1 / 37

Translating Data Driven Language Learning into French

Translating Data Driven Language Learning into French. Tom Cobb Dép. de Linguistique Université du Québec à Montréal. Peut-on augmenter le rythme d’acquisition lexicale par la lecture ?.

fawzi
Download Presentation

Translating Data Driven Language Learning into French

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Translating Data Driven Language Learning into French Tom Cobb Dép. de Linguistique Université du Québec à Montréal

  2. Peut-on augmenter le rythme d’acquisition lexicale par la lecture ? Une expérience de lecture en français appuyée sur une série de ressources en ligne.Tom Cobb, Université du Québec à Montréal

  3. Can the rate of lexical acquisition from reading be increased? An experiment in reading French with a suite of on-line resources.Tom Cobb, Université du Québec à Montréal

  4. Background: Data-Driven Language Learning On-line • Discovery learning • Learner-as-linguist • Alternatives to rules & definitions • Concordancing • Grammar Safari • Concordancing • Concordancing on-line • Concordancing on-line in French

  5. The idea of shortcuts to L2 • It has long been known that the time available for LL through experience is inadequate in most cases • Learner’s time is short • Database is dispersed • Much time is needed to expose patterns in data

  6. The traditional shortcut to L2: Explicit declarative knowledge • ‘Rules’ in grammar • ‘Definitions’ in vocabulary • Never all that successful • Linguistic computing makes another kind of shortcut possible • Data aggregation & compression • Rapid pattern exposure

  7. ‘Rules’ in grammar • Error: * This is one of the biggest car in the world • Solution: We tell students the rule: “After one of the comes a plural noun”

  8. Or, tell them to go check the data 10 of 396 examples in Brown Corpus…

  9. Advantages of data based learning • Learners initiate search themselves • Patterns are large, crystal clear • Linguistic authenticity is assured • Learners have positive role to play: they are linguists (Cobb, 1999) • Cf. negative ‘mistake maker’ role in traditional approach • Technology is used in a non-gaming context • And used well, since concordances can not be generated by any other means

  10. Building a second lexicon - big need for data aggregation • Contextual inference problematic • On learner-side (inferences generally unsuccessful; Laufer, Haynes et al studies) • On data-side (poor contexts, vast distances between) • Dictionary information hard to use by those who need it • Direct instruction runs up against task-size problem

  11. Can computer data-aggregation help build a second lexicon?Two ideas: 1. List-driven learning: Corpus and concordance linked to frequency lists • Frequency based testing to find level • Make yourself a dictionary at the level where you are weak • Example: Lexical Tutor

  12. Problems with list-driven learning: • Needed frequency information seems unavailable except in English • List is not everyone’s cup of tea So, another idea: Adapt computational tools to the less structured context of extensive reading

  13. Introducing R-READ ReadingExtended Authentic Documents withResources …of a kind that are increasingly capable of Internet delivery

  14. Brief History of Computer-Assisted L2 Reading • Pre-Internet Age: Skills based, no proof of transfer, “too little to read” • Internet Age: Too much to read, reading reduced to scanning

  15. R-READ as a middle way • that uses Internet resources to • make extensive authentic documents readable, and • target specific learning

  16. Personal Anecdote • Me, 1980, French reading test looming… • Method: read one book, several times, aided by a ‘language consultant’ • Voltaire’s Candide • Francophone girlfriend • Look into every word; deconstruct every structure • Repeat pronunciations • Stick-on concordances • Little notebooks • Stick-on’s removed, fewer look-ups • First Hurdle clear in about a week

  17. Equity problem: • Not everyone can find a personal language consultant • Question: Would it be possible to itemise what the consultant was doing and reproduce these services universally?

  18. An electronic language consultant? Go online VLC

  19. User lexicon

  20. Research Base (1) • Listen & read • Draper & Moeller, 1971; Stanovich, 1896. Lightbown,1992 • Concordance: computer aided contextual inference • Huckin, Haynes & Coady, 1991; Cobb, 1999; Zahar, Cobb, & Spada, in press • Database as take-home learning outcome • Minimal time-off-task (Cobb, 1997) • Collaborative (Horst & Cobb, in prep)

  21. Research Base (2) • Dictionary • Can disrupt reading, cause misconception (Noblitt et al, 1990) • Useful pair with context if it follows effort to infer (Fraser, 1990) • Click-on interface • Even if useful, dictionary will not be used if effortful (Hulsteijn et al, 1996)

  22. Research Base (3) • R-READ as middle position between stark choices of the past on extensive reading • Alternative 1: Natural extensive reading is an adequate source of vocabulary growth in L1 (Krashen, 1989) or L2 (Nagy, 1997) • Alternative 2: Vocabulary growth will not happen if conditions are not in place; assure they are in place by pre-teaching wordlists, out of context if necessary (Nation & Waring, 1997)

  23. Middle approach made possible through ‘NTIC’ • Vocabulary enhanced reading (Hulstijn, Holander, & Greidanus, 1996) • Learners make their own way through roughly tuned texts with support of resources • In-context feature preserved • But is it useful? • What follows is a substantial test of this middle approach

  24. Pilot Test of de Maupassant’s Boule de Suif with R-READ • How do vocabulary learning results of reading with online lexical resources compare to results of reading without these tools? • Baseline for comparison: Repeated-reading case studies of lexical acquisition by Horst (2000)

  25. R – motivated adult intermediate learner German novella 9500 words 300 unique targets (1:32) 45% rated unknown at pretest 20% rated known at pretest Treatment 3 readings Av. 3 hrs / reading (3167 wds/hr) R’s reading of German novella (Horst, 2000)

  26. J – motivated adult intermediate learner Boule de Suif 13,400 words 400 unique targets (1:33) 45% rated unknown at pretest 27% rated known at pretest Treatment 3 readings Av. 4.6 hrs/reading(2913 wds/hr) J’s reading of Boule de Suif

  27. R – motivated adult intermediate learner German novella 9500 words 300 unique targets (1:32) 45% rated unknown at pretest 20% rated known at pretest Treatment 3 readings Av. 3 hrs / reading (3167 wds/hr) J – motivated adult intermediate learner Boule de Suif 13,400 words 400 unique targets (1:33) 45% rated unknown at pretest 27% rated known at pretest Treatment 3 readings Av. 4.6 hrs/reading(2913 wds/hr) R’s German novella vs. J’s Boule de Suif

  28. Rating scaleused at end of each reading • 0 = I don't know what this word means • 1 = I am not sure what this word means • 2 = I think I know what this word means • 3 = I definitely know what this word means (Underlining added) Non-binary measure, Horst & Meara, 1999

  29. Results

  30. Pretest Posttest 1 Posttest 2 Posttest 3 0 (unknown) 180 wds 74 49 28 1,2 (unsure) 142 wds 189 165 170 3 (known) 78 wds 137 186 202 J’s word knowledge ratings before reading and after each of three readings (resource assisted) Summary: Unknown reduced from 180 to 128 Known increased from 78 to 202

  31. Results for R (unassisted)n=300 words Results for J (R-READ)n=400 words Pretest 3rd posttest Pretest 3rd posttest 0 (not known) 45% 38 45 7 1 or 2 (unsure) 28% 33 36 43 3 (known) 27% 29 20 51 Comparison to baseline Percentage of targets in each category at outset and after three readings, unassisted and assisted

  32. Results for R (unassisted)n=300 words Results for J (R-READ)n=400 words Pretest 3rd posttest Pretest 3rd posttest 0 (not known) 45% 38 45 7 1 or 2 (unsure) 28% 33 36 43 3 (known) 27% 29 20 51 Comparison to baseline R’s results typical of many acquisition-from-reading studies;J 250% greater in ‘known’ category.

  33. Self-assessment check • J (after 3 readings) and R (after 10 readings) asked for translations of words judged known • Js responses 94% accurate(Three readings with R-READ) • Rs responses 77% accurate • (10 unassisted readings)

  34. Conclusion (1) • This is only a pilot study • Suggests significant learning increase for minor time increase • These are learning figures seen in previous research only for tiny word sets via ‘rich’ instruction (Beck, McKeown… 1982)

  35. Conclusion (2) • Suggests viablity of middle-way model of acquisition-through-reading • Suggests that low-cost language consultants can be brought into wide-spread use

  36. Conclusion (3) J. B. Carroll (1964) expressed a wish that a way could be found to mimic the effects of natural contextual learning, except more efficiently.... • Maybe this ancient educational cul-de-sac can be solved through the principled application of computer technology – how many others?

  37. Acknowledgements • This Web page incorporates the labours of many: • The roman 'Boule de Suif' • Guy de Maupassant (1870) • Concordance program, true click-on hypertext  • Chris Greaves, Virtual Language Centre, Polytechnic University, Hong Kong • French-English Dictionary • Neil Coffey  http://www.french-linguistics.co.uk/dictionary/ • Complete Corpus of de Maupassant oeuvre • Thierry de Selva, Laboratoire d'Informatique, Université de Franche-Compté, Besançon • Read-aloud of 'Boule de Suif' • Dominique Daguier, for «Le livre qui parle» • Perl scripting for User Lexicon • Mutassem Abdulahab & Monet, EZScripting. • Web formatting of 'Boule de Suif' • Carole Netter,Clicnet, Swarthmore College. • Historical Background • Luc et Eric Dodument, Skylink, Hombourg, Belgium. • Movie poster • http://perso.wanadoo.fr/lester/fifiaffiche.htm • Frequency List • Association des Bibliophiles Universels (ABU), De Maupassant, CEDRIC/CNAM, Paris

More Related