 8 We took 10000 8 Our data for NKJP were merged, so cross-validation was experiments (kpwr, nkjp). Asterisks denote significant differences between an experiment and a baseline in one-tailed test at \u03b1 = 0.05 bootstrap resamplings for each measure (P, R, F1), \u03b1 was equal to 0.05 for each one-tailed test and CI (a percentile CI need not be symmetrical).", |
| "cite_spans": [ |
| { |
| "start": 140, |
| "end": 163, |
| "text": "(Uchiyama et al., 2008)", |
| "ref_id": "BIBREF32" |
| }, |
| { |
| "start": 1056, |
| "end": 1082, |
| "text": "(DiCiccio and Efron, 1996)", |
| "ref_id": "BIBREF2" |
| }, |
| { |
| "start": 1085, |
| "end": 1112, |
| "text": "(DiCiccio and Romano, 1988)", |
| "ref_id": "BIBREF3" |
| }, |
| { |
| "start": 1205, |
| "end": 1206, |
| "text": "8", |
| "ref_id": null |
| }, |
| { |
| "start": 1221, |
| "end": 1222, |
| "text": "8", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [ |
| { |
| "start": 1018, |
| "end": 1022, |
| "text": "Tab.", |
| "ref_id": null |
| } |
| ], |
| "eq_spans": [], |
| "section": "Results and conclusions", |
| "sec_num": "5" |
| }, |
| { |
| "text": "In nkjp we have beaten both idealistic and realistic baselines. Precision, recall and F1 for kpwr are higher than Baseline #2. Only idealistic Baseline #1 for the KPWr corpus has overtaken our rule-based algorithm with regard to recall (42.2% vs. 32.8%), while its precision is lower and F1's are statistically indistinguishable.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Results and conclusions", |
| "sec_num": "5" |
| }, |
| { |
| "text": "Results are promising, precisions go above 50% (the lower endpoint for the kpwr confidence intervel), for nkjp we may assess it even more precisely as 60%-65%. Some semantic relations are recognized with higher precision: Proto-Agent (nkjp: 89-100%, kpwr: 90-100%, \u03b1 = 0.05), Proto-Patient (nkjp: 88-95%, kpwr: 83%-98%), family (nkjp: 90-100%) and order (nkjp: 91-100%). Our system is thus comparable in this aspect to the systems described in Sec. 2. 9 Overall recall is low, but higher than realistic baselines. In kpwr we obtained R = 27-38%, while for nkjp we got statistically higher interval of 37-41%. It seems that recall was not affected by lack of marked NP/PP borders in the corpus (these should have been brought out by the CRF shallow parser). F-measures calculated on our both corpora are also much higher than realistic baselines #2.", |
| "cite_spans": [ |
| { |
| "start": 452, |
| "end": 453, |
| "text": "9", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Results and conclusions", |
| "sec_num": "5" |
| }, |
| { |
| "text": "We can already conclude that our preliminary experiments turned successful. Now we are aiming at improving our operators to raise their recall and at expanding the semantic role set (e.g., for Agent, Causer, Experiencer, Possessor or Result). Parallel, we start work on construction of automatic algorithms for relation recognition. not avaiable.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Results and conclusions", |
| "sec_num": "5" |
| }, |
| { |
| "text": "9 Not directly, of course.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Results and conclusions", |
| "sec_num": "5" |
| }, |
| { |
| "text": "Work financed by The National Centre for Research and Development project SP/I/1/77065/10.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "Similarly to system presented in(Gamallo et al., 2002).", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "Random sample of 200 NPs/PPs taken from KPWr, 331 relation instances, bootstrap confidence intervals are following P = 83\u00f791%, R = 76\u00f784%, F = 79\u00f787%, \u03b1 = 0.05. The corpus was divided by us into two parts: one working set for testing and preparing parsing rules and semantic operatorsconsisting of 300 texts, and a smaller evaluation part of 26 texts.6 Since we do not use any word sense disambiguation system, we simply take the first sense of every given word.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "We focused on one-million balanced version of the much bigger corpus.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "Automatic extraction of semantic relations between medical entities: a rule based approach", |
| "authors": [ |
| { |
| "first": "Asma", |
| "middle": [], |
| "last": "Ben Abacha", |
| "suffix": "" |
| }, |
| { |
| "first": "Pierre", |
| "middle": [], |
| "last": "Zweigenbaum", |
| "suffix": "" |
| } |
| ], |
| "year": 2011, |
| "venue": "Journal of Biomedical Semantics", |
| "volume": "2", |
| "issue": "5", |
| "pages": "1--11", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Asma Ben Abacha and Pierre Zweigenbaum. 2011. Automatic extraction of semantic relations between medical entities: a rule based approach. Journal of Biomedical Semantics, 2(5):1-11.", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "KPWr: Towards a Free Corpus of Polish", |
| "authors": [ |
| { |
| "first": "Bartosz", |
| "middle": [], |
| "last": "Broda", |
| "suffix": "" |
| }, |
| { |
| "first": "Micha\u0142", |
| "middle": [], |
| "last": "Marci\u0144czuk", |
| "suffix": "" |
| }, |
| { |
| "first": "Marek", |
| "middle": [], |
| "last": "Maziarz", |
| "suffix": "" |
| }, |
| { |
| "first": "Adam", |
| "middle": [], |
| "last": "Radziszewski", |
| "suffix": "" |
| }, |
| { |
| "first": "Adam", |
| "middle": [], |
| "last": "Wardy\u0144ski", |
| "suffix": "" |
| } |
| ], |
| "year": 2012, |
| "venue": "Proceedings of LREC'12", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Bartosz Broda, Micha\u0142 Marci\u0144czuk, Marek Maziarz, Adam Radziszewski, and Adam Wardy\u0144ski. 2012. KPWr: Towards a Free Corpus of Polish. In Nico- letta Calzolari, Khalid Choukri, Thierry Declerck, Mehmet Ugur Dogan, Bente Maegaard, Joseph Mar- iani, Jan Odijk, and Stelios Piperidis, editors, Pro- ceedings of LREC'12, Istanbul, Turkey. ELRA.", |
| "links": null |
| }, |
| "BIBREF2": { |
| "ref_id": "b2", |
| "title": "Bootstrap confidence intervals", |
| "authors": [ |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Thomas", |
| "suffix": "" |
| }, |
| { |
| "first": "Bradley", |
| "middle": [], |
| "last": "Diciccio", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Efron", |
| "suffix": "" |
| } |
| ], |
| "year": 1996, |
| "venue": "Statistical Science", |
| "volume": "11", |
| "issue": "3", |
| "pages": "189--212", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Thomas J. DiCiccio and Bradley Efron. 1996. Boot- strap confidence intervals. Statistical Science, 11(3):189-212.", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "A review of bootstrap confidence intervals", |
| "authors": [ |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Thomas", |
| "suffix": "" |
| }, |
| { |
| "first": "Joseph", |
| "middle": [ |
| "P" |
| ], |
| "last": "Diciccio", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Romano", |
| "suffix": "" |
| } |
| ], |
| "year": 1988, |
| "venue": "Journal of the Royal Statistical Society. Series B (Methodological)", |
| "volume": "50", |
| "issue": "3", |
| "pages": "338--354", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Thomas J. DiCiccio and Joseph P. Romano. 1988. A review of bootstrap confidence intervals. Journal of the Royal Statistical Society. Series B (Methodolog- ical), 50(3):338-354.", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "Thematic proto-roles and argument selection", |
| "authors": [ |
| { |
| "first": "David", |
| "middle": [], |
| "last": "Dowty", |
| "suffix": "" |
| } |
| ], |
| "year": 1991, |
| "venue": "Language", |
| "volume": "67", |
| "issue": "3", |
| "pages": "547--619", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "David Dowty. 1991. Thematic proto-roles and argu- ment selection. Language, 67(3):547-619.", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "Argument co-occurrence matrix as a description of verb valence", |
| "authors": [ |
| { |
| "first": "\u0141ukasz", |
| "middle": [], |
| "last": "D\u0119bowski", |
| "suffix": "" |
| }, |
| { |
| "first": "Marcin", |
| "middle": [], |
| "last": "Woli\u0144ski", |
| "suffix": "" |
| } |
| ], |
| "year": 2007, |
| "venue": "Proc. of the 3rd Language & Technology Conference", |
| "volume": "3", |
| "issue": "", |
| "pages": "260--264", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "\u0141ukasz D\u0119bowski and Marcin Woli\u0144ski. 2007. Argu- ment co-occurrence matrix as a description of verb valence. In Proc. of the 3rd Language & Technol- ogy Conference, volume 3, pages 260-264, Pozna\u0144, Poland.", |
| "links": null |
| }, |
| "BIBREF6": { |
| "ref_id": "b6", |
| "title": "Polish valence dictionary", |
| "authors": [ |
| { |
| "first": "\u0141ukasz", |
| "middle": [], |
| "last": "D\u0119bowski", |
| "suffix": "" |
| } |
| ], |
| "year": 2013, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "\u0141ukasz D\u0119bowski. 2013. Polish valence dictionary (http://nlp.ipipan.waw.pl/ppjp/ slownik/swigra/kon- cowy.txt).", |
| "links": null |
| }, |
| "BIBREF7": { |
| "ref_id": "b7", |
| "title": "Mapping syntactic dependencies onto semantic relations", |
| "authors": [ |
| { |
| "first": "Pablo", |
| "middle": [], |
| "last": "Gamallo", |
| "suffix": "" |
| }, |
| { |
| "first": "Marco", |
| "middle": [], |
| "last": "Gonzalez", |
| "suffix": "" |
| }, |
| { |
| "first": "Alexandre", |
| "middle": [], |
| "last": "Agustini", |
| "suffix": "" |
| }, |
| { |
| "first": "Gabriel", |
| "middle": [], |
| "last": "Lopes", |
| "suffix": "" |
| }, |
| { |
| "first": "Vera", |
| "middle": [ |
| "S De" |
| ], |
| "last": "Lima", |
| "suffix": "" |
| } |
| ], |
| "year": 2002, |
| "venue": "ECAI Workshop on Machine Learning and Natural Language Processing for Ontology Engineering", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Pablo Gamallo, Marco Gonzalez, Alexandre Agustini, Gabriel Lopes, and Vera S. De Lima. 2002. Map- ping syntactic dependencies onto semantic relations. In ECAI Workshop on Machine Learning and Natu- ral Language Processing for Ontology Engineering.", |
| "links": null |
| }, |
| "BIBREF8": { |
| "ref_id": "b8", |
| "title": "Semantic role labelling without deep syntactic parsing", |
| "authors": [ |
| { |
| "first": "Konrad", |
| "middle": [], |
| "last": "Go\u0142uchowski", |
| "suffix": "" |
| }, |
| { |
| "first": "Adam", |
| "middle": [], |
| "last": "Przepi\u00f3rkowski", |
| "suffix": "" |
| } |
| ], |
| "year": 2012, |
| "venue": "Advances in Natural Language Processing", |
| "volume": "7614", |
| "issue": "", |
| "pages": "192--197", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Konrad Go\u0142uchowski and Adam Przepi\u00f3rkowski. 2012. Semantic role labelling without deep syntac- tic parsing. In Hitoshi Isahara and Kyoko Kanzaki, editors, Advances in Natural Language Processing, volume 7614 of Lecture Notes in Computer Science, pages 192-197. Springer Berlin Heidelberg.", |
| "links": null |
| }, |
| "BIBREF9": { |
| "ref_id": "b9", |
| "title": "Automatic acquisition of hyponyms from large text corpora", |
| "authors": [ |
| { |
| "first": "A", |
| "middle": [], |
| "last": "Marti", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Hearst", |
| "suffix": "" |
| } |
| ], |
| "year": 1992, |
| "venue": "Proceedings of the 14th conference on Computational linguistics", |
| "volume": "2", |
| "issue": "", |
| "pages": "539--545", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Marti A. Hearst. 1992. Automatic acquisition of hy- ponyms from large text corpora. In Proceedings of the 14th conference on Computational linguistics - Volume 2, COLING '92, pages 539-545, Strouds- burg, PA, USA. Association for Computational Lin- guistics.", |
| "links": null |
| }, |
| "BIBREF10": { |
| "ref_id": "b10", |
| "title": "Semantic relation extraction by automatically constructed rules", |
| "authors": [ |
| { |
| "first": "Rongfeng", |
| "middle": [], |
| "last": "Huang", |
| "suffix": "" |
| } |
| ], |
| "year": 2009, |
| "venue": "Proceedings of the International Conference on Artificial Intelligence and Computational Intelligence, AICI '09", |
| "volume": "", |
| "issue": "", |
| "pages": "425--434", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Rongfeng Huang. 2009. Semantic relation extrac- tion by automatically constructed rules. In Proceed- ings of the International Conference on Artificial Intelligence and Computational Intelligence, AICI '09, pages 425-434, Berlin, Heidelberg. Springer- Verlag.", |
| "links": null |
| }, |
| "BIBREF11": { |
| "ref_id": "b11", |
| "title": "Nominalizacje w systemie i tekstach wsp\u00f3\u0142czesnej polszczyzny", |
| "authors": [ |
| { |
| "first": "Ewa", |
| "middle": [], |
| "last": "J\u0119drzejko", |
| "suffix": "" |
| } |
| ], |
| "year": 1993, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Ewa J\u0119drzejko. 1993. Nominalizacje w systemie i tek- stach wsp\u00f3\u0142czesnej polszczyzny. Uniwersytet\u015al\u0105ski, Katowice.", |
| "links": null |
| }, |
| "BIBREF13": { |
| "ref_id": "b13", |
| "title": "WordNet: Understanding English Grammar", |
| "authors": [ |
| { |
| "first": "Martha", |
| "middle": [ |
| "J" |
| ], |
| "last": "Kolln", |
| "suffix": "" |
| } |
| ], |
| "year": 1990, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Martha J. Kolln. 1990. WordNet: Understanding En- glish Grammar. Macmillan.", |
| "links": null |
| }, |
| "BIBREF14": { |
| "ref_id": "b14", |
| "title": "Knowledge of Meaning: An Introduction to Semantic Theory", |
| "authors": [ |
| { |
| "first": "Gabriel", |
| "middle": [], |
| "last": "Larson", |
| "suffix": "" |
| }, |
| { |
| "first": "&", |
| "middle": [], |
| "last": "Richard", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Segal", |
| "suffix": "" |
| } |
| ], |
| "year": 1996, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Gabriel Larson, Richard & Segal. 1996. Knowledge of Meaning: An Introduction to Semantic Theory. The MIT Press.", |
| "links": null |
| }, |
| "BIBREF15": { |
| "ref_id": "b15", |
| "title": "Gramatyka wsp\u00f3\u0142czesnego j\u0119zyka polskiego. Morfologia", |
| "authors": [], |
| "year": 1997, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Roman Laskowski and Henryk Wr\u00f3bel, editors. 1997. Gramatyka wsp\u00f3\u0142czesnego j\u0119zyka polskiego. Mor- fologia [Grammar of contemporary Polish: Mor- phology]. PWN.", |
| "links": null |
| }, |
| "BIBREF16": { |
| "ref_id": "b16", |
| "title": "Cyc: a large-scale investment in knowledge infrastructure", |
| "authors": [ |
| { |
| "first": "B", |
| "middle": [], |
| "last": "Douglas", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Lenat", |
| "suffix": "" |
| } |
| ], |
| "year": 1995, |
| "venue": "Luna project: Spoken language understanding in multilingual communication systems", |
| "volume": "38", |
| "issue": "", |
| "pages": "33--38", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Douglas B. Lenat. 1995. Cyc: a large-scale invest- ment in knowledge infrastructure. Commun. ACM, 38(11):33-38, November. 2009. Luna project: Spoken language under- standing in multilingual communication systems (http://zil.ipipan.waw.pl/luna).", |
| "links": null |
| }, |
| "BIBREF17": { |
| "ref_id": "b17", |
| "title": "Chunking of Polish: guidelines, discussion and experiments with Machine Learning", |
| "authors": [ |
| { |
| "first": "Marek", |
| "middle": [], |
| "last": "Maziarz", |
| "suffix": "" |
| }, |
| { |
| "first": "Adam", |
| "middle": [], |
| "last": "Radziszewski", |
| "suffix": "" |
| } |
| ], |
| "year": 2011, |
| "venue": "Proc. of the LTC", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Marek Maziarz, Adam Radziszewski, and Jan Wiec- zorek. 2011. Chunking of Polish: guidelines, dis- cussion and experiments with Machine Learning. In Proc. of the LTC.", |
| "links": null |
| }, |
| "BIBREF18": { |
| "ref_id": "b18", |
| "title": "Approaching plWordNet 2.0", |
| "authors": [ |
| { |
| "first": "Marek", |
| "middle": [], |
| "last": "Maziarz", |
| "suffix": "" |
| }, |
| { |
| "first": "Maciej", |
| "middle": [], |
| "last": "Piasecki", |
| "suffix": "" |
| }, |
| { |
| "first": "Stanis\u0142aw", |
| "middle": [], |
| "last": "Szpakowicz", |
| "suffix": "" |
| } |
| ], |
| "year": 2012, |
| "venue": "Proceedings of the 6th Global Wordnet Conference", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Marek Maziarz, Maciej Piasecki, and Stanis\u0142aw Sz- pakowicz. 2012. Approaching plWordNet 2.0. In Proceedings of the 6th Global Wordnet Conference, Matsue, Japan, January.", |
| "links": null |
| }, |
| "BIBREF19": { |
| "ref_id": "b19", |
| "title": "Models for the semantic classification of noun phrases", |
| "authors": [ |
| { |
| "first": "Dan", |
| "middle": [], |
| "last": "Moldovan", |
| "suffix": "" |
| }, |
| { |
| "first": "Adriana", |
| "middle": [], |
| "last": "Badulescu", |
| "suffix": "" |
| }, |
| { |
| "first": "Marta", |
| "middle": [], |
| "last": "Tatu", |
| "suffix": "" |
| }, |
| { |
| "first": "Daniel", |
| "middle": [], |
| "last": "Antohe", |
| "suffix": "" |
| }, |
| { |
| "first": "Roxana", |
| "middle": [], |
| "last": "Girju", |
| "suffix": "" |
| } |
| ], |
| "year": 2004, |
| "venue": "Proceedings of the HLT-NAACL Workshop on Computational Lexical Semantics, CLS '04", |
| "volume": "", |
| "issue": "", |
| "pages": "60--67", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Dan Moldovan, Adriana Badulescu, Marta Tatu, Daniel Antohe, and Roxana Girju. 2004. Models for the semantic classification of noun phrases. In Pro- ceedings of the HLT-NAACL Workshop on Compu- tational Lexical Semantics, CLS '04, pages 60-67, Stroudsburg, PA, USA. Association for Computa- tional Linguistics.", |
| "links": null |
| }, |
| "BIBREF20": { |
| "ref_id": "b20", |
| "title": "Learning noun-modifier semantic relations with corpus-based and wordnet-based features", |
| "authors": [ |
| { |
| "first": "Vivi", |
| "middle": [], |
| "last": "Nastase", |
| "suffix": "" |
| }, |
| { |
| "first": "Jelber", |
| "middle": [], |
| "last": "Sayyad Shirabad", |
| "suffix": "" |
| }, |
| { |
| "first": "Marina", |
| "middle": [], |
| "last": "Sokolova", |
| "suffix": "" |
| }, |
| { |
| "first": "Stan", |
| "middle": [], |
| "last": "Szpakowicz", |
| "suffix": "" |
| } |
| ], |
| "year": 2006, |
| "venue": "Proceedings of the 21st National Conference on Artificial Intelligence", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Vivi Nastase, Jelber Sayyad Shirabad, Marina Sokolova, and Stan Szpakowicz. 2006. Learning noun-modifier semantic relations with corpus-based and wordnet-based features. In Proceedings of the 21st National Conference on Artificial Intelligence.", |
| "links": null |
| }, |
| "BIBREF21": { |
| "ref_id": "b21", |
| "title": "Semantic Role Labeling", |
| "authors": [ |
| { |
| "first": "Martha", |
| "middle": [ |
| "&" |
| ], |
| "last": "Palmer", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Gilldea", |
| "suffix": "" |
| }, |
| { |
| "first": "&", |
| "middle": [], |
| "last": "Daniel", |
| "suffix": "" |
| }, |
| { |
| "first": "Nianwen", |
| "middle": [], |
| "last": "Xue", |
| "suffix": "" |
| } |
| ], |
| "year": 2010, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Palmer, Martha & Gilldea, Daniel & Xue, and Nian- wen. 2010. Semantic Role Labeling. Morgan & Claypool Publishers.", |
| "links": null |
| }, |
| "BIBREF22": { |
| "ref_id": "b22", |
| "title": "2012. Narodowy Korpus Jezyka Polskiego", |
| "authors": [ |
| { |
| "first": "Adam", |
| "middle": [], |
| "last": "Przepi\u00f3rkowski", |
| "suffix": "" |
| }, |
| { |
| "first": "Miroslaw", |
| "middle": [], |
| "last": "Banko", |
| "suffix": "" |
| } |
| ], |
| "year": null, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Adam Przepi\u00f3rkowski, Miroslaw Banko, Rafal G\u00f3rski, and Barbara Lewandowska-Tomaszczyk, editors. 2012. Narodowy Korpus Jezyka Polskiego.", |
| "links": null |
| }, |
| "BIBREF24": { |
| "ref_id": "b24", |
| "title": "Large-scale experiments with NP chunking of Polish", |
| "authors": [ |
| { |
| "first": "Adam", |
| "middle": [], |
| "last": "Radziszewski", |
| "suffix": "" |
| }, |
| { |
| "first": "Adam", |
| "middle": [], |
| "last": "Pawlaczek", |
| "suffix": "" |
| } |
| ], |
| "year": 2012, |
| "venue": "Proceedings of TSD 2012", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Adam Radziszewski and Adam Pawlaczek. 2012. Large-scale experiments with NP chunking of Pol- ish. In Proceedings of TSD 2012, Brno, Czech Re- public. Springer.", |
| "links": null |
| }, |
| "BIBREF25": { |
| "ref_id": "b25", |
| "title": "WCCL: A Morpho-syntactic Feature Toolkit", |
| "authors": [ |
| { |
| "first": "Adam", |
| "middle": [], |
| "last": "Radziszewski", |
| "suffix": "" |
| }, |
| { |
| "first": "Adam", |
| "middle": [], |
| "last": "Wardy\u0144ski", |
| "suffix": "" |
| }, |
| { |
| "first": "Tomas\u017a", |
| "middle": [], |
| "last": "Sniatowski", |
| "suffix": "" |
| } |
| ], |
| "year": 2011, |
| "venue": "Proceedings of the Balto-Slavonic Natural Language Processing Workshop", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Adam Radziszewski, Adam Wardy\u0144ski, and Tomas\u017a Sniatowski. 2011. WCCL: A Morpho-syntactic Feature Toolkit. In Proceedings of the Balto- Slavonic Natural Language Processing Workshop. Springer.", |
| "links": null |
| }, |
| "BIBREF26": { |
| "ref_id": "b26", |
| "title": "Shallow syntactic annotation in the Corpus of Wroc\u0142aw University of Technolog. Cognitive Studies", |
| "authors": [ |
| { |
| "first": "Adam", |
| "middle": [], |
| "last": "Radziszewski", |
| "suffix": "" |
| }, |
| { |
| "first": "Marek", |
| "middle": [], |
| "last": "Maziarz", |
| "suffix": "" |
| } |
| ], |
| "year": 2012, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Adam Radziszewski, Marek Maziarz, and Jan Wiec- zorek. 2012. Shallow syntactic annotation in the Corpus of Wroc\u0142aw University of Technolog. Cog- nitive Studies, 12.", |
| "links": null |
| }, |
| "BIBREF27": { |
| "ref_id": "b27", |
| "title": "A tiered CRF tagger for Polish", |
| "authors": [ |
| { |
| "first": "Adam", |
| "middle": [], |
| "last": "Radziszewski", |
| "suffix": "" |
| } |
| ], |
| "year": 2013, |
| "venue": "R. Bembenik, \u0141. Skonieczny, H. Rybi\u0144ski, M. Kryszkiewicz, and M. Niezg\u00f3dka, editors, Intelligent Tools for Building a Scientific Information Platform: Advanced Architectures and Solutions", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Adam Radziszewski. 2013. A tiered CRF tagger for Polish. In R. Bembenik, \u0141. Skonieczny, H. Ry- bi\u0144ski, M. Kryszkiewicz, and M. Niezg\u00f3dka, edi- tors, Intelligent Tools for Building a Scientific Infor- mation Platform: Advanced Architectures and Solu- tions.", |
| "links": null |
| }, |
| "BIBREF28": { |
| "ref_id": "b28", |
| "title": "Utd: Classifying semantic relations by combining lexical and semantic resources", |
| "authors": [ |
| { |
| "first": "Bryan", |
| "middle": [], |
| "last": "Rink", |
| "suffix": "" |
| }, |
| { |
| "first": "Sanda", |
| "middle": [], |
| "last": "Harabagiu", |
| "suffix": "" |
| } |
| ], |
| "year": 2010, |
| "venue": "Proceedings of the 5th International Workshop on Semantic Evaluation, Se-mEval '10", |
| "volume": "", |
| "issue": "", |
| "pages": "256--259", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Bryan Rink and Sanda Harabagiu. 2010. Utd: Clas- sifying semantic relations by combining lexical and semantic resources. In Proceedings of the 5th In- ternational Workshop on Semantic Evaluation, Se- mEval '10, pages 256-259, Stroudsburg, PA, USA. Association for Computational Linguistics.", |
| "links": null |
| }, |
| "BIBREF29": { |
| "ref_id": "b29", |
| "title": "Classifying the semantic relations in noun compounds via a domain-specific lexical hierarchy", |
| "authors": [ |
| { |
| "first": "Barbara", |
| "middle": [], |
| "last": "Rosario", |
| "suffix": "" |
| }, |
| { |
| "first": "Marti", |
| "middle": [], |
| "last": "Hearst", |
| "suffix": "" |
| } |
| ], |
| "year": 2001, |
| "venue": "Proceedings of the 2001 Conference on Empirical Methods in Natural Language Processing", |
| "volume": "", |
| "issue": "", |
| "pages": "82--90", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Barbara Rosario and Marti Hearst. 2001. Classify- ing the semantic relations in noun compounds via a domain-specific lexical hierarchy. In Proceedings of the 2001 Conference on Empirical Methods in Nat- ural Language Processing (EMNLP-01), pages 82- 90.", |
| "links": null |
| }, |
| "BIBREF30": { |
| "ref_id": "b30", |
| "title": "Isi: Automatic classification of relations between nominals using a maximum entropy classifier", |
| "authors": [ |
| { |
| "first": "Stephen", |
| "middle": [], |
| "last": "Tratz", |
| "suffix": "" |
| }, |
| { |
| "first": "Eduard", |
| "middle": [], |
| "last": "Hovy", |
| "suffix": "" |
| } |
| ], |
| "year": 2010, |
| "venue": "Proc. of the 5th Intern. Workshop on Semantic Evaluation", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Stephen Tratz and Eduard Hovy. 2010. Isi: Automatic classification of relations between nominals using a maximum entropy classifier. In Proc. of the 5th Intern. Workshop on Semantic Evaluation, Sweden. Association for Computational Linguistics.", |
| "links": null |
| }, |
| "BIBREF31": { |
| "ref_id": "b31", |
| "title": "Fbk-irst: Semantic relation extraction using cyc", |
| "authors": [ |
| { |
| "first": "Kateryna", |
| "middle": [], |
| "last": "Tymoshenko", |
| "suffix": "" |
| }, |
| { |
| "first": "Claudio", |
| "middle": [], |
| "last": "Giuliano", |
| "suffix": "" |
| } |
| ], |
| "year": 2010, |
| "venue": "Proceedings of the 5th International Workshop on Semantic Evaluation", |
| "volume": "", |
| "issue": "", |
| "pages": "214--217", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Kateryna Tymoshenko and Claudio Giuliano. 2010. Fbk-irst: Semantic relation extraction using cyc. In Proceedings of the 5th International Workshop on Semantic Evaluation, pages 214-217, Uppsala, Sweden, July. Association for Computational Lin- guistics.", |
| "links": null |
| }, |
| "BIBREF32": { |
| "ref_id": "b32", |
| "title": "Identifying semantic relations in japanese compound nouns for patent documents analysis", |
| "authors": [ |
| { |
| "first": "Kiyoko", |
| "middle": [], |
| "last": "Uchiyama", |
| "suffix": "" |
| }, |
| { |
| "first": "Shunsuke", |
| "middle": [], |
| "last": "Aihara", |
| "suffix": "" |
| }, |
| { |
| "first": "Shun", |
| "middle": [], |
| "last": "Ishizaki", |
| "suffix": "" |
| } |
| ], |
| "year": 2008, |
| "venue": "Proceedings of the 3rd international conference on Large-scale knowledge resources: construction and application, LKR'08", |
| "volume": "", |
| "issue": "", |
| "pages": "75--81", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Kiyoko Uchiyama, Shunsuke Aihara, and Shun Ishizaki. 2008. Identifying semantic relations in japanese compound nouns for patent documents analysis. In Proceedings of the 3rd international conference on Large-scale knowledge resources: construction and application, LKR'08, pages 75- 81, Berlin, Heidelberg. Springer-Verlag.", |
| "links": null |
| }, |
| "BIBREF33": { |
| "ref_id": "b33", |
| "title": "Semantic macroroles in role and reference grammar", |
| "authors": [ |
| { |
| "first": "D", |
| "middle": [], |
| "last": "Robert", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Van Valin", |
| "suffix": "" |
| } |
| ], |
| "year": 2004, |
| "venue": "Semantische Rollen", |
| "volume": "", |
| "issue": "", |
| "pages": "62--82", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Robert D. Van Valin. 2004. Semantic macroroles in role and reference grammar. In Rolf Kailuweit and Martin Hummel, editors, Semantische Rollen, pages 62-82. Tuebingen Narr.", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "FIGREF0": { |
| "num": null, |
| "text": "noun or adj. & accusative in(class[1], {subst,depr,ger,adj}), equal(cas[1],acc) ) )", |
| "type_str": "figure", |
| "uris": null |
| }, |
| "FIGREF1": { |
| "num": null, |
| "text": "Listing 2: A WCCL operator for the Proto-Agent relation @b:\"PROTO-AGENT-ger-przez-acc\" ( [2], {ger}), lex(base[2], \"ger_base\"), base[2]), \"noun_domain\"), [\"czas\"])) ) )", |
| "type_str": "figure", |
| "uris": null |
| }, |
| "FIGREF2": { |
| "num": null, |
| "text": ", \"noun_domain\"), [\"os\"]), in(class[1], {ger, subst, depr}), ) )", |
| "type_str": "figure", |
| "uris": null |
| }, |
| "TABREF1": { |
| "content": "<table><tr><td colspan=\"5\">: Results of the algorithm on a sample from KPWr:</td></tr><tr><td colspan=\"5\">P = Precision, R = recall, F1 = F-measure, TP = true positives,</td></tr><tr><td colspan=\"5\">FP = false positives, FN = false negatives, sp. = spatial, t. =</td></tr><tr><td colspan=\"5\">temporal. Percentile bootstrap confidence intervals are cal-</td></tr><tr><td colspan=\"5\">culated at \u03b1 = 0.05. Asterisks denote significant differences</td></tr><tr><td colspan=\"4\">between kpwr and nkjp in one-tailed tests, \u03b1 = 0.05</td><td/></tr><tr><td>Relation</td><td>TP/FP/FN</td><td>P [%]</td><td>R [%]</td><td>F1 [%]</td></tr><tr><td>Proto-Agent</td><td>75/7/143</td><td>91.5</td><td>34.4</td><td>50.0</td></tr><tr><td>Proto-Patient</td><td>181/17/230</td><td>91.4</td><td>44.0</td><td>59.4</td></tr><tr><td>Instrument</td><td>2/1/8</td><td>66.7</td><td>20.0</td><td>30.8</td></tr><tr><td>Material</td><td>3/4/36</td><td>42.9</td><td>7.7</td><td>13.0</td></tr><tr><td>Purpose</td><td>13/7/94</td><td>65.0</td><td>12.2</td><td>20.5</td></tr><tr><td>location</td><td>90/75/202</td><td>54.6</td><td>30.8</td><td>39.4</td></tr><tr><td>sp. meronymy</td><td>12/11/25</td><td>52.2</td><td>32.4</td><td>40.0</td></tr><tr><td>time</td><td>25/16/75</td><td>61.0</td><td>25.0</td><td>35.5</td></tr><tr><td>t. meronymy</td><td>2/0/66</td><td>100</td><td>2.9</td><td>57.1</td></tr><tr><td>attribute</td><td>200/248/64</td><td>44.6</td><td>75.8</td><td>56.2</td></tr><tr><td>family</td><td>18/0/6</td><td>100.0</td><td>60.0</td><td>85.7</td></tr><tr><td>order</td><td>33/0/100</td><td>100.0</td><td>24.8</td><td>39.8</td></tr><tr><td>quantity</td><td>113/68/146</td><td>62.4</td><td>43.6</td><td>51.4</td></tr><tr><td>All</td><td colspan=\"4\">767/454/1195 60.1-65.6 *36.9-41.2 *46.0-50.3</td></tr></table>", |
| "type_str": "table", |
| "html": null, |
| "num": null, |
| "text": "" |
| }, |
| "TABREF2": { |
| "content": "<table/>", |
| "type_str": "table", |
| "html": null, |
| "num": null, |
| "text": "Results of the algorithm on a sample from NKJP, labels as in the previous table. Percentile bootstrap confidence intervals are calculated at \u03b1 = 0.05. Asterisks denote significant differences between kpwr and nkjp in one-tailed tests, \u03b1 = 0.05" |
| }, |
| "TABREF4": { |
| "content": "<table/>", |
| "type_str": "table", |
| "html": null, |
| "num": null, |
| "text": "Precision, recall and F1 for baselines (#1 & #2) and" |
| } |
| } |
| } |
| } |
