Ongoing work (in mid 2016)

Transcript

1. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Automatic scansion of poetry Manex Agirrezabal Advisors: Mans Hulden, I˜ naki Alegria eta Bertol Arrieta May 31, 2016 1 / 51
2. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Table of contents 1 Thesis topic 2 Motivation 3 Scansion examples 4 Corpora 5 Techniques 6 Results 7 Discussion and FW 2 / 51
3. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Table of contents 1 Thesis topic 2 Motivation 3 Scansion examples 4 Corpora 5 Techniques 6 Results 7 Discussion and FW 3 / 51
4. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Thesis: redirection At the beginning: Natural Language (poetry) Generation Very hard There are simple solutions, but we developed them in the Master Thesis The gap between these solutions and our main goal was too big Hard to get good results 4 / 51
5. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Thesis: redirection At the beginning: Natural Language (poetry) Generation Very hard There are simple solutions, but we developed them in the Master Thesis The gap between these solutions and our main goal was too big Hard to get good results So, because of that, we decided to change the course of the thesis 4 / 51
6. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Automatic scansion of poetry Manex Agirrezabal Advisors: Mans Hulden, I˜ naki Alegria eta Bertol Arrieta May 31, 2016 5 / 51
7. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Table of contents 1 Thesis topic 2 Motivation 3 Scansion examples 4 Corpora 5 Techniques 6 Results 7 Discussion and FW 6 / 51
8. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW What’s poetry? Poetry Writing that formulates a concentrated imaginative awareness of experience in language chosen and arranged to create a specific emotional response through meaning, sound, and rhythm Merriam-Webster dictionary 7 / 51
9. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW What’s poetry? Poetry Writing that formulates a concentrated imaginative awareness of experience in language chosen and arranged to create a specific emotional response through meaning, sound, and rhythm Merriam-Webster dictionary 8 / 51
10. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW What’s scansion of poetry? Scansion Scansion is the act of determining and graphically representing the metrical character of a line of verse. en.wikipedia.org 9 / 51
11. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW What’s scansion of poetry? Scansion Scansion is the act of determining and graphically representing the metrical character of a line of verse. en.wikipedia.org Different metrical patterns are used around the world 9 / 51
12. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Why scanned poetry? That will help online culture databases organizing the poems according to their meter We can get a better understanding of poems knowing their meter. 10 / 51
13. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Research questions Research questions Which is the mininum knowledge required to analyze a poem? Which are interesting features when analyzing a poem? To what extent do language-specific knowledge contribute when analyzing poetry? Is it possible to analyze a poem without having any information about language? Can we learn to do it? 11 / 51
14. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Contributions Contributions ZeuScansion: A rule-based tool for scansion of english poetry Application of Machine Learning techniques to poetry scansion Conventional ML + CRF Deep Learning Unsupervised Learning Comparable corpus in English, Spanish and Basque 12 / 51
15. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Table of contents 1 Thesis topic 2 Motivation 3 Scansion examples 4 Corpora 5 Techniques 6 Results 7 Discussion and FW 13 / 51
16. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW IAMB [to be] [or not] [to be] ... - ’ - ’ _ ’ 14 / 51
17. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW IAMB [to be] [or not] [to be] ... - ’ - ’ _ ’ TROCHEE [By the], [shores of], [Gitche] [Gumee] ’ - ’ - ’ - ’ - [By the] [shining] [big-sea-] [water] ’ - ’ - ’ - ’ - 14 / 51
18. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW IAMB [to be] [or not] [to be] ... - ’ - ’ _ ’ TROCHEE [By the], [shores of], [Gitche] [Gumee] ’ - ’ - ’ - ’ - [By the] [shining] [big-sea-] [water] ’ - ’ - ’ - ’ - DACTYL [ Woman much] [missed, how you] [call to me], [call to me] ... ’ _ _ ’ _ _ ’ _ _ ’ _ _ 14 / 51
19. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW 15 / 51
20. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Azkeneko ola-gizona (Orixe) Beti tiriki-tauki - ’ - ’ - ’ - eztarrian kanta, - ’ -’ - ’ ardoz ondo bustita, - ’ - ’ - ’ - etzedin marranta. - ’ - ’ - ’ Dantza-sokarik bazan, an Gabi-arotza; tobera-jotzen, berriz, ez bait-zan arrotza... .... gaberdian gaberdi, eztairikan bazan, gure Ola-gizona antxen izango zan. 16 / 51
21. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Eleg´ ıa I (Garcilaso de la Vega) Y luego con gracioso movimiento - ’ - - - ’ - - - ’ - se fue su paso por el verde suelo, - ’ - ’ - - - ’ - ’ - con su guirlanda usada y su ornamento; - - - ’ - - ’ - - - - ’ - desordenaba con lascivo vuelo - - - ’ - - - ’ - ’ - el viento sus cabellos; con su vista - ’ - - - ’ - - - ’ - s’alegraba la tierra, el mar y el cielo. - - ’ - - ’ - - ’ - - ’ - 17 / 51
22. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW English Spanish Basque Table of contents 1 Thesis topic 2 Motivation 3 Scansion examples 4 Corpora English Spanish Basque 5 Techniques 6 Results 7 Discussion and FW 18 / 51
23. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW English Spanish Basque Why corpora? We need corpora 19 / 51
24. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW English Spanish Basque Why corpora? We need corpora 1 To train our systems 19 / 51
25. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW English Spanish Basque Why corpora? We need corpora 1 To train our systems 2 To evaluate our systems 19 / 51
26. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW English Spanish Basque Why corpora? We need corpora 1 To train our systems 2 To evaluate our systems English Spanish Basque 19 / 51
27. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW English Spanish Basque For Better For Verse (University of Virginia) http://prosody.lib.virginia.edu/ https://github.com/waynegraham/for_better_for_verse/tree/master/poems 20 / 51
28. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW English Spanish Basque Two options: The Spanish Verse Corpus (The structure of verse, Svetlana Bochaver & Dmitri Sitchinava) Corpus de Sonetos Siglo de Oro https: //github.com/bncolorado/CorpusSonetosSigloDeOro 21 / 51
29. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW English Spanish Basque I’m tagging it! 22 / 51
30. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW English Spanish Basque I’m tagging it! Urkizu, P., Poes´ ıa vasca: Antolog´ ıa biling¨ ue 22 / 51
31. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW English Spanish Basque I’m tagging it! Urkizu, P., Poes´ ıa vasca: Antolog´ ıa biling¨ ue 1 Ixa-pipes (tokenization) 22 / 51
32. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW English Spanish Basque I’m tagging it! Urkizu, P., Poes´ ıa vasca: Antolog´ ıa biling¨ ue 1 Ixa-pipes (tokenization) 2 Line group, line, word and syllable division 22 / 51
33. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW English Spanish Basque I’m tagging it! Urkizu, P., Poes´ ıa vasca: Antolog´ ıa biling¨ ue 1 Ixa-pipes (tokenization) 2 Line group, line, word and syllable division We follow TEI P5 guidelines for annotation. 22 / 51
34. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW English Spanish Basque Nire aitaren etxea defendituko dut. Otsoen kontra, sikatearen kontra, lukurreriaren kontra. justiziaren kontra, defenditu eginen dut nire aitaren etxea. Galduko ditut ... 23 / 51
35. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW English Spanish Basque <titleStmt> <title>Nire aitaren etxea</title> <author>Gabriel Aresti</author> </titleStmt> <publicationStmt><date>1976</date></publicationStmt> <sourceDesc default="false"><p/></sourceDesc> </fileDesc> </teiHeader> <text id="POEM_MARKUP"> <body> <lg n="1"> <l n="1" met="" real=""> <!--Nire aitaren etxea--> <seg type="syll" doc="NAF_FILE" targetId="w1">Ni</seg> <seg type="syll" doc="NAF_FILE" targetId="w1">re</seg> <seg type="space"> </seg> <seg type="syll" doc="NAF_FILE" targetId="w2">ai</seg> ... <seg type="syll" doc="NAF_FILE" targetId="w3">a</seg> </l> <l n="2" met="" real=""> <!--defendituko dut.--> ... <l n="36" met="" real=""> <!--zutik.--> <seg type="syll" doc="NAF_FILE" targetId="w112">zu</seg> ... 24 / 51
36. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW English Spanish Basque Call for linguists/volunteers/poetry enthusiasts 25 / 51
37. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW English Spanish Basque <titleStmt> <title>Nire aitaren etxea</title> <author>Gabriel Aresti</author> </titleStmt> <publicationStmt><date>1976</date></publicationStmt> <sourceDesc default="false"><p/></sourceDesc> </fileDesc> </teiHeader> <text id="POEM_MARKUP"> <body> <lg n="1"> <l n="1" met="" real="+--+--+-"> <!--Nire aitaren etxea--> <seg type="syll" doc="NAF_FILE" targetId="w1">Ni</seg> <seg type="syll" doc="NAF_FILE" targetId="w1">re</seg> <seg type="space"> </seg> <seg type="syll" doc="NAF_FILE" targetId="w2">ai</seg> ... <seg type="syll" doc="NAF_FILE" targetId="w3">a</seg> </l> <l n="2" met="" real="-+---+"> <!--defendituko dut.--> ... <l n="36" met="" real="+-"> <!--zutik.--> <seg type="syll" doc="NAF_FILE" targetId="w112">zu</seg> ... 26 / 51
38. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW English Spanish Basque XVI: 2 (10) XVII: 1 (10) XVIII: 1 (10) XIX: 8 (9) XX: 13 (13) Total: 25 (52) 27 / 51
39. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Rule-based scansion Supervised learning Unsupervised learning Table of contents 1 Thesis topic 2 Motivation 3 Scansion examples 4 Corpora 5 Techniques Rule-based scansion Supervised learning Unsupervised learning 6 Results 7 Discussion and FW 28 / 51
40. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Rule-based scansion Supervised learning Unsupervised learning Techniques We have several systems that perform automatic scansion of English poetry (which we want to extend to Spanish & Basque) 29 / 51
41. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Rule-based scansion Supervised learning Unsupervised learning Techniques ZeuScansion: a tool for scansion of English poetry (Agirrezabal et al., 2013) (Agirrezabal et al., 2016) 30 / 51
42. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Rule-based scansion Supervised learning Unsupervised learning Techniques ZeuScansion: a tool for scansion of English poetry (Agirrezabal et al., 2013) (Agirrezabal et al., 2016) Simple heuristics based on Groves (1998). Inference from Lexical stress + POS-tag 30 / 51
43. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Rule-based scansion Supervised learning Unsupervised learning Techniques ZeuScansion: a tool for scansion of English poetry (Agirrezabal et al., 2013) (Agirrezabal et al., 2016) Simple heuristics based on Groves (1998). Inference from Lexical stress + POS-tag The problem of out-of-vocabulary words 30 / 51
44. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Rule-based scansion Supervised learning Unsupervised learning Techniques ZeuScansion: a tool for scansion of English poetry (Agirrezabal et al., 2013) (Agirrezabal et al., 2016) Simple heuristics based on Groves (1998). Inference from Lexical stress + POS-tag The problem of out-of-vocabulary words FST-based Closest Word Finder (Improvements in Agirrezabal et al., (2014)) 30 / 51
45. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Rule-based scansion Supervised learning Unsupervised learning Techniques ZeuScansion: a tool for scansion of English poetry (Agirrezabal et al., 2013) (Agirrezabal et al., 2016) Simple heuristics based on Groves (1998). Inference from Lexical stress + POS-tag The problem of out-of-vocabulary words FST-based Closest Word Finder (Improvements in Agirrezabal et al., (2014)) Global analysis system 30 / 51
46. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Rule-based scansion Supervised learning Unsupervised learning Techniques Single predictors: Naive Bayes Averaged Perceptron Support Vector Machines 31 / 51
47. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Rule-based scansion Supervised learning Unsupervised learning Techniques Single predictors: Naive Bayes Averaged Perceptron Support Vector Machines Sequence-based predictors Hidden Markov Models Conditional Random Fields 31 / 51
48. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Rule-based scansion Supervised learning Unsupervised learning Techniques Single predictors: Naive Bayes Averaged Perceptron Support Vector Machines Sequence-based predictors Hidden Markov Models Conditional Random Fields Neural Networks (Deep Learning) 31 / 51
49. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Rule-based scansion Supervised learning Unsupervised learning Techniques Two main steps: 1 Syllabification 2 Finding patterns among syllables 32 / 51
50. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Baselines + SoA Rule-based scansion Supervised learning Unsupervised scansion Table of contents 1 Thesis topic 2 Motivation 3 Scansion examples 4 Corpora 5 Techniques 6 Results Baselines + SoA Rule-based scansion Supervised learning Unsupervised scansion 7 Discussion and FW 33 / 51
51. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Baselines + SoA Rule-based scansion Supervised learning Unsupervised scansion Baselines Baselines Per syllable accuracy Per line accuracy Always stressed 0.50877 0.00274 Always unstressed 0.49764 0.00000 Lexical stress 0.73574 0.05764 Syllable weight 0.62135 0.00549 Iambic (Based on first) 0.75890 0.11253 Iambic (Based on last) 0.83639 0.33303 Naive Bayes 0.85306 0.269295 Table: Accuracies of baselines 34 / 51
52. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Baselines + SoA Rule-based scansion Supervised learning Unsupervised scansion State of the art Scandroid: Automatic scansion system Per-syllable accuracy: %87.42 Per-line accuracy: %34.49 35 / 51
53. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Baselines + SoA Rule-based scansion Supervised learning Unsupervised scansion ZeuScansion ZeuScansion: We evaluated against the whole corpus, as it is an expert system Program Per syllable accuracy Per line accuracy ZeuScansion 0.86165 0.29369 Scandroid 0.8742 0.3449 Table: Accuracies of rule-based systems 36 / 51
54. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Baselines + SoA Rule-based scansion Supervised learning Unsupervised scansion Feature selection We performed feature selection in three different ways: 37 / 51
55. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Baselines + SoA Rule-based scansion Supervised learning Unsupervised scansion Feature selection We performed feature selection in three different ways: Feature set 1: Ablation study 37 / 51
56. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Baselines + SoA Rule-based scansion Supervised learning Unsupervised scansion Feature selection We performed feature selection in three different ways: Feature set 1: Ablation study Feature set 2: Feature ranking with RELIEF algorithm 37 / 51
57. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Baselines + SoA Rule-based scansion Supervised learning Unsupervised scansion Feature selection We performed feature selection in three different ways: Feature set 1: Ablation study Feature set 2: Feature ranking with RELIEF algorithm Feature set 3: Feature ranking based on SVM weights (Guyon et al., 2002) 37 / 51
58. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Baselines + SoA Rule-based scansion Supervised learning Unsupervised scansion Single prediction results Table: Results using the features selected in the ablation study. Per syllable Per line Naive Bayes 0.82160 0.17019 Linear SVM 0.86970 0.31965 Perceptron 0.85215 0.35596 38 / 51
59. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Baselines + SoA Rule-based scansion Supervised learning Unsupervised scansion Single prediction results Table: Results using the 25 best ranked features (according to RELIEF). Per syllable Per line Naive Bayes 0.85738 0.27936 Linear SVM 0.87764 0.37245 Perceptron 0.87034 0.44310 39 / 51
60. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Baselines + SoA Rule-based scansion Supervised learning Unsupervised scansion Single prediction results Table: Results using the 25 best ranked features (according to SVM weights). Per syllable Per line Naive Bayes 0.85534 0.28811 Linear SVM 0.88129 0.37561 Perceptron 0.86691 0.42632 40 / 51
61. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Baselines + SoA Rule-based scansion Supervised learning Unsupervised scansion Structured prediction results Table: Results of Hidden Markov Models. Per syllable Per line HMM 0.90426 0.49875 41 / 51
62. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Baselines + SoA Rule-based scansion Supervised learning Unsupervised scansion Structured prediction results Table: Results of previous feature sets using Conditional Random Fields. Per syllable Per line Ablation study (13 features) 0.91926 0.57184 RELIEF (25 features) 0.91626 0.55515 (Guyon et al., 2002) (25 features) 0.90914 0.52766 42 / 51
63. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Baselines + SoA Rule-based scansion Supervised learning Unsupervised scansion Comparison of best results Per syllable Per line Iambic (Based on last) 0.83639 0.33303 Naive Bayes 0.85306 0.269295 ZeuScansion 0.86165 0.29369 Scandroid 0.8742 0.3449 Linear SVM (ABLATION STUDY) 0.86970 0.31965 Perceptron (ABLATION STUDY) 0.85215 0.35596 Linear SVM (RELIEF-FEATURES (25)) 0.87764 0.37245 Perceptron (RELIEF-FEATURES (25)) 0.87034 0.44310 Linear SVM (SVM WEIGHT-FEATURES (25)) 0.88129 0.37561 Perceptron (SVM WEIGHT-FEATURES (25)) 0.86691 0.42632 HMM 0.90426 0.49875 CRF-Ablation study (13 features) 0.91926 0.57184 CRF-RELIEF (25 features) 0.91626 0.55515 CRF-SVM WEIGHTS (25 features) 0.90914 0.52766 43 / 51
64. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Baselines + SoA Rule-based scansion Supervised learning Unsupervised scansion Neural Networks ...work in progress... 44 / 51
65. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Baselines + SoA Rule-based scansion Supervised learning Unsupervised scansion Unsupervised poetry analysis Beginning... 45 / 51
66. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Baselines + SoA Rule-based scansion Supervised learning Unsupervised scansion Unsupervised poetry analysis Beginning... 1st trial: 1.- Create an automatic syllabifier in an unsupervised manner 2.- Syllabify poem 3.- Train syllable embeddings 4.- Check the distance between consecutive syllables 45 / 51
67. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Baselines + SoA Rule-based scansion Supervised learning Unsupervised scansion Unsupervised poetry analysis kantuz sortu naiz eta kantuz nahi bizi kantuz igortzen ditut nik penak ihesi kantuz zan dudanian zerbait irabazi kantuz gostura ditut guziak iretsi kantuz ni bezalakoak hiltzia du merezi 46 / 51
68. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Baselines + SoA Rule-based scansion Supervised learning Unsupervised scansion Unsupervised poetry analysis kan.tuz sor.tu na.iz e.ta kan.tuz na.hi bi.zi kan.tuz i.gortzen di.tut nik pe.nak i.he.si kan.tuz zan du.da.nian zer.ba.it i.ra.ba.zi kan.tuz gos.tu.ra di.tut gu.ziak i.ret.si kan.tuz ni be.za.la.ko.ak hiltzia du me.re.zi 47 / 51
69. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Table of contents 1 Thesis topic 2 Motivation 3 Scansion examples 4 Corpora 5 Techniques 6 Results 7 Discussion and FW 48 / 51
70. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Discussion and FW We have done several experiments on poetry scansion in English. Now we are finishing the supervised part of our work and we will include some experiments based on Neural Networks. We should extrapolate these works to other languages (Spanish & Basque). Our final goal is to be able to (at least minimally) analyze poems without knowing a language. 49 / 51
71. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Discussion and FW We have done several experiments on poetry scansion in English. Now we are finishing the supervised part of our work and we will include some experiments based on Neural Networks. We should extrapolate these works to other languages (Spanish & Basque). Our final goal is to be able to (at least minimally) analyze poems without knowing a language. My goals: 49 / 51
72. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Discussion and FW We have done several experiments on poetry scansion in English. Now we are finishing the supervised part of our work and we will include some experiments based on Neural Networks. We should extrapolate these works to other languages (Spanish & Basque). Our final goal is to be able to (at least minimally) analyze poems without knowing a language. My goals: 1.- Finish ressearch 49 / 51
73. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Discussion and FW We have done several experiments on poetry scansion in English. Now we are finishing the supervised part of our work and we will include some experiments based on Neural Networks. We should extrapolate these works to other languages (Spanish & Basque). Our final goal is to be able to (at least minimally) analyze poems without knowing a language. My goals: 1.- Finish ressearch 2.- Write thesis 49 / 51
74. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Discussion and FW We have done several experiments on poetry scansion in English. Now we are finishing the supervised part of our work and we will include some experiments based on Neural Networks. We should extrapolate these works to other languages (Spanish & Basque). Our final goal is to be able to (at least minimally) analyze poems without knowing a language. My goals: 1.- Finish ressearch 2.- Write thesis Just that 49 / 51
75. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Eskerrak Mila esker! Galderarik? 50 / 51
76. Thesis topic Motivation Scansion examples Corpora Techniques Results Discussion and FW Automatic scansion of poetry Manex Agirrezabal Advisors: Mans Hulden, I˜ naki Alegria eta Bertol Arrieta May 31, 2016 51 / 51