| { | |
| "paper_id": "2016", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T01:03:19.448590Z" | |
| }, | |
| "title": "plWordNet in Word Sense Disambiguation", | |
| "authors": [ | |
| { | |
| "first": "Maciej", | |
| "middle": [], | |
| "last": "Piasecki", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Wroc\u0142aw University of Technology", | |
| "location": { | |
| "country": "Poland" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Pawe\u0142", | |
| "middle": [], | |
| "last": "K\u0119dzia", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Wroc\u0142aw University of Technology", | |
| "location": { | |
| "country": "Poland" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Marlena", | |
| "middle": [], | |
| "last": "Orli\u0144ska", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Wroc\u0142aw University of Technology", | |
| "location": { | |
| "country": "Poland" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "{maciej", | |
| "middle": [], | |
| "last": "Piasecki", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Wroc\u0142aw University of Technology", | |
| "location": { | |
| "country": "Poland" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Marlena", | |
| "middle": [], | |
| "last": "Kedzia", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Wroc\u0142aw University of Technology", | |
| "location": { | |
| "country": "Poland" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [ | |
| "Edu" | |
| ], | |
| "last": "Orlinska}", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Wroc\u0142aw University of Technology", | |
| "location": { | |
| "country": "Poland" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Pl", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Wroc\u0142aw University of Technology", | |
| "location": { | |
| "country": "Poland" | |
| } | |
| }, | |
| "email": "" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "The paper explores the application of plWordNet, a very large wordnet of Polish, in weakly supervised Word Sense Disambiguation (WSD). Because plWord-Net provides only partial descriptions by glosses and usage examples, and does not include sense-disambiguated glosses, PageRank-based WSD methods perform slightly worse than for English. However, we show that the use of weights for the relation types and the order in which lexical units have been added for sense re-ranking can significantly improve WSD precision. The evaluation was done on two Polish corpora (KPWr and Sk\u0142adnica) including manual WSD. We discuss the fundamental difference in the construction of both corpora and very different test results.", | |
| "pdf_parse": { | |
| "paper_id": "2016", | |
| "_pdf_hash": "", | |
| "abstract": [ | |
| { | |
| "text": "The paper explores the application of plWordNet, a very large wordnet of Polish, in weakly supervised Word Sense Disambiguation (WSD). Because plWord-Net provides only partial descriptions by glosses and usage examples, and does not include sense-disambiguated glosses, PageRank-based WSD methods perform slightly worse than for English. However, we show that the use of weights for the relation types and the order in which lexical units have been added for sense re-ranking can significantly improve WSD precision. The evaluation was done on two Polish corpora (KPWr and Sk\u0142adnica) including manual WSD. We discuss the fundamental difference in the construction of both corpora and very different test results.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Abstract", | |
| "sec_num": null | |
| } | |
| ], | |
| "body_text": [ | |
| { | |
| "text": "Large wordnets are often treated as sense inventories that describe and enumerate word senses. If we want to process texts at the level of wordnet senses, a very useful operation, we first must map text words to those senses, i.e. to perform Word Sense Disambiguation (henceforth WSD). This is only trivial for monosemous words. WSD methods built upon supervised Machine Learning achieve good accuracy but are intrinsically impractical in their dependence on corpora that have been manually disambiguated with respect to word senses. Needless to say, such corpora are very laborious to annotate.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "Weakly supervised WSD methods that use a wordnet as the basic knowledge source, but do not depend on a manually annotated corpus, can fully utilise wordnet senses, i.e. they can in theory assign any sense stored in a wordnet to words in text. So, in spite of their lower precision they seem to be noteworthy as a potentially practical solution. Most wordnet-based weakly supervised WSD methods are based on the idea of spreading activation in the wordnet graph, where the initial activation comes from the words in a textual context.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "Several methods based on this general scheme were proposed. A short overview is presented in Section 2. Most such methods were developed and tested on Princeton WordNet (PWN) (Fellbaum, 1998) that is slightly different than plWordNet (Piasecki et al., 2009 , Maziarz et al., 2013a , currently the world largest wordnet. First attempts to transfer the methods with good performance on PWN to plWordNet (K\u0119dzia et al., 2015) were encouraging; the performance is relatively close to the performance of the supervised methods observed for Polish on limited test sets (Ba\u015b et al., 2008, M\u0142odzki and Przepi\u00f3rkowski, 2009) . In addition to the differences between both wordnets, PWN has been enriched with various other resources in order to obtain better performance of unsupervised WSD. First of all, additional links between synsets were created on the basis of the manually disambiguated SemCore corpus (Miller et al., 1993) . Such links have contributed significantly to the increase of WSD performance. There is no Polish corpus similar to SemCore.", | |
| "cite_spans": [ | |
| { | |
| "start": 175, | |
| "end": 191, | |
| "text": "(Fellbaum, 1998)", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 234, | |
| "end": 256, | |
| "text": "(Piasecki et al., 2009", | |
| "ref_id": "BIBREF24" | |
| }, | |
| { | |
| "start": 257, | |
| "end": 280, | |
| "text": ", Maziarz et al., 2013a", | |
| "ref_id": "BIBREF13" | |
| }, | |
| { | |
| "start": 401, | |
| "end": 422, | |
| "text": "(K\u0119dzia et al., 2015)", | |
| "ref_id": "BIBREF10" | |
| }, | |
| { | |
| "start": 563, | |
| "end": 593, | |
| "text": "(Ba\u015b et al., 2008, M\u0142odzki and", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 594, | |
| "end": 615, | |
| "text": "Przepi\u00f3rkowski, 2009)", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 900, | |
| "end": 921, | |
| "text": "(Miller et al., 1993)", | |
| "ref_id": "BIBREF18" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "The goal of the work presented here is to explore the structure and specific properties of plWordNet in order to improve the precision of the WSD methods based on the spreading activation in the wordnet graph, here the plWordNet graph.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "In the rest of the paper, first we will briefly overview the existing wordnet-based unsupervised WSD methods, including their known applications to plWordNet. Next, the plWordNet model will be discussed and compared with PWN from the perspective of utilising different features in WSD method. On this basis, several possible versions of unsupervised WSD will be introduced. Finally, we will present data sets used in the evaluation and the results achieved for different settings used in WSD methods. Based on the results, we will analyse the the specific properties of plWordNet and its development process and its influence on wordnetbased unsupervised WSD methods for Polish.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "Unsupervised WSD methods (Pantel, 2003) use corpora to induce word senses and tune mechanisms for assignment of the induced senses to words. However, it is difficult to map the induced word senses to the wordnet. Weakly supervised WSD that are based on a wordnet as the knowledge base work directly on wordnet synsets and do not depend on manually disambiguated corpus.", | |
| "cite_spans": [ | |
| { | |
| "start": 25, | |
| "end": 39, | |
| "text": "(Pantel, 2003)", | |
| "ref_id": "BIBREF22" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Wordnet-based WSD", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "Lesk's algorithm (Lesk, 1986) can be applied to textual definitions constructed on the basis on of synsets, e.g. from glosses, examples and synset members. The definitions are next compared with the occurrence contexts of words. Different similarity measures can be applied. The main problems are limited lenghts of the constructed definitions and high computational complexity, because many word sets must be compared.", | |
| "cite_spans": [ | |
| { | |
| "start": 17, | |
| "end": 29, | |
| "text": "(Lesk, 1986)", | |
| "ref_id": "BIBREF11" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Wordnet-based WSD", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "Weakly supervised wordnet-based WSD algorithms assume that if we map words senses pertaining to a text fragment onto the wordnet graph, we can expect that the \"hits\" are located in short distances (in terms of paths) from each other in the wordnet graph. Moreover, we can use a kind of spreading activation algorithm in order to move this information along the wordnet graph, analyse the \"hot\" areas and identify word sense, i.e. lexical units (LUs), 1 located in them or close to them. Those LUs should be the most likely senses for words in the text. There are several parameters to set in this general scheme: the initial activation (text words vs LUs), spreading algorithm (topology and relations) and identification of association between \"hot\" areas and LUs to be chosen. Various methods propose a range of decisions.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Wordnet-based WSD", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "Weakly supervised WSD methods are mostly based on the PageRank algorithm (Page et al., 1999) for spreading. Mihalcea et al. (2004) proposed application of the original PageRank to WSD called Static PageRank.", | |
| "cite_spans": [ | |
| { | |
| "start": 73, | |
| "end": 92, | |
| "text": "(Page et al., 1999)", | |
| "ref_id": "BIBREF21" | |
| }, | |
| { | |
| "start": 108, | |
| "end": 130, | |
| "text": "Mihalcea et al. (2004)", | |
| "ref_id": "BIBREF17" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Wordnet-based WSD", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "Page Rank algorithm (henceforth PR) is an iterative method for ranking nodes in the graph G. In WSD the nodes in G represent synsets and the 1 See Section 3 for more on LUs. edges of G correspond to wordnet relations (between synsets and in other case between synsets and between LUs). The spreading is done iteratively in the following way:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Wordnet-based WSD", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "P (new) = cM P (old) + (1 \u2212 c)v", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Wordnet-based WSD", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "(1) M N \u00d7N ins the adjacency matrix of the wordnet graph with N nodes (synsets), where", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Wordnet-based WSD", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "m ij = 1 d i", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Wordnet-based WSD", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "if the edge from the node s i to s j exists, 0 otherwise; d i is degree of the node s i (representing the synset i); where c is the damping factor; v N\u00d71 is the vector of the initial scores for nodes and P N\u00d71 is a vector of node scores updated in every iteration. In Static PageRank (SPR) all values in v are equal 1/N . Agirre and Soroa (2009) , Agirre et al. (2014) proposed a modified version called Personalised PageRank (PPR) in which the values in v, called personalised vector, depends on the text context of the disambiguated word. The non-zero score values are assigned to those nodes which are contextually supported. In PPR all words from the context are disambiguated at once. The v values are equal to:", | |
| "cite_spans": [ | |
| { | |
| "start": 322, | |
| "end": 345, | |
| "text": "Agirre and Soroa (2009)", | |
| "ref_id": "BIBREF0" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Wordnet-based WSD", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "EQUATION", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [ | |
| { | |
| "start": 0, | |
| "end": 8, | |
| "text": "EQUATION", | |
| "ref_id": "EQREF", | |
| "raw_str": "v[i] = 1 CS N S(i) , i = 1, 2, ..., N", | |
| "eq_num": "(2)" | |
| } | |
| ], | |
| "section": "Wordnet-based WSD", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "where CS is the number of different lemmas in the context, N S(i) -the number of synsets sharing the same context lemma with the synset i. Agirre and Soroa (2009) , Stevenson et al. (2012) proposed a modified version of PPR called Personalised PageRank Word-to-Word (PPR_W2W), in which a word to be disambiguated is excluded from the occurrence contexts, i.e. all synsets of this word have initial scores in v set to zero. Thus, PPR_W2W cannot be run once for all ambiguous words in the context. The vector v must be initialised individually for each ambiguous word in the context -this is a disadvantage of PPR_W2W. A potential advantage is the removal of the effect of mutual amplification of the closely connected senses of the word being disambiguated. The best results (measured in recall) are obtained on the Senseval-2 dataset for a graph built from WordNet 1.7 and eXtended WordNet (Harabagiu et al., 1999) . For nouns the best results are obtained using PPR (recall 71.1%), for verbs and adjectives with PPR_W2W recall was between 38.9% and 58.3%. For adverbs SPR achieved the best result of 70.8%. The best result for nouns, 71.9%, was achieved by PPR_W2W on the basis of the combination of WordNet 3.0 with disambiguated glosses.", | |
| "cite_spans": [ | |
| { | |
| "start": 139, | |
| "end": 162, | |
| "text": "Agirre and Soroa (2009)", | |
| "ref_id": "BIBREF0" | |
| }, | |
| { | |
| "start": 165, | |
| "end": 188, | |
| "text": "Stevenson et al. (2012)", | |
| "ref_id": "BIBREF27" | |
| }, | |
| { | |
| "start": 890, | |
| "end": 914, | |
| "text": "(Harabagiu et al., 1999)", | |
| "ref_id": "BIBREF7" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Wordnet-based WSD", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "In , SPR algorithm for Polish was based on plWordNet 2.1. The graph consisted of synsets linked by edges representing a selected subset of the synset relations. The precision on nouns (43%) and verbs (28%) was low in comparison to the works for English. The algorithm was evaluated on the KPWr corpus of Polish discussed in Section 5. In the second version, a Measure of Semantic Relatedness was utilised to add links to plWordNet. The measure had been extracted automatically from a large corpus of 1.8 billion words. However, there was no improvement: the precision for nouns was 37% and 27% for verbs. Nevertheless, we observed that even a WSD method of limited precision can be helpful in improving the performance of text clustering.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Wordnet-based WSD", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "Next we adapted several algorithms: SPR, PPR and PPR_W2W -to Polish resources K\u0119dzia et al. (2015) . plWordNet 2.2 was used with all synset relations for the edges. Due to the lack of word-sense disambiguation of glosses, no additional synset links could be added. The achieved precision (on KPWr) was in the range 42.79%-50.73% for nouns and 29.79%-32.94% for verbs. PPR_W2W produced the best results. We also tested different variants of combining plWord-Net with the Suggested Upper Merged Ontology (SUMO) (Pease, 2011) on the basis of the mapping constructed in . All three PR-based algorithm were evaluated. A slight improvement of the precision for nouns up to 50.89% for PPR_W2W could be observed when the two joined graphs were treated as one large graph.", | |
| "cite_spans": [ | |
| { | |
| "start": 78, | |
| "end": 98, | |
| "text": "K\u0119dzia et al. (2015)", | |
| "ref_id": "BIBREF10" | |
| }, | |
| { | |
| "start": 509, | |
| "end": 522, | |
| "text": "(Pease, 2011)", | |
| "ref_id": "BIBREF23" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Wordnet-based WSD", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "plWordNet is a very large wordnet built independently from PWN and expresses several unique features. Word senses are represented in plWord-Net as lexical units (LUs), i.e. pairs: lemma 2 plus sense identifier. LUs are the basic building blocks of plWordNet, but one LU belongs to exactly one synset. plWordNet includes about 40 main types of lexico-semantic relation. Half of them links synsets, the rest directly link LUs (Piasecki et al., 2009 , 2013a . Many relations, e.g. meronymy, have subtypes, so the total number of lexico-semantic relations in plWordNet 2.3 exceeds 90.", | |
| "cite_spans": [ | |
| { | |
| "start": 424, | |
| "end": 446, | |
| "text": "(Piasecki et al., 2009", | |
| "ref_id": "BIBREF24" | |
| }, | |
| { | |
| "start": 447, | |
| "end": 454, | |
| "text": ", 2013a", | |
| "ref_id": "BIBREF13" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "plWordNet properties", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "The detailed description of the model underlying plWordNet can be found in (Maziarz et al., 2013b) , below we present only a concise overview due to the space limit. LUs that share a set of constitutive lexico-semantic relations are grouped into synsets that are considered to consists of near synonyms. Synset relations are notational abbreviations for the relations shared between LUs from the linked synsets. The relations are the basic means of describing word senses. Different types of relations express different semantic associations, and provide different semantic information. This properties can be explored in WSD to improve the use of knowledge during spreading activation in the graph. plWordNet provides as well some additional means of semantic description: stylistic registers, glosses and use examples. Stylistic registers signal pragmatic constraints on the use of LUs. However, such subtle differences are difficult to explore in WSD methods, so we have not done it. Glosses in plWordNet are comments to the LUs (not to synsets like in PWN) provided for a human reader in order to explain the motivation behind the given word sense and clarify its difference from other senses of the same lemma. Glosses are short descriptions but they are not proper lexicographic definitions and are much less elaborated from the point of view of their application in Lesk's algorithm (Lesk, 1986) . Glosses are intended to be secondary and additional to the lexico-semantic relations that are the primary tool for the description of the lexical meanings in plWordNet, e.g. the genus information is expressed by hypernymy and should not be provided in a gloss. As such they have been added only to a subset of LUs. In addition to glosses, LU can be described by one or more use examples. They are also focused on human readers, but they can be used in WSD as an additional source of information. There have been not attempts so far to disambiguate word senses in the plWordNet glosses and examples. plWordNet has been automatically mapped onto SUMO with high precision. The extended graph, plWordNet plus SUMO, has been already used in WSD with positive signals, discussed in Section 2.", | |
| "cite_spans": [ | |
| { | |
| "start": 75, | |
| "end": 98, | |
| "text": "(Maziarz et al., 2013b)", | |
| "ref_id": "BIBREF14" | |
| }, | |
| { | |
| "start": 1390, | |
| "end": 1402, | |
| "text": "(Lesk, 1986)", | |
| "ref_id": "BIBREF11" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "plWordNet properties", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "plWordNet LUs are not clustered into semantic domains, but only into PWN-like, i.e. domains that correspond to the lexicographer files introduced in early stages of PWN development (Fellbaum, 1998) . They do not seem to provide important knowledge for WSD.", | |
| "cite_spans": [ | |
| { | |
| "start": 181, | |
| "end": 197, | |
| "text": "(Fellbaum, 1998)", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "plWordNet properties", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "Finally, there is no information about the frequency or salience of LUs, e.g. in comparison to other LUs of the same lemma. Numerical identifiers of LUs and the order of synsets in the plWord-Net database mostly originate from the order in which editors introduced them into the database.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "plWordNet properties", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "Taking as a starting point the work of K\u0119dzia et al. (2015) and the observations in the previous section, we explored several ways of using the knowledge present in plWordNet to improve WSD performance.", | |
| "cite_spans": [ | |
| { | |
| "start": 39, | |
| "end": 59, | |
| "text": "K\u0119dzia et al. (2015)", | |
| "ref_id": "BIBREF10" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Exploring plWordNet in WSD", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "As the number of glosses and examples has been increased in the version 2.3 of plWordNet 3 we can apply Lesk's algorithm in a straightforward wayfurther on called basic Lesk's:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Glosses and Examples", | |
| "sec_num": "4.1" | |
| }, | |
| { | |
| "text": "1. For a word w to be disambiguated, we select all synsets s i that include LUs with lemma identical to the lemma of w. 3. For each occurrence of w a context set C(w) is collected, such that it contains all lemmas from the fixed size context of the w occurrence.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Glosses and Examples", | |
| "sec_num": "4.1" | |
| }, | |
| { | |
| "text": "4. s i such that the set D(s i ) that have the maximal intersection with C(w) is selected as the sense of the given occurrence of w.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Glosses and Examples", | |
| "sec_num": "4.1" | |
| }, | |
| { | |
| "text": "The results obtained with the basic Lesk's algorithm are presented in Table 5 .", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 70, | |
| "end": 77, | |
| "text": "Table 5", | |
| "ref_id": "TABREF7" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Glosses and Examples", | |
| "sec_num": "4.1" | |
| }, | |
| { | |
| "text": "In all experiments presented in (K\u0119dzia et al., 2015) the wordnet graph was treated as a direct but uniform graph, i.e. every relation link was represented in the same way independent of the relation type. In order to increase the density of the graph LU relations were mapped on the synset level, i.e. if there was a link between LUs, then a link between their synsets was added. However, different relations represent different types of semantic association and provide different descriptions for the elements (synsets or LUs) they are attached to. On the basis of preliminary experiments, we assumed that synset relations and LU relations convey information of different importance for WSD and we assigned different weights to both types of links: w LU = 0.3 for LU relations and w S = 0.7 for synset relation 4 . The assigned weights can be next used in the spreading activation algorithm.", | |
| "cite_spans": [ | |
| { | |
| "start": 32, | |
| "end": 53, | |
| "text": "(K\u0119dzia et al., 2015)", | |
| "ref_id": "BIBREF10" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Structural Description", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "In the case of highly polysemous words, some word senses located close to each other in the word graph are difficult to be distinguished. However, for practical applications, sometimes there is no need to differentiate such closely related word senses. So, we also tested partial WSD in which the top-ranked LUs within the range of k = 30% of the maximal score from the WSD algorithm were selected as a joint result. In a natural way, this relaxation of the task resulted in significantly improved precision. It is well known that the most frequent sense baseline is difficult to be beaten by WSD. This is due to the mostly skewed distribution of word senses, in which one or few senses dominate among occurrences. Having LUs ordered according to their frequency in plWordNet, we could use this information to boots WSD performance. However, both Polish corpora annotated with word senses are much too small to provide such data. Regardless, LUs are numbered in plWordNet according to the order in which they have been added for the given lemma. The detailed guidelines for plWordNet editors say nothing about the order in which LUs should be defined 5 , and our null hypothesis was that this would be almost a random factor from the point of WSD, i.e. the use of this information should not have any positive effect on the WSD performance. Nevertheless, we suspected that the null hypothesis does not match the data and that the order of LUs identifiers is not accidental. We assumed that LUs with the highest identifiers represent the most salient senses of lemmas. Thus, selecting them should bring us closer to selecting the most frequent sense.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Sense order", | |
| "sec_num": "4.3" | |
| }, | |
| { | |
| "text": "The relatively good results, presented in Section 5, seem to be in favour of rejecting the null hypothesis. They give some insights into the work of plWordNet editors, see Section 5.2.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Sense order", | |
| "sec_num": "4.3" | |
| }, | |
| { | |
| "text": "Evaluation was based on applying the analysed algorithms to a corpus with manually disambiguated LUs (word senses). As a main criterion for evaluation we used the precision, calculated by comparing the LUs assigned by annotators and the algorithms, see Equation 3:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Results and evaluation", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "P r = t t + f", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Results and evaluation", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "(3)", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Results and evaluation", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "\u2022 t: the number of correctly disambiguated instances,", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Results and evaluation", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "\u2022 f : the number of incorrectly disambiguated instances.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Results and evaluation", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "Two corpora including disambiguated assignment of LUs to words were used during the evaluation. They have different character and were built by two independent teams but both are based on plWordNet, so that seems to be an interesting opportunity for evaluation. The KPWr corpus (Corpus of the Wroc\u0142aw University of Technology) (Broda et al., 2012) , available under the Creative Commons license, 6 contains 1,127 documents (\u2248250,000 tokens) divided into 11 thematic categories. KPWr has been manually annotated and disambiguated at several levels: morpho-syntactic, syntactic relations, semantic relations, Named Entities. The documents are also described with manually assigned keywords and meta-information, like genre, author, etc.", | |
| "cite_spans": [ | |
| { | |
| "start": 327, | |
| "end": 347, | |
| "text": "(Broda et al., 2012)", | |
| "ref_id": "BIBREF3" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Experimental settings", | |
| "sec_num": "5.1" | |
| }, | |
| { | |
| "text": "In the case of 88 different lemmas, all their occurrences have been manually described with LUs from plWordNet by two annotators plus a superannotator, who was responsible for solving conflicts. In the case of all lemmas annotated, their descriptions in plWordNet have been verified according to the defined set of LUs and the information provided for them, i.e. relation links, glosses and usage examples. In the case of lacking LUs (missing word senses), they have been added. If for some LU of one of the 88 lemmas there was no usage examples in KPWr or the number was very small, KPWr was expanded with some new texts. The WSD part of KPWr has been built in two stages, and in the second stage all previous annotations have been verified.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Experimental settings", | |
| "sec_num": "5.1" | |
| }, | |
| { | |
| "text": "The WSD lemma set includes 58 different nouns and 30 verbs, see the statistics in Table 1 . The lemmas were not selected randomly, but were chosen by linguists in such a way that all the lemmas are polysemous and represent different types of homonymy and polysemy. Moreover they vary according to numbers of possible lexical meanings, i.e. possible LUs. From the very beginning this set of WSD annotations was meant to be a gold standard for the evaluation of WSD methods. For 58 nouns and 30 verbs, the average number of word senses per word are 5.98 and 7.50 respectively. The standard deviation is 4.30 for nouns and 3.96 for verbs. The median of number of senses for the nouns is 5; 4 nouns have the number of senses equal to the median. 28 nouns have more senses than the median, and 26 have fewer. The median number of senses for the verbs is 6; 5 verbs have a number of senses equal to the median. 12 verbs have fewer senses than the median, and 13 have more. Thus, the annotated words are quite diversified and challenging for WSD.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 82, | |
| "end": 89, | |
| "text": "Table 1", | |
| "ref_id": "TABREF1" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Experimental settings", | |
| "sec_num": "5.1" | |
| }, | |
| { | |
| "text": "Sk\u0142adnica (Hajnicz, 2014a) , a treebank of Polish, is the second test set used during the evaluation. It includes 20,000 sentences among which more than 8,200 have manually assigned parse trees. For all these sentences, nouns, verbs and adjectives occurring in them have been manually mapped to LUs from plWordNet 1.6 (Hajnicz, 2014b) . Proper Names included in them have been marked and semantically classified. Lemmas or word senses not found in plWordNet have been marked. Sk\u0142adnica includes sentences randomly selected from the open part of NKPJ (National Corpus of Polish) (Przepi\u00f3rkowski et al., 2009 ). All sentences are described by identifiers and links to the original paragraphs, so it is possible to use the whole paragraphs as contexts for WSD. Sk\u0142adnica differs significantly from KPWr with respect to words disambiguated with word senses: the selection was made at the level of sentences, so in the case of most lemmas only selected senses are covered. In KPWr all senses of every selected word are represented. Moreover, the KPWr builders paid attention to acquiring as many usage examples as possible for every senses, including those that are infrequent. WSD annotations in Sk\u0142adnica has been provided not only for polysemous words, but also for monosemous -in Table 2 the column MN contains statistics for monosemous nouns, PN for polysemous nouns, MV for monosemous verbs, PV polysemous verbs.", | |
| "cite_spans": [ | |
| { | |
| "start": 10, | |
| "end": 26, | |
| "text": "(Hajnicz, 2014a)", | |
| "ref_id": "BIBREF5" | |
| }, | |
| { | |
| "start": 318, | |
| "end": 334, | |
| "text": "(Hajnicz, 2014b)", | |
| "ref_id": "BIBREF6" | |
| }, | |
| { | |
| "start": 578, | |
| "end": 606, | |
| "text": "(Przepi\u00f3rkowski et al., 2009", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 1279, | |
| "end": 1286, | |
| "text": "Table 2", | |
| "ref_id": "TABREF3" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Experimental settings", | |
| "sec_num": "5.1" | |
| }, | |
| { | |
| "text": "As a baseline, we repeated experiments from (K\u0119dzia et al., 2015) using plWordNet 2.2 as originally, but also version 2.3 as a basis for the WSD algorithm. All tests were performed on KPWr ; the results are shown in Table 3 . The columns grouped under the label PPR include results achieved by the application of the Personalized PageRank algorithm, while the joint label Static signals the application of Static PageRank. The description of the tested combinations (algorithm parameters and the wordnet version) could make the table too large, so the combinations have been encoded as follows:", | |
| "cite_spans": [ | |
| { | |
| "start": 44, | |
| "end": 65, | |
| "text": "(K\u0119dzia et al., 2015)", | |
| "ref_id": "BIBREF10" | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 216, | |
| "end": 223, | |
| "text": "Table 3", | |
| "ref_id": "TABREF5" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Baseline PageRank approaches", | |
| "sec_num": "5.2.1" | |
| }, | |
| { | |
| "text": "C1 the results achieved on plWordNet 2.2, Table 4 : Precision of disambiguation achieved on KPWr and Sk\u0142adnica.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 42, | |
| "end": 49, | |
| "text": "Table 4", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Baseline PageRank approaches", | |
| "sec_num": "5.2.1" | |
| }, | |
| { | |
| "text": "PPR Static V N All V N All", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Baseline PageRank approaches", | |
| "sec_num": "5.2.1" | |
| }, | |
| { | |
| "text": "C2 as above, but for plWordNet 2.3, C3 and C4 the results achieved on plWordNet versions 2.2 and 2.3, respectively, merged with the SUMO ontology; in both only nodes belonging to plWordNet are initialised (i.e. receive non-zero values in the initial vector).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Baseline PageRank approaches", | |
| "sec_num": "5.2.1" | |
| }, | |
| { | |
| "text": "In Table 3 we can observe that the increasing size of plWordNet affects positively the precision when the same configuration of the algorithm is applied. This effect can be caused by the increasing number of text words covered by the wordnet that results in the increasing number of initially activated nodes in the PR graph. Moreover, in plWordNet 2.3 the number of adjectives and relation links between adjectives and nouns have been increased significantly. Thus cross-categorial connections have been improved, facilitating the activation flow in PR-based algorithms.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 3, | |
| "end": 10, | |
| "text": "Table 3", | |
| "ref_id": "TABREF5" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Baseline PageRank approaches", | |
| "sec_num": "5.2.1" | |
| }, | |
| { | |
| "text": "Next, we performed similar tests but using both data sets, i.e. KPWr and Sk\u0142adnica. Once again algorithms and parameters from (K\u0119dzia et al., 2015) were applied, but this time we concentrated only on plWordNet 2.3. This resulted in better precision in the experiments presented above. Table 4 contains the results achieved for the following configuration of the algorithms: clearly boosted by the monosemous words, while monosemous words are not annotated KPWr. However this influence is too small to be the only reason for the difference, e.g. in Tab. 6 in the case of Sk\u0142adnica only polysemous words were evaluated, i.e. for polysemous and monosemous words the precision of C9 is: 69.08% for nouns, 53.86% for verbs and 63.46% for all. The higher precision on Sk\u0142adnica can be also caused by the different way of selecting words for WSD annotation. In Sk\u0142adnica they come from the running text and we can expect some bias towards most frequent LUs (word senses), while the authors of KPWr tried to cover in WSD annotation all LUs for the selected lemmas, so less frequent LUs received more occurrences than we could expect in a text sample. Tests on KPWr illustrate the ability of the algorithms to distinguish between all possible senses, while tests on Sk\u0142adnica are a better picture of average precision we can expect in practical applications (especially when monosemous words are included in the result).", | |
| "cite_spans": [ | |
| { | |
| "start": 126, | |
| "end": 147, | |
| "text": "(K\u0119dzia et al., 2015)", | |
| "ref_id": "BIBREF10" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Baseline PageRank approaches", | |
| "sec_num": "5.2.1" | |
| }, | |
| { | |
| "text": "C5 Static", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Baseline PageRank approaches", | |
| "sec_num": "5.2.1" | |
| }, | |
| { | |
| "text": "The results of the simple Lesk's algorithm based on plWordNet 2.3 run on both corpora are presented in Tab. 5, where the precision is given for verbs and nouns in percentage points. This algorithm can be treated as the second baselines. The results illustrate the amount of disambiguating information included in the textual descriptions of plWordNet. They are much lower than obtained by PageRank-based algorithms, that explore the rich structure of plWordNet relations", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Glosses and Examples", | |
| "sec_num": "5.2.2" | |
| }, | |
| { | |
| "text": "Tab. 6 presents a comparison of the best baseline configuration for KPWr, namely C8 with the ap- 38.57 43.20 41.62 48.77 61.74 56.69 C11 39.76 39.30 39.46 49.28 61.12 56.51 Table 7 : PageRank-based WSD algorithms supported by re-ranking based on the synset order in plWordNet.", | |
| "cite_spans": [ | |
| { | |
| "start": 97, | |
| "end": 172, | |
| "text": "38.57 43.20 41.62 48.77 61.74 56.69 C11 39.76 39.30 39.46 49.28 61.12 56.51", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 173, | |
| "end": 180, | |
| "text": "Table 7", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Structural Description", | |
| "sec_num": "5.2.3" | |
| }, | |
| { | |
| "text": "KPWr Sk\u0142adnica V N All V N All C10", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Structural Description", | |
| "sec_num": "5.2.3" | |
| }, | |
| { | |
| "text": "proach using the information about the relation types called C9. In C9 Static algorithm based on plWordNet 2.3 was used, but synset relations were assigned weights equal to 0.7 and LU relations weights equal to 0.3. Moreover, the top-scoring LUs within the range of 10% from the best score (according to the WSD algorithm) are re-ranked according to their order (i.e. their identifiers) in the plWordNet database. The re-ranking is limited to those cases in which the values from WSD are very close and the differences can be insignificant.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Structural Description", | |
| "sec_num": "5.2.3" | |
| }, | |
| { | |
| "text": "On KPWr, the use of weighting gave improvement only for verbs. Verbs have a higher ratio of LU relations in comparison to synset relations than nouns, so this supports the intuition that synset relations provide more information for WSD. However, a more in-depth analysis of different weights for different relations is needed. Such an optimisation would need larger training-testing WSD data sets. The situation was completely different in tests on Sk\u0142adnica -here in all cases a significant improvement can be observed. It seems that the higher weights for synset relations and synonymy (the weight 1.0) favour the most frequent senses.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Structural Description", | |
| "sec_num": "5.2.3" | |
| }, | |
| { | |
| "text": "Finally, we tested the use of the order of adding LUs to plWordNet for a given lemma as an additional source of knowledge for WSD algorithms. In all cases this knowledge was used for post-reranking. Two configurations were tested: C10 Static algorithm, plWordNet 2.3 synset graph only, WSD results post-processed by reranking of the top highest scored LUs within the range of k = 30% of the maximal score, the re-ranking is based on LUs numbers in plWordNet.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Sense order", | |
| "sec_num": "5.2.4" | |
| }, | |
| { | |
| "text": "C11 Similar to C10, but re-ranking is limited to k = 40% of the maximal score.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Sense order", | |
| "sec_num": "5.2.4" | |
| }, | |
| { | |
| "text": "The results obtained with the help of C10 and C11 are presented in Tab. 7. In comparison to the baselines shown in Tab. 4, we can notice that reranking brought significant improvement in tests on Sk\u0142adnica for both configurations. The situation is different for KPWr. KPWr includes more occurrences of less frequent senses, while Sk\u0142adnica has a bias towards more frequent senses as built on randomly selected sentences. This difference supports our assumptions that LU numbers in plWordNet are correlated with their frequency in corpora. This correlation is next transferred to re-ranking. This observation is important for practical applications. Thus, we guess that the wordnet editors share some notion of the word sense saliency or their frequency. For a new lemma being edited, they seem to add to the plWordNet its more prominent and more frequent senses first. plWordNet 1.6 noun synsets were automatically ordered according to the estimated frequency of the word senses they represent (McCarthy et al., 2004 (McCarthy et al., , 2007 . However, this method is of limited accuracy and all synsets added later (a large number, the majority) were not ordered in this way.", | |
| "cite_spans": [ | |
| { | |
| "start": 994, | |
| "end": 1016, | |
| "text": "(McCarthy et al., 2004", | |
| "ref_id": "BIBREF15" | |
| }, | |
| { | |
| "start": 1017, | |
| "end": 1041, | |
| "text": "(McCarthy et al., , 2007", | |
| "ref_id": "BIBREF16" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Sense order", | |
| "sec_num": "5.2.4" | |
| }, | |
| { | |
| "text": "In Tab. 1 and 2 the analysis of the relation between the re-ranking threshold and precision is presented. In the case of KPWr the best results were obtained for the 10% re-ranking threshold. However, in the case of Sk\u0142adnica the highest results are concentrated around the threshold 30% and decrease beyond it, so scores produced by the WSD algorithm are at least useful in selecting the most likely LUs for a given word.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Sense order", | |
| "sec_num": "5.2.4" | |
| }, | |
| { | |
| "text": "Weakly supervised WSD methods based on plWordNet have slightly lower precision in tests on Polish WSD corpora than similar PWN-based methods. However, plWordNet does not provide glosses for all LUs and the existing glosses are not disambiguated. Instead we looked into utilisation of other features. We showed that except glosses and examples, we can explore relation types by weighting them for the needs of WSD and the order in which LUs have been added to plWordNet. Both resulted in the increased precision of WSD on one of the test corpora -the one that seems to be closer to the practical applications. While the positive influence of the relations weights on PageRank-based WSD algorithm had been expected, the positive influence of the LUs adding order is a surprise, as the wordnet editors were not asked to use any specific order in introducing new LUs into plWordNet. Thus they have to share some idea of the salience or frequency of the individual LUs for the given lemma. This effect may not be visible when we analyse lists of LUs of individual lemmas, but it seems to be the most probable explanation for the results WSD algorithms using this order as a knowledge source. In future work we plan to develop more sophisticated system of weights assigned to relations for WSD and to work on combining different knowledge sources in one complex WSD algorithm.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Conclusions", | |
| "sec_num": "6" | |
| }, | |
| { | |
| "text": "A lemma is a basic morphological form representing a group of word forms that have the same meaning but differ in the values of the morphological categories.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "However, most glosses take the form of short comments that are several words long.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "The highest weight of 1.0 was implicitly assigned to the synonymy relation that was not present in the graph structure but was expressed by synsets. The synsets collected activations from the occurrence of their members in the contexts of disambiguation.5 In fact it would be very difficult to define this in guidelines in a way resulting in consistent decisions of editors.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "http://nlp.pwr.edu.pl/kpwr", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| } | |
| ], | |
| "back_matter": [ | |
| { | |
| "text": "Work supported by the Polish Ministry of Education and Science, Project CLARIN-PL, the European Innovative Economy Programme project POIG.01.01.02-14-013/09, and by the EU's 7FP under grant agreement No. 316097 [ENGINE].", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Acknowledgment", | |
| "sec_num": null | |
| } | |
| ], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "Personalizing PageRank for Word Sense Disambiguation", | |
| "authors": [ | |
| { | |
| "first": "Eneko", | |
| "middle": [], | |
| "last": "Agirre", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Aitor", | |
| "middle": [], | |
| "last": "Soroa", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2009, | |
| "venue": "Proceedings of the 12th Conference of the European Chapter of the Association for Computational Linguistics, EACL '09", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "33--41", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Eneko Agirre and Aitor Soroa. Personaliz- ing PageRank for Word Sense Disambigua- tion. In Proceedings of the 12th Confer- ence of the European Chapter of the Asso- ciation for Computational Linguistics, EACL '09, pages 33-41, Stroudsburg, PA, USA, 2009. Association for Computational Lin- guistics. URL http://dl.acm.org/ citation.cfm?id=1609067.1609070.", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "Random walks for Knowledge-Based Word Sense Disambiguation", | |
| "authors": [ | |
| { | |
| "first": "Eneko", | |
| "middle": [], | |
| "last": "Agirre", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Oier", | |
| "middle": [], | |
| "last": "Lopez De Lacalle", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Aitor", | |
| "middle": [], | |
| "last": "Soroa", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "Computational Linguistics", | |
| "volume": "40", | |
| "issue": "1", | |
| "pages": "57--84", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Eneko Agirre, Oier Lopez de Lacalle, and Aitor Soroa. Random walks for Knowledge-Based Word Sense Disambiguation. Computational Linguistics, 40(1):57-84, 2014.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Towards Word Sense Disambiguation of Polish", | |
| "authors": [ | |
| { | |
| "first": "Dominik", | |
| "middle": [], | |
| "last": "Ba\u015b", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Bartosz", | |
| "middle": [], | |
| "last": "Broda", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Maciej", | |
| "middle": [], | |
| "last": "Piasecki", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2008, | |
| "venue": "Proceedings of the International Multiconference on Computer Science and Information Technology -3rd International Symposium Advances in Artificial Intelligence and Applications (AAIA'08)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "65--71", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Dominik Ba\u015b, Bartosz Broda, and Maciej Pi- asecki. Towards Word Sense Disambiguation of Polish. In Proceedings of the International Multiconference on Computer Science and In- formation Technology -3rd International Sym- posium Advances in Artificial Intelligence and Applications (AAIA'08), pages 65-71, 2008. URL http://www.proceedings2008. imcsit.org/pliks/162.pdf.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "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. KPWr: Towards a free corpus of Polish. In Proceedings of LREC'12, Istanbul, Turkey, 2012. ELRA.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "WordNet: An Electronic Lexical Database (Language, Speech, and Communication)", | |
| "authors": [], | |
| "year": 1998, | |
| "venue": "", | |
| "volume": "026206197", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Christiane Fellbaum, editor. WordNet: An Elec- tronic Lexical Database (Language, Speech, and Communication). The MIT Press, May 1998. ISBN 026206197X.", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "The procedure of lexicosemantic annotation of Sk\u0142adnica treebank", | |
| "authors": [ | |
| { | |
| "first": "El\u017cbieta", | |
| "middle": [], | |
| "last": "Hajnicz", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": ";", | |
| "middle": [], | |
| "last": "", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Khalid", | |
| "middle": [], | |
| "last": "Choukri", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Thierry", | |
| "middle": [], | |
| "last": "Declerck", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Hrafn", | |
| "middle": [], | |
| "last": "Loftsson", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Bente", | |
| "middle": [], | |
| "last": "Maegaard", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Joseph", | |
| "middle": [], | |
| "last": "Mariani", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Asuncion", | |
| "middle": [], | |
| "last": "Moreno", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC'14)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "El\u017cbieta Hajnicz. The procedure of lexico- semantic annotation of Sk\u0142adnica treebank. In Nicoletta Calzolari (Conference Chair), Khalid Choukri, Thierry Declerck, Hrafn Loftsson, Bente Maegaard, Joseph Mariani, Asuncion Moreno, Jan Odijk, and Stelios Piperidis, ed- itors, Proceedings of the Ninth International Conference on Language Resources and Eval- uation (LREC'14), Reykjavik, Iceland, may 2014a. European Language Resources Associ- ation (ELRA). ISBN 978-2-9517408-8-4.", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "Lexico-semantic annotation of sk\u0142adnica treebank by means of PLWN lexical units", | |
| "authors": [ | |
| { | |
| "first": "El\u017cbieta", | |
| "middle": [], | |
| "last": "Hajnicz", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "Proceedings of the 7th International WordNet Conference", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "23--31", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "El\u017cbieta Hajnicz. Lexico-semantic annotation of sk\u0142adnica treebank by means of PLWN lexical units. In Proceedings of the 7th International WordNet Conference, pages 23-31, 2014b.", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "title": "WordNet 2 -a morphologically and semantically enhanced resource", | |
| "authors": [ | |
| { | |
| "first": "Sanda", | |
| "middle": [ | |
| "M" | |
| ], | |
| "last": "Harabagiu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "George", | |
| "middle": [ | |
| "A" | |
| ], | |
| "last": "Miller", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Dan", | |
| "middle": [ | |
| "I" | |
| ], | |
| "last": "Moldovan", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1999, | |
| "venue": "SIGLEX99: Standardizing Lexical Resources", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "1--8", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Sanda M. Harabagiu, George A. Miller, and Dan I. Moldovan. WordNet 2 -a morphologically and semantically enhanced resource. In SIGLEX99: Standardizing Lexical Resources, pages 1-8, 1999. URL http://www.aclweb.org/ anthology/W99-0501.", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "Ruled-based, interlingual motivated mapping of plWordNet onto SUMO ontology", | |
| "authors": [ | |
| { | |
| "first": "Pawe\u0142", | |
| "middle": [], | |
| "last": "K\u0119dzia", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Maciej", | |
| "middle": [], | |
| "last": "Piasecki", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC'14)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Pawe\u0142 K\u0119dzia and Maciej Piasecki. Ruled-based, interlingual motivated mapping of plWordNet onto SUMO ontology. In Nicoletta Cal- zolari (Conference Chair), Khalid Choukri, Thierry Declerck, Hrafn Loftsson, Bente Mae- gaard, Joseph Mariani, Asuncion Moreno, Jan Odijk, and Stelios Piperidis, editors, Pro- ceedings of the Ninth International Confer- ence on Language Resources and Evalua- tion (LREC'14), Reykjavik, Iceland, may 2014. European Language Resources Associa- tion (ELRA). ISBN 978-2-9517408-8-4.", | |
| "links": null | |
| }, | |
| "BIBREF9": { | |
| "ref_id": "b9", | |
| "title": "Distributionally extended network-based Word Sense Disambiguation in semantic clustering of Polish texts", | |
| "authors": [ | |
| { | |
| "first": "Pawe\u0142", | |
| "middle": [], | |
| "last": "K\u0119dzia", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Maciej", | |
| "middle": [], | |
| "last": "Piasecki", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jan", | |
| "middle": [], | |
| "last": "Koco\u0144", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Agnieszka", | |
| "middle": [], | |
| "last": "Indyka-Piasecka", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "IERI Procedia", | |
| "volume": "10", | |
| "issue": "", | |
| "pages": "38--44", | |
| "other_ids": { | |
| "DOI": [ | |
| "10.1016/j.ieri.2014.09.073" | |
| ] | |
| }, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Pawe\u0142 K\u0119dzia, Maciej Piasecki, Jan Koco\u0144, and Agnieszka Indyka-Piasecka. Distributionally extended network-based Word Sense Disam- biguation in semantic clustering of Polish texts. IERI Procedia, 10(Complete):38-44, 2014. doi: 10.1016/j.ieri.2014.09.073.", | |
| "links": null | |
| }, | |
| "BIBREF10": { | |
| "ref_id": "b10", | |
| "title": "Orli\u0144ska. Word sense disambiguation based on large scale Polish CLARIN heterogeneous lexical resources", | |
| "authors": [ | |
| { | |
| "first": "Pawe\u0142", | |
| "middle": [], | |
| "last": "K\u0119dzia", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Maciej", | |
| "middle": [], | |
| "last": "Piasecki", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Marlena", | |
| "middle": [ | |
| "J" | |
| ], | |
| "last": "", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2015, | |
| "venue": "Cognitive Studies", | |
| "volume": "14", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Pawe\u0142 K\u0119dzia, Maciej Piasecki, and Marlena J. Or- li\u0144ska. Word sense disambiguation based on large scale Polish CLARIN heterogeneous lexi- cal resources. Cognitive Studies, 14(To appear), 2015.", | |
| "links": null | |
| }, | |
| "BIBREF11": { | |
| "ref_id": "b11", | |
| "title": "Automatic sense disambiguation using machine readable dictionaries: how to tell a pine cone from an ice cream cone", | |
| "authors": [ | |
| { | |
| "first": "Michael", | |
| "middle": [], | |
| "last": "Lesk", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1986, | |
| "venue": "Proceedings SIGDOC '86 Proceedings of the 5th annual international conference on Systems documentation", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "24--26", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Michael Lesk. Automatic sense disambiguation using machine readable dictionaries: how to tell a pine cone from an ice cream cone. In Proceed- ings SIGDOC '86 Proceedings of the 5th annual international conference on Systems documen- tation, pages 24-26. ACM, 1986.", | |
| "links": null | |
| }, | |
| "BIBREF12": { | |
| "ref_id": "b12", | |
| "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 Szpakowicz. Approaching plWordNet 2.0. In Proceedings of the 6th Global Wordnet Confer- ence, Matsue, Japan, January 2012.", | |
| "links": null | |
| }, | |
| "BIBREF13": { | |
| "ref_id": "b13", | |
| "title": "Beyond the transfer-and-merge wordnet construction: plWordNet and a comparison with Word-Net", | |
| "authors": [ | |
| { | |
| "first": "Marek", | |
| "middle": [], | |
| "last": "Maziarz", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Maciej", | |
| "middle": [], | |
| "last": "Piasecki", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ewa", | |
| "middle": [], | |
| "last": "Rudnicka", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Stanis\u0142aw", | |
| "middle": [], | |
| "last": "Szpakowicz", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2013, | |
| "venue": "Recent Advances in Natural Language Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "443--452", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Marek Maziarz, Maciej Piasecki, Ewa Rud- nicka, and Stanis\u0142aw Szpakowicz. Beyond the transfer-and-merge wordnet construction: plWordNet and a comparison with Word- Net. In Recent Advances in Natural Language Processing, RANLP 2013, 9-11 September, 2013, Hissar, Bulgaria, pages 443-452, 2013a. URL http://aclweb.org/anthology/ R/R13/R13-1058.pdf.", | |
| "links": null | |
| }, | |
| "BIBREF14": { | |
| "ref_id": "b14", | |
| "title": "The chicken-and-egg problem in wordnet design: Synonymy, synsets and constitutive relations. Language Resources and Evaluation", | |
| "authors": [ | |
| { | |
| "first": "Marek", | |
| "middle": [], | |
| "last": "Maziarz", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Maciej", | |
| "middle": [], | |
| "last": "Piasecki", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Stanis\u0142aw", | |
| "middle": [], | |
| "last": "Szpakowicz", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2013, | |
| "venue": "", | |
| "volume": "47", | |
| "issue": "", | |
| "pages": "769--796", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Marek Maziarz, Maciej Piasecki, and Stanis\u0142aw Szpakowicz. The chicken-and-egg problem in wordnet design: Synonymy, synsets and consti- tutive relations. Language Resources and Eval- uation, 47(3):769-796, 2013b. doi: 10.1007/ s10579-012-9209-9.", | |
| "links": null | |
| }, | |
| "BIBREF15": { | |
| "ref_id": "b15", | |
| "title": "Finding predominant word senses in untagged text", | |
| "authors": [ | |
| { | |
| "first": "Diana", | |
| "middle": [], | |
| "last": "Mccarthy", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Rob", | |
| "middle": [], | |
| "last": "Koeling", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Julie", | |
| "middle": [], | |
| "last": "Weeds", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "John", | |
| "middle": [], | |
| "last": "Carroll", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2004, | |
| "venue": "Proceedings of the 42Nd Annual Meeting on Association for Computational Linguistics, ACL '04", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": { | |
| "DOI": [ | |
| "10.3115/1218955.1218991" | |
| ] | |
| }, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Diana McCarthy, Rob Koeling, Julie Weeds, and John Carroll. Finding predominant word senses in untagged text. In Proceedings of the 42Nd Annual Meeting on Association for Computational Linguistics, ACL '04, Strouds- burg, PA, USA, 2004. Association for Com- putational Linguistics. doi: 10.3115/1218955. 1218991. URL http://dx.doi.org/10. 3115/1218955.1218991.", | |
| "links": null | |
| }, | |
| "BIBREF16": { | |
| "ref_id": "b16", | |
| "title": "Unsupervised acquisition of predominant word senses", | |
| "authors": [ | |
| { | |
| "first": "Diana", | |
| "middle": [], | |
| "last": "Mccarthy", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Rob", | |
| "middle": [], | |
| "last": "Koeling", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Julie", | |
| "middle": [], | |
| "last": "Weeds", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "John", | |
| "middle": [], | |
| "last": "Carroll", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2007, | |
| "venue": "Comput. Linguist", | |
| "volume": "33", | |
| "issue": "4", | |
| "pages": "553--590", | |
| "other_ids": { | |
| "DOI": [ | |
| "10.1162/coli.2007.33.4.553" | |
| ] | |
| }, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Diana McCarthy, Rob Koeling, Julie Weeds, and John Carroll. Unsupervised acquisi- tion of predominant word senses. Com- put. Linguist., 33(4):553-590, December 2007. ISSN 0891-2017. doi: 10.1162/coli.2007.33. 4.553. URL http://dx.doi.org/10. 1162/coli.2007.33.4.553.", | |
| "links": null | |
| }, | |
| "BIBREF17": { | |
| "ref_id": "b17", | |
| "title": "PageRank on semantic networks, with application to Word Sense Disambiguation", | |
| "authors": [ | |
| { | |
| "first": "Rada", | |
| "middle": [], | |
| "last": "Mihalcea", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Paul", | |
| "middle": [], | |
| "last": "Tarau", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Elizabeth", | |
| "middle": [], | |
| "last": "Figa", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2004, | |
| "venue": "Proceedings of the 20th International Conference on Computational Linguistics, COLING '04", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": { | |
| "DOI": [ | |
| "10.3115/1220355.1220517" | |
| ] | |
| }, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Rada Mihalcea, Paul Tarau, and Elizabeth Figa. PageRank on semantic networks, with appli- cation to Word Sense Disambiguation. In Proceedings of the 20th International Confer- ence on Computational Linguistics, COLING '04, Stroudsburg, PA, USA, 2004. Association for Computational Linguistics. doi: 10.3115/ 1220355.1220517. URL http://dx.doi. org/10.3115/1220355.1220517.", | |
| "links": null | |
| }, | |
| "BIBREF18": { | |
| "ref_id": "b18", | |
| "title": "Association for Computational Linguistics", | |
| "authors": [ | |
| { | |
| "first": "George", | |
| "middle": [ | |
| "A" | |
| ], | |
| "last": "Miller", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Claudia", | |
| "middle": [], | |
| "last": "Leacock", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Randee", | |
| "middle": [], | |
| "last": "Tengi", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ross", | |
| "middle": [ | |
| "T" | |
| ], | |
| "last": "Bunker", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1993, | |
| "venue": "Proceedings of the Workshop on Human Language Technology, HLT '93", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "303--308", | |
| "other_ids": { | |
| "DOI": [ | |
| "10.3115/1075671.1075742" | |
| ] | |
| }, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "George A. Miller, Claudia Leacock, Randee Tengi, and Ross T. Bunker. A semantic concordance. In Proceedings of the Work- shop on Human Language Technology, HLT '93, pages 303-308, Stroudsburg, PA, USA, 1993. Association for Computational Linguis- tics. ISBN 1-55860-324-7. doi: 10.3115/ 1075671.1075742. URL http://dx.doi. org/10.3115/1075671.1075742.", | |
| "links": null | |
| }, | |
| "BIBREF19": { | |
| "ref_id": "b19", | |
| "title": "The WSD development environment", | |
| "authors": [ | |
| { | |
| "first": "Rafa\u0142", | |
| "middle": [], | |
| "last": "M\u0142odzki", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Adam", | |
| "middle": [], | |
| "last": "Przepi\u00f3rkowski", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": null, | |
| "venue": "Lecture Notes in Computer Science", | |
| "volume": "6562", | |
| "issue": "", | |
| "pages": "224--233", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Rafa\u0142 M\u0142odzki and Adam Przepi\u00f3rkowski. The WSD development environment. In Zygmunt Vetulani, editor, LTC, volume 6562 of Lecture Notes in Computer Science, pages 224-233.", | |
| "links": null | |
| }, | |
| "BIBREF21": { | |
| "ref_id": "b21", | |
| "title": "The PageRank citation ranking: Bringing order to the Web", | |
| "authors": [ | |
| { | |
| "first": "Lawrence", | |
| "middle": [], | |
| "last": "Page", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Sergey", | |
| "middle": [], | |
| "last": "Brin", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Rajeev", | |
| "middle": [], | |
| "last": "Motwani", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Terry", | |
| "middle": [], | |
| "last": "Winograd", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1999, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Lawrence Page, Sergey Brin, Rajeev Motwani, and Terry Winograd. The PageRank citation ranking: Bringing order to the Web, 1999.", | |
| "links": null | |
| }, | |
| "BIBREF22": { | |
| "ref_id": "b22", | |
| "title": "Clustering by Committee", | |
| "authors": [ | |
| { | |
| "first": "Patrick", | |
| "middle": [ | |
| "A" | |
| ], | |
| "last": "Pantel", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2003, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Patrick A. Pantel. Clustering by Committee. PhD thesis, University of Alberta Edmonton, Alta., Canada, 2003.", | |
| "links": null | |
| }, | |
| "BIBREF23": { | |
| "ref_id": "b23", | |
| "title": "Ontology: A Practical Guide", | |
| "authors": [ | |
| { | |
| "first": "Adam", | |
| "middle": [], | |
| "last": "Pease", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2011, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Adam Pease. Ontology: A Practical Guide. 2011.", | |
| "links": null | |
| }, | |
| "BIBREF24": { | |
| "ref_id": "b24", | |
| "title": "A Wordnet from the Ground up. Oficyna Wydawnicza Politechniki Wroclawskiej", | |
| "authors": [ | |
| { | |
| "first": "Maciej", | |
| "middle": [], | |
| "last": "Piasecki", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Stanis\u0142aw", | |
| "middle": [], | |
| "last": "Szpakowicz", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Bartosz", | |
| "middle": [], | |
| "last": "Broda", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2009, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Maciej Piasecki, Stanis\u0142aw Szpakowicz, and Bar- tosz Broda. A Wordnet from the Ground up. Ofi- cyna Wydawnicza Politechniki Wroclawskiej, 2009.", | |
| "links": null | |
| }, | |
| "BIBREF25": { | |
| "ref_id": "b25", | |
| "title": "Introduction to the special issue: On wordnets and relations", | |
| "authors": [ | |
| { | |
| "first": "Maciej", | |
| "middle": [], | |
| "last": "Piasecki", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Stan", | |
| "middle": [], | |
| "last": "Szpakowicz", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Christiane", | |
| "middle": [], | |
| "last": "Fellbaum", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Bolette", | |
| "middle": [], | |
| "last": "Sandford Pedersen", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2013, | |
| "venue": "Language Resources and Evaluation", | |
| "volume": "47", | |
| "issue": "3", | |
| "pages": "757--767", | |
| "other_ids": { | |
| "DOI": [ | |
| "10.1007/s10579-013-9247-y" | |
| ] | |
| }, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Maciej Piasecki, Stan Szpakowicz, Christiane Fellbaum, and Bolette Sandford Pedersen. Introduction to the special issue: On word- nets and relations. Language Resources and Evaluation, 47(3):757-767, 2013. ISSN 1574-020X. doi: 10.1007/s10579-013-9247-y. URL http://dx.doi.org/10.1007/ s10579-013-9247-y.", | |
| "links": null | |
| }, | |
| "BIBREF27": { | |
| "ref_id": "b27", | |
| "title": "Exploiting domain information for Word Sense Disambiguation of medical documents", | |
| "authors": [ | |
| { | |
| "first": "Mark", | |
| "middle": [], | |
| "last": "Stevenson", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Eneko", | |
| "middle": [], | |
| "last": "Agirre", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Aitor", | |
| "middle": [], | |
| "last": "Soroa", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2012, | |
| "venue": "JAMIA", | |
| "volume": "19", | |
| "issue": "2", | |
| "pages": "235--240", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Mark Stevenson, Eneko Agirre, and Aitor Soroa. Exploiting domain information for Word Sense Disambiguation of medical documents. JAMIA, 19(2):235-240, 2012. URL http://dblp. uni-trier.de/db/journals/jamia/ jamia19.html#StevensonAS12.", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "FIGREF0": { | |
| "type_str": "figure", | |
| "num": null, | |
| "text": "Description sets D(s i ) encompass all lemmas that are included in glosses and examples describing LUs from s i , as well lemmas from s i .", | |
| "uris": null | |
| }, | |
| "FIGREF1": { | |
| "type_str": "figure", | |
| "num": null, | |
| "text": "Influence of ranking % on precision evaluated on KPWr with Static and PPR.", | |
| "uris": null | |
| }, | |
| "FIGREF2": { | |
| "type_str": "figure", | |
| "num": null, | |
| "text": "Influence of ranking % on precision evaluated on Sk\u0142adnica with Static and PPR.", | |
| "uris": null | |
| }, | |
| "TABREF1": { | |
| "type_str": "table", | |
| "html": null, | |
| "content": "<table/>", | |
| "text": "", | |
| "num": null | |
| }, | |
| "TABREF3": { | |
| "type_str": "table", | |
| "html": null, | |
| "content": "<table/>", | |
| "text": "Statistics of WSD annotations in Sk\u0142adnica.", | |
| "num": null | |
| }, | |
| "TABREF4": { | |
| "type_str": "table", | |
| "html": null, | |
| "content": "<table/>", | |
| "text": "C1 28.64 47.25 40.45 28.14 43 37.57 C2 33.70 50.23 44.58 34.11 44.17 40.73 C3 29.57 48.06 37.57 29.79 42.79 38.05 C4 32.61 52.22 45.52 32.19 44.63 40.38", | |
| "num": null | |
| }, | |
| "TABREF5": { | |
| "type_str": "table", | |
| "html": null, | |
| "content": "<table><tr><td/><td>KPWr</td><td/><td/><td>Sk\u0142adnica</td><td/></tr><tr><td>V</td><td>N</td><td>All</td><td>V</td><td>N</td><td>All</td></tr><tr><td>C5</td><td/><td/><td/><td/><td/></tr></table>", | |
| "text": "Comparison of disambiguation precision using PLWN 2.2 and PLWN 2.3 evaluated on KPWr 34.11 44.17 40.73 47.08 57.37 53.37 C6 33.70 50.23 44.58 42.05 54.15 49.44 C7 32.19 44.63 40.38 47.00 57.97 53.70 C8 32.61 52.22 45.52 41.99 55.40 50.17", | |
| "num": null | |
| }, | |
| "TABREF6": { | |
| "type_str": "table", | |
| "html": null, | |
| "content": "<table><tr><td>V Lesk 16</td><td>algorithm, only plWordNet 2.3 synset graph used, C6 PPR algorithm, only plWordNet 2.3 synsets, C7 Static algorithm, plWordNet 2.3 synset graph merged with SUMO ontology, but only nodes from plWordNet are initialised, C8 PPR algorithm, as above, plWordNet 2.3 synset graph merged with SUMO ontology, but only nodes from plWordNet are ini-tialised for disambiguation. Results on KPWr Sk\u0142adnica N All V N All</td></tr></table>", | |
| "text": "Sk\u0142adnica are higher and close to the results obtained for English. The precision is .80 18.80 18.12 39.34 38.56 38.87", | |
| "num": null | |
| }, | |
| "TABREF7": { | |
| "type_str": "table", | |
| "html": null, | |
| "content": "<table><tr><td/><td>KPWr</td><td/><td/><td>Sk\u0142adnica</td><td/></tr><tr><td>V</td><td>N</td><td>All</td><td>V</td><td>N</td><td>All</td></tr><tr><td colspan=\"6\">C8 32.61 52.22 45.52 49.02 64.02 58.48</td></tr><tr><td colspan=\"6\">C9 42.66 47.91 46.12 47.51 61.67 56.16</td></tr></table>", | |
| "text": "Simple Lesk algorithm run on KPWr and Sk\u0142adnica", | |
| "num": null | |
| }, | |
| "TABREF8": { | |
| "type_str": "table", | |
| "html": null, | |
| "content": "<table/>", | |
| "text": "Static PageRank WSD algorithm based on the weighted plWordNet graph (C9) in comparison to the PPR algorithm.", | |
| "num": null | |
| } | |
| } | |
| } | |
| } |