A Taxonomy of Weeds: A Field Guide for Corpus Curators to Winnowing - - PowerPoint PPT Presentation

▶
a taxonomy of weeds a field guide for corpus curators to
SMART_READER_LITE
LIVE PREVIEW

A Taxonomy of Weeds: A Field Guide for Corpus Curators to Winnowing - - PowerPoint PPT Presentation

Mar 09, 2023 587 likes •991 views

A Taxonomy of Weeds: A Field Guide for Corpus Curators to Winnowing the Parallel Text Harvest Katherine M. Young Jeremy Gwinnup Lane O. Schwartz N-Space Analysis, LLC katherine.young.ctr.1@us.af.mil Air Force Research Laboratory

slide-1
SLIDE 1

A Taxonomy of Weeds: A Field Guide for Corpus Curators to Winnowing the Parallel Text Harvest

Katherine M. Young Jeremy Gwinnup Lane O. Schwartz

N-Space Analysis, LLC

katherine.young.ctr.1@us.af.mil

Air Force Research Laboratory

jeremy.gwinnup.1@us.af.mil

University of Illinois at Urbana-Champaign

lanes@illinois.edu

31 October 2016

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-2
SLIDE 2

Modern MT requires corpora

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-3
SLIDE 3

Weeds among the wheat

“Weed is possibly a beer metaphor than dirt for many of these issues with SMT training data: a wild plant growing where it is not wanted and in compeon with culvated plants” – Simard (2014)

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-4
SLIDE 4

Weeds among the wheat

ἐν δὲ τῷ καθεύδειν τοὺς ἀνθρώπους ἦλθεν αὐτοῦ ὁ ἐχθρὸς καὶ ἐπέσπειρεν ζιζάνια ἀνὰ μέσον τοῦ σίτου καὶ ἀπῆλθεν. Parable of the weeds —Greek New Testament

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-5
SLIDE 5

Some types of weeds

zizania ex machina

  • Wrong language text
  • Historical encoding errors
  • Bidireconal reversal
  • Sentence alignment errors
  • Sentence-internal fragments
  • Harvested MT

zizania ex homine

  • Mixed alphabet spellings
  • Mixed morphology
  • Transliteraons of NE and

Borrowings

  • Underachieving translaon
  • Overachieving translaon
  • Translaon direconality

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-6
SLIDE 6

Zizania ex machina

Weeds of mechanical origin

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-7
SLIDE 7

Wrong language text

IWSLT 2014

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-8
SLIDE 8

Wrong language text

IWSLT 2014

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-9
SLIDE 9

Wrong language text

Common Crawl

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-10
SLIDE 10

Wrong language text

Common Crawl

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-11
SLIDE 11

Wrong language text

  • U I (і І)
  • Y(ї Ї)
  • G W U (ґ Ґ)
  • I (є Є).

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-12
SLIDE 12

Wrong language text

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-13
SLIDE 13

Wrong language text

+0.34 BLEU on WMT’15.

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-14
SLIDE 14

Historical encoding errors

Ñïðàâêà ïî ãîðîäàì Ðîññèè è ìèðà

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-15
SLIDE 15

Historical encoding errors

Ñïðàâêà ïî ãîðîäàì Ðîññèè è ìèðà Справка по городам России и мира

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-16
SLIDE 16

Historical encoding errors

In ru-en common crawl, some Russian text is actually encoded as Windows-1251 but is incorrectly interpreted as UTF-8 Correct by shiing 350hex code points into Unicode Cyrillic range

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-17
SLIDE 17

Historical encoding errors

йquipe chвteau

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-18
SLIDE 18

Historical encoding errors

йquipe chвteau équipe château

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-19
SLIDE 19

Historical encoding errors

In fr-en common crawl, some French text has the reverse problem Correct by shiing −350hex code points Unicode Lan range

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-20
SLIDE 20

Bidireconal reversal

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-21
SLIDE 21

Bidireconal reversal

Extracon of RTL text from PDF

  • PDF is a display format
  • Extracted text is reversed,

character-by-character Noceable because the teh-marbuta character [ة] should not begin a word

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-22
SLIDE 22

Sentence-internal fragments

Last year I showed these two slides so that demonstrate that the arcc ice cap, which for most of the last three million years has been the size of the lower 48 states, has shrunk by 40 percent. L’année dernière, je vous ai présenté ces deux diaposives qui mon- traient que la caloe glacière arcque, qui pendant ces 3 derniers millions d’année avait la taille des Etats-Unis sans l’Alaska, qui pen- dant ces 3 derniers millions d’année avait la taille des Etats-Unis sans l’Alaska, avait diminué de 40%.

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-23
SLIDE 23

Sentence-internal fragments

+1.53 BLEU on IWSLT’14

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-24
SLIDE 24

Sentence-internal fragments

It’s the difference between divergent thinkingand convergent

  • thinking. You have to separate

the two so that you can diverge your thoughtsand come up with this great collecon of ideas, and then once you have this great colleconof ideas, you focus on the convergent thinking. Chunking errors in the QED Corpus. Resolved using greedy search for two valid words.

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-25
SLIDE 25

Harvested MT

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-26
SLIDE 26

Zizania ex homine

Weeds of human origin

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-27
SLIDE 27

Under-achieving translaon

Errors in human translaons:

  • Transliteraon instead of translaon

Awards and Reviews → Награды и Ревью /nagradɨ i rev’ju/

  • Codeswitching

English: In the top ten, India comes in the last Urdu: “cerficaon”, “top ten”, and “number”

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-28
SLIDE 28

Over-achieving translaon

Explicitaon in human translaons:

  • Explanaons added
  • Acronyms expanded
  • Visual aids described in target side of TED talks
  • Disfluencies removed from translaon of speech

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-29
SLIDE 29

Mixed alphabets

  • нa

cейчас

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-30
SLIDE 30

Mixed alphabets

Word Lan (L) or Cyrillic (C) Meaning

  • нa

LCL she cейчас LCCCCC now MP3-плеер LLL-CCCCC MP3-player MP3плеер LLLCCCCC MP3player амазон.com CCCCCC.LLL amazon.com iпациент LCCCCCCC iPaent Russian word oнa is encoded such that the Lan characters o and a are used instead of the more appropriate (but visually indisnguishable) Cyrillic equivalents.

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-31
SLIDE 31

Mixed alphabets

Urdu:

  • U+002D - L H
  • U+06D4۔ U F S

French:

  • U+00A8 ¨ L D
  • U+0022 '' L Q M

Russian:

  • U+0431 б C S L B
  • U+0036 6 L D S

English:

  • U+006F o L S L O
  • U+00B0 ° L D S

Correcng such errors typically requires human intervenon

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-32
SLIDE 32

Mixed morphology

Mixed use of Urdu and English plural markings

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-33
SLIDE 33

Mixed morphology

Mixed use of null and Russian plural markings

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-34
SLIDE 34

Mixed morphology Soluons:

  • Stem OOVs before transliterang
  • Replace inflected OOV with in-vocabulary variant

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-35
SLIDE 35

Transliteraon of Names and Borrowings

Муаммар Каддафи

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-36
SLIDE 36

Transliteraon of Names and Borrowings

Муаммар Каддафи

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-37
SLIDE 37

Transliteraon of Names and Borrowings

Rule-based transliteraon recovery of NE

  • Transform target language pronunciaon diconary entries;

map to English entries

  • Pre-translate known NEs from exisng list
  • Third-language mappings

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-38
SLIDE 38

Third-language mappings

(a) 翟志刚 (b) Чжай Чжиган (c) Chzhay Chzhigan (d) Zhai Zhigang Chinese name (a), with transliteraons into Cyrillic (b) and Lan using normal Cyrillic-to-Lan transliteraon (c) and reverse Palladius transliteraon (d). The output in (d) is correct.

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz

slide-39
SLIDE 39

Weeds among the wheat

Movaon Weeds of mechanical origin Weeds of human origin Conclusion AMTA 2016 Katherine M. Young, Jeremy Gwinnup, Lane O. Schwartz