June 2, 2014 Institut national des sciences appliquées de Lyon

Size: px

Start display at page:

Download "June 2, 2014 Institut national des sciences appliquées de Lyon"

Melvyn O’Connor’
5 years ago
Views:

1 June 2, 2014 Institut national des sciences appliquées de Lyon

2 Search Searchable corpora: All Perseus Greek and Latin Selection of English texts

3 Latin Latin Search Results

4 How Tesserae Works: Stage 1 - Gathering Stage 1: User selects Texts A and B for comparison. Then: For given units (line of poetry or sentence) the software will: Compare each unit in Text A with each unit in Text B Find all instances where they share two or more matching words

5 How Tesserae Works: Stage 1 - Gathering Caesar, Gallic War Vergil, Aeneid ensis 4.579, 7.523, , moneo , periculum , 3.3.2, , 3.710, Horace, Odes , Ovid, Amores , strictus , volito 7.102, , Martial, Epigrams 1.8.1, , , , , , , Tacitus, Annales (For purposes of illustration. Not all occurrences of each word shown.)

6 How Tesserae Works: Stage 1 - Gathering Caesar, Gallic War Vergil, Aeneid ensis 4.579, 7.523, , moneo , periculum , 3.3.2, , 3.710, Horace, Odes , Ovid, Amores , strictus , volito 7.102, , Martial, Epigrams 1.8.1, , , , , , , Tacitus, Annales (For purposes of illustration. Not all occurrences of each word shown.)

7 How Tesserae Works: Stage 1 - Gathering Caesar, Gallic War Vergil, Aeneid ensis 4.579, 7.523, , moneo , periculum , 3.3.2, , 3.710, Horace, Odes , Ovid, Amores , strictus , volito 7.102, , Martial, Epigrams 1.8.1, , , , , , , Tacitus, Annales (For purposes of illustration. Not all occurrences of each word shown.)

8 How Tesserae Works: Stage 1 - Gathering Caesar, Gallic War Vergil, Aeneid ensis 4.579, 7.523, , moneo , periculum , 3.3.2, , 3.710, Horace, Odes , Ovid, Amores , strictus , volito 7.102, , Martial, Epigrams 1.8.1, , , , , , , Tacitus, Annales (For purposes of illustration. Not all occurrences of each word shown.)

How Tesserae Works: Stage 1 - Gathering Caesar, Gallic War Vergil, Aeneid Horace, Odes Ovid ensis Amores 2.2.63 4.579, 7.523, non scelus adgredimur, non ad 10.

615, quod magni Thraseae 3.3.2, consummatique 3.710, Catonis dogmata sic sequeris, 8.41.6 salvos ut 9.

9 How Tesserae Works: Stage 1 - Gathering Caesar, Gallic War Vergil, Aeneid Horace, Odes Ovid ensis Amores , 7.523, non scelus adgredimur, non ad , miscenda coimus toxica, non stricto fulminat ense manus. moneo , , Martial Epigrams Ovid, Amores , periculum , 1.615, quod magni Thraseae 3.3.2, consummatique 3.710, Catonis dogmata sic sequeris, salvos ut esse velis, pectore nec nudo strictos incurris in ensis, quod strictus fecisse velim te, Deciane, facis , volito 7.102, , Martial, Epigrams 1.8.1, , , , , , , Tacitus, Annales (For purposes of illustration. Not all occurrences of each word shown.)

10 How Tesserae Works: Stage 2 - Sorting Stage 1: User selects Texts A and B for comparison. Then: For given units (line of poetry or sentence): Compare each unit in Text A with each unit in Text B Find all instances where they share two or more matching words Stage 2: Scoring system ranks selected parallels by: Rarity of the matched words Proximity of matched words in each phrase Highest scoring parallels have phrases with rare words, close together

11 How Tesserae Works: Stage 2 - Sorting

12 Testing: Hand Rank System Type Characteristics 5 High formal similarity in analogous context. 4 Moderate formal similarity in analogous context; or High formal similarity in moderately analogous context. 3 High / moderate formal similarity with very common phrase or words; or High / moderate formal similarity with no analogous context; or Moderate formal similarity with moderate / highly analogous context. 2 Very common words in very common phrase; or Words too distant to form a phrase. 1 Error in discovery algorithm, words should not have matched.

13 What Can Tesserae Capture? Tesserae V3 only 759 Shared 103 Commentators only 64 62% + of intertexts identified by scholars Unique interpretable (type 4-5) parallels between Lucan Civil War 1 and Vergil Aeneid found by Tesserae Version 3 (projected), commentators, and both. 82% of intertexts identified by scholars + projected bigram intertexts

14 Hand Rank Automatic Scoring 5 Average Hand Rank of Parallels per Automatic Score for BC 1 - Aeneid Test Fall Automatically Assigned Score

15 Multitext Search

16 Multitext Search

17 Multitext Search

18 Unique Lines with Parallels Large-scale Intertextuality Parallels with Aeneid Score 8 or above % Unique in Prior Corpus vs. % of Lines with Parallels 80% 70% 60% 50% 40% 30% 20% 10% 0% Ov. Met. Luc. BC Stat. Theb. Sil. Pun. VF Arg. DRP Cons. Stil. 35% 30% 25% 20% 15% 10% 5% 0% nec tu divinam Aeneida tempta, sed longe sequere et vestigia semper adora. Thebaid

19 Large-scale Intertextuality Network analysis and graphing by: Alexander Nikolaev, Alireza Farasat, Eric Yu U. Buffalo Dept. of Industrial and Systems Engineering

20 Large-scale Intertextuality

21 Other Search Types - Sound

22 Other Search Types - Sound

23 Other Search Types: Semantics Homer and Vergil Homer Iliad compared with Vergil Aeneid 4

24 Semantic Match: Homer and Vergil Homer Iliad compared with Vergil Aeneid 4

25 Ongoing Work Improved semantic matching by word and topic Feature fusion Work on intertextual repositories

26 Intertextual Repository

27 Intertextual Repository

28 PIs Neil Coffee (Buffalo Classics) J.-P. Koenig (Buffalo Linguistics) Walter Scheirer (U. Colorado, Colorado Springs Computer Science) Graduate Christopher Forstall, Tesserae Fellow and Version 3 Architect James Gawley, Tesserae Fellow Konnor Clark Caitlin Diddams Xia Lu Amy Miu Katie Roache is Collaborators Neil Bernstein (Ohio University) Damien Nelis (University of Geneva) Undergraduate Jessica Micceri Bridget Murray Ashley Cercone

29 thanks University at Buffalo Departments of Classics and Linguistics, College of Arts and Sciences Gregory Crane and Perseus team Helma Dik NEH Office of Digital Humanities

30 Links Advanced Latin search by lemma and sound: Catullus Georgics 1 by lemma (line and phrase (no.31)) and character tri-gram Prose - poetry: Aeneid and Seneca s letters Greek search: Iliad Argonautica 1 by line and phrase Multi-text Latin search: Aeneid Psychomachia results: Greek-Latin search: Iliad Aeneid 1 by word: Building a Greek-Latin dictionary Exporting results

31 Scoring System Function Demonstration of a calculation of Tesserae s automatic score using one of the top-ranked parallels from the Lucan Vergil test set: si licet, exclamat, Romani maxime rector nominis et ius est, veras expromere voces. (Civil War ) If it is permitted, he cries, o greatest ruler of the Roman name, and it is just to bring forth true utterances. ultro flens ipse videbar compellare virum et maestas expromere voces. (Aeneid ) For my part, I myself was seen from far away, weeping, to accost the hero and to bring forth sad utterances.

32 Scoring System Function These two phrases are returned by the stage-one bigram search because of the shared lemmata expromo and vox-voco. Stage two, the automatic scoring, depends on two criteria: the frequencies of the matching words, and the distance between them. The frequencies of the matching words are calculated for each text as follows. Target (Civil War) Source (Aeneid) expromo Number of occurrences 2 1 Total words in text 4,375 63,719 Frequency vox-voco Number of occurrences Total words in text 4,375 63,719 Frequency Because the form voces is ambiguous, automated counts include both the noun and the verb. Unsupervised disambiguation in such cases is a long-term goal.

33 Scoring System Function Distance is meant to measure the number of words spanned by the intertext. It thus includes the two endpoints: for example, expromere voces comprises two words in each phrase. Because the minimum distance is 2, we avoid having 0 in the denominator of the scoring equation in cases where matching words are adjacent, as in the present example. Calculation of the score for this parallel then proceeds in the following way: 1 f t + 1 f(s) score = ln d t + d s 1 = ln = ln , , = ln 66,693 4 = ln 16,673 = 9.7 where f(t) is the frequency of each matching term in the target phrase; f(s) is the frequency of each matching term in the source phrase; d t is the distance in the target; d s is the distance in the source.

34 Scoring System Function Frequency is the number of times a word occurs in its respective text divided by the total number of words in that text. The frequency of the same word may thus be different in different texts. Distance is measured between the two lowestfrequency matching words in a phrase. We assume that, where an allusion involves more than two shared words, the lowest-frequency words are likely the most important.

NOTE: THIS IS A PRE-PRINT DRAFT VERSION OF: The Tesserae Project: intertextual analysis of Latin poetry published in Literary and Linguistic

1 NOTE: THIS IS A PRE-PRINT DRAFT VERSION OF: The Tesserae Project: intertextual analysis of Latin poetry published in Literary and Linguistic Computing. The published version contains several editorial