A learner generated corpus to direct learner-centered courses

1
A learner generated corpus to direct
learner-centered courses

• Clara Inés López Rodríguez
• University of Granada (Spain)
• Bryan Robinson
• María Isabel Tercedor Sánchez
• Maastricht, May 2005

2
Research and teaching context

• RD project of the Spanish Ministry of Education
• PUERTOTERM knowledge representation and the
  generation of terminological resources within the
  domain of Coastal Engineering (Reference BF
  2003-04720)
• Scientific and Technical translation classroom
• English ltgt Spanish
• Teaching innovation action
• Localización del Texto multimedia generación de
  recursos en el aula de traducción científica y
  técnica
• (importance of audiovisual and multimedia
  material)
• Close collaboration between teachers

3
Social constructivism, collaborative learning
Kiraly (2000 72)
4
Objectives

• To build a sound foundation for new classroom
  activities
• to make students aware of their mistakes/errors
  deriving from lack of understanding of the
  subject field and their inability to grasp the
  meaning of the text.
• Analyze data extracted from a corpus of student
  translations so as to focus on the patterns and
  problems regularly generated by our learners.
• To see the effect translation problems related to
  meaning have on the type of mistakes/errors
  students make.

5
Theoretical foundations use of corpora in the
translation classroom

• DIY corpus Zannetin (2002), López Rodríguez
  (2002)
• Learner corpus Uzar (2002)
• Evaluation corpus Bowker (2001)
• Quality comparable corpus
• Quantity corpus
• Inappropriate corpus

6
Theoretical foundations evaluation

• Learner error vs. learning mistake (Miller 1966
  Corder 1973)
• Approaches to evaluation Waddington (2001)
• Analytical Martínez Melis and Hurtado (2001)
• Analytical holistic González Davies (2001)
• Holistic Waddington (1999, 2001) and
• Robinson (1998)

Adapted from Bryan Robinson 1998. Traducción
transparente métodos cuantitativos y
cualitativos en la evaluación de la traducción.
Revista de Enseñanza Universitaria (Número
extraordinario)577-589. ISSN 1131-5245.
7
Criterion descriptors (Robinson 1998)
DECODING
DECODING DECODING ENCODING ENCODING ENCODING
Content Register, vocabulary, terminology Translation brief and orientation to target text type Written expression
0 The text fails to meet minimum requirements The text fails to meet minimum requirements The text fails to meet minimum requirements The text fails to meet minimum requirements
1-2 Comprehension limited. Major content errors. Substantial omissions of ST content. Choice of register sometimes inappropriate or inconsistent. Vocabulary limited with some basic errors. Limited awareness of appropriate terminology. Formal or literal translation. Little or no use of rhetorical devices appropriate to TT type. Limited, with errors in basic constructions.
3-4 Comprehension adequate. Minor content errors. Some omissions of ST content. Choice of register occasionally inappropriate or inconsistent. Occasional mistakes of basic vocabulary. Clear awareness of appropriate terminology although some errors. Elements of formal or literal translation. Limited use of rhetorical devices appropriate to TT type. Ineffective, with errors in complex structures and mistakes in basic structures.
5-6 Comprehension good. Minor omissions of less relevant ST content. Over- or under-translation distorts ST content or results in ambiguity Choice of register mostly appropriate and consistent. Vocabulary effective despite mistakes. Terminology appropriate despite occasional errors. Clear orientation towards TT type. Appropriate use of many TT type rhetorical devices Effective. Complex structures are clear. Occasional mistakes. Some errors in use of articles, prepositions or spelling of less common words.
7-8 Comprehension very good. Over- or under-translation does not distort ST content or result in ambiguity. Choice of register appropriate and consistent. Vocabulary effective despite occasional mistakes. Terminology appropriate despite mistakes. Effective production of TT type. Consistently appropriate use of TT type rhetorical devices Good and effective. Occasional systematic errors of advanced usage only. No mistakes.
9-10 Comprehension excellent. ST content, including subtle detail, fully understood. Choice of register consistently effective and appropriate. Sophisticated, highly effective choice of vocabulary. Terminology appropriate and wholly accurate. Effective, sophisticated production of TT type with few or no mistakes. Very good. Sophisticated use of language with no mistakes and occasional or no errors of advanced usage.
8
Advantages of this scale

• Formative
• Holistic
• Transparent, easy to understand and conducive to
  self-assessment
• The learner acquires editor like training
• You can adapt the scale and give different
  weights to different aspects of the scale,
  depending on the level of the course and the
  directionality of translation

9
Adapting the descriptors to final year students
translating into their mother tongue
4. Written expression (ES gt EN) 4. Translation
brief and professional aspects
3. Translation brief and orientation to target
text type (ES gt EN) 3. Fluency and orientation to
target text type
10
Implementing learner-centered courses (I)

• 3rd and 4th year translation students (ENltgtES)
• Scientific translation, technical translation and
  localisation, audiovisual translation
• Subject field coastal engineering

11
Implementing learner-centered courses (II)

• Design of learner-centered activities
• Acquisition of field knowledge
• Getting familiar with corpora and annotation
• Identifying translation problems
• Developing translation strategies
• Evaluating the solutions given by peers and by
  themselves self-assessment of mistakes/errors
  and acquisition of editor like training

12
1. Acquisition of field knowledge and relevant
cognitive and lexical structures

• Analysis of visual and multimedia aids
• Reading skills skimming and scanning
• Compiling a DIY corpus
• Conceptual modeling elaborating ontologies and
  glossaries

13
Analysis of visual/multimedia aids
http//cil-www.coas.oregonstate.edu8080/frames/mo
tivate/motivate.html http//coastview.ims.plym.ac.
uk/video.html http//wldelft.nl/cons/appl/argus/in
dex.html
a. Describing and comparing images using the
target language (Scene-and-frame semantics,
Kussmaul 1991, 1995). b. Matching images with
their captions.
PROCESS-ORIENTED TERMINOLOGY AND TRANSLATION
14

• CAPTIONS
• Instrument deployment during SandyDuck97.
• Diver installing pressure transducer on the sea
  bead.
• Time-lapse movie camera on Oregon Coast.

15
Argus collects three basic image products for
beach studies. These include the traditional
snapshot, showing wave activity, ten-minute
time-exposure (timex) images of the wave
dissipation patterns (revealing submerged sand
bars and rip channels) and variance images
(separating dynamic from steady areas of the
image. The greatest scientific value currently
come from the timex images.
16
c. Associating images with the body of the text
and with animations
White bands in a timex image, locations of
enhanced breaker dissipation, provide a proxy for
submerged sand bars.
Click for an AVI or RealPlayer version of an
animation demonstrating the conversion of live
video like that from Duck, NC on the left, into a
finished timex image, below.
17
DIY corpus and conceptual modeling

• assessing reliability of texts
• improving documentation and terminology skills
  elaborating ontologies and glossaries
• acquiring the basic concepts that shape knowledge
• interrelation between concepts
• dynamic and process-oriented conceptual
  structure Barsalou 2003 (apud Faber, Márquez y
  Vega in press)
• PROCESS-ORIENTED TERMINOLOGY AND TRANSLATION

18

• COASTAL ENGINEERING EVENT
• (Faber, Márquez and Vega (in press)

MICRO-EVENTS After reading different texts on a
specific topic (monitoring systems in coastal
engineering-video imaging), students elaborate
their own ontology MindManager X5
Pro http//www.mindjet.com Protégé http//prote
ge.stanford.edu/ Cmap http//cmap.ihmc.us

19
(No Transcript)
20
(No Transcript)
21
2. Corpus-based exercises and annotation tagging
the corpus

• Initial tags referring to style, translation
  brief and professional aspects
• ltstyle5gtltt-brief1gtltprofessional3gt
• Tags placed after the problem or mistake/error
• Tags specify
• Type of problem
• Type of error/mistake
• Adequacy /appropriateness of translated sentences

22
3. Identifying translation problems (tagging)

• ltNumbergt referring to sentence number n
• Conceptualization ltCONgt
• Procedural ltPROgt
• Transfer (due to linguistic and cultural
  differences) ltTRAgt
• Lack of quality of the Source text ltQTOgt

López and Tercedor (2004)
23
and consists of 3 Snap Shotslt3gtltPROgt These
images are instantaneo 4 Time exposure images
(Timex)lt4agtltPROgt These images average the ver a
time period of 10 minuteslt4bgtltCONgt. 5 Typically
images are rec frequently over shallow
sanbarslt6gtltQTOgtltCONgt generating foam appear a .
7 The wave-breaking patternslt7gtltPROgt highlighted
in the timex imag the surface. 8 Variance
imageslt8gtltPROgt These images represent the v ude
the waters edge (swash zone)lt10gtltCONgtltPROgt where
the beach is p projected on the ground
planelt11agtltCONgt, resulting in rectified image s
with real world co-ordinateslt11bgtltCONgt. 12
These rectified images ously, so-called merged
imageslt13agtltCONgt can be obtained, which give a
lan view of the nearshore zonelt13bgtltPROgt. 14
The figure below shows time domain of optical
signalslt15agtltCONgt, by rapidly sampling the inte l
elements of the image (pixels)lt15bgtltTRAgt. 16 It
is now possible ow possible to define an
arraylt16agtltPROgt of pixels in the image which l
moving oceanographic targetslt16bgtltCONgt. 17
Tests have shown pixels ampled at high
frequencies (1Hz)lt18gtltCONgt, so that a time series
of
24
4. Translation of the text(tagging their own TT)

• Students use translation strategies and produce a
  translation which includes tags pointing to
  problematic segments
• Numbers lt1gt, lt16agt, lt16bgt)

25
5a. Evaluating translation segments of peers
(tagging)

• Students evaluate potentially problematic
  segments of texts from group assessments
• Tagging these segments according to the type of
  mistake/error
• Tagging according to adequacy or appropriateness
  of translated sentences
• Relating meaning problems with mistakes/errors

26
Tagging type of error/mistake and adequacy of
translation

• (1) Offer students a list of filtered
  concordances with their own rendering of the
  problematic segment
• (2) Students identify type of error/mistake and
  tag the segments
• (3) They grade the segments according to quality
  parameters (Laurscher 2000, López and Tercedor
  2004) adequacy /appropriateness of translated
  sentences
• (4) Justify own choice of best segment

27
SOURCE TEXT Regions where waves break frequently
over shallow sanbarslt6gtltQTOgtltCONgt generating foam
appear as bright white bands in the timex images.
The wave-breaking patternslt7gtltPROgt highlighted in
the timex images can be used to infer the
position and shape of sandbars, even though the
bars are not visible above the surface.

• FILTERED CONCORDANCES
• 1 bre bancos de arena poco profundoslt6gtltlxgt
  generando espuma, en las imá
• 2 en bancos de arena poco profundoslt6gtltlxgt
  generando espuma aparecen co
• re barras de arenas poco profundaslt6gtltcolxgt,
  aparecen indicadas en la
• arras de arena en zonas de poca
  profundidadlt6gtltAAgt, generando espuma, apa
• 5 las barras de arena poco profundaslt6gtltsegt, lo
  que genera espuma, aparecen
• n barras de arenas no muy profundalt6gtltsegt,
  provocando espuma, se represent
• as barreras de arena superficialeslt6gtltlxgt,
  generaltccsxgtltFFgt la aparici
• en una barra de arena superficiallt6gtltlxgt, lo que
  hace que se produzca
• bancos de arena son poco profundoslt6gtltlxgt. Para
  inferir la posición y
• superficie. Imágenes de varianzalt6gtltlxgt estas
  imágenes representan
• ltsxgtltFFgt sobre las barras de arenalt6gtltmisegt
  generando espuma en el agu
• arras de arena de poca profundidadlt6gtltsegt
  produciendo espuma aparecen
• obre las barras de arena del fondolt6gtltsegt,
  generando espuma, aparecen

28
Type of error/mistake (from corpus)
1. CONTENT ES gt EN EN gtES (x 2)

• ltsegt meaning
• ltchsegt Lack of cohesion
• ltmisegt less information than ST
• ltplusegtmore info than ST
• lttvsegt wrong tense that causes change in meaning
• ltcosegt change in meaning due to wrong
  collocation
• ltdtsegt changes in the data

29
2. REGISTER AND LEXIS ES gt EN EN gtES

• ltlxgt lexis and terminology
• ltcolxgt wrong collocation
• ltrglxgt term that is not appropriate for the
  register of the text
• ltrggt register (inconsistencies)

30
3. TRANSLATION BRIEF AND ORIENTATION TO TARGET
TEXT TYPE ES gt EN

• ltogt organization
• ltprgt pragmatic mistakes
• ltrtprgt Grammatically correct but it sounds
  unnatural. The rhetorical effect of the ST is
  missing. Literal translation.
• ltistgt inappropriate style
• ltotgt orthotypography
• ltfgt layout, wrong accomplishment of style sheet
  or computer requirements

31
3. FLUENCY AND ADEQUACY TO ORTHOTYPOGRAPHICAL,
TEXTUAL AND PRAGMATIC CONVENTIONS OF TARGET
LANGUAGE EN gt ES

• ltogt organization
• ltprgt pragmatic mistakes
• ltrtprgt Grammatically correct but it sounds
  unnatural. The rhetorical effect of the ST is
  missing. Literal translation.
• ltistgt inappropriate style
• ltotgt orthotypography

32
4. WRITTEN EXPRESSION ES gt EN

• ltorgt spelling
• ltptorgt punctuation
• ltsxgt syntax
• ltccsxgt lack of concord

33
4. TRANSLATION BRIEF AND PROFESSIONAL ASPECTS EN
gt ES
A syntactic, spelling or punctuation mistake will
reduce the mark in the fourth column to the
minimum

• ltfgt layout, wrong accomplishment of style sheet
  or computer requirements
• ltorgt spelling
• ltptorgt punctuation
• ltsxgt syntax
• ltccsxgt lack of concordance

34
Adequacy of translated sentences

• Excellent solution
• ltAAgt
• Inappropriate
• lttype of error/mistakegt
• Very bad-serious mistake/error
• lttype of error/mistakegtltFFgt

35
Relating problems with mistakes

• Students see that when they concentrate on
  specific problems and do not revise, inadvertent
  mistakes slip in. As a result, the quality of
  their translation diminishes dramatically.
• Students keep a record of mistakes and errors in
  a learner diary.
• Conceptual problems -gt serious mistakes.

36
Merged image (panoramic and rectified view) of
Noordwijk. Five individual time-exposure images
are used to compose these merged images. In the
rectified image (lower panel)lt12agtltTRAgt, the
shoreline is located at the lower side, around x
0 m lt12bgtltCONgt. The bright band of 3000 m
length at about 900 m off-shorelt13agtltCONgt
indicates the location of a 1.25 Mm³lt13bgtltTRAgt
shoreface nourishmentlt13cgtltCONgt.
En la imagen rectificada (panel
inferior)lt12agtltlxgt el litoral estaltorgtltFFgt
situado en el lado inferior, aproximadamente x
0 mlt12bgtltsegt. La banda luminosa de 3000
metrosltotgt de longitud a aproximadamente 900 m de
altamarlt13agtltsegtltFFgt indica la localización de
una regeneración de la zona costeraltsegt de 1.25
Mm3lt13bgtltotgtltFFgt.
37
5b. Evaluation of full translations tagging and
criterion descriptors

• They evaluate the translation of other students
  and their own translation.
• They tag the translation and give a mark
  according to the criterion descriptors

38
6. Additional exercises

• Corpus analysis with lexical analysis software
  (Wordsmith Tools)
• Proposal of new texts to be translated
• Access to DIY corpora
• Access to quality corpus (Puertoterm)
• Access to the learner corpus solutions to
  translation problems

39
(No Transcript)
40
Conclusions and future research

• In specialised translation, the aim of activities
  should be to increase learner autonomy different
  skills and tasks involved in the translation
  process and in evaluation.
• Students who understand the basic conceptual
  structure of the subject field (visual aids and
  DIY corpus) do better translations.

41
Conclusions and future research

• Bottom-up, top-down approach to translation
• Learner corpora as input for lexical analysis
  software
• consistency analysis and statistical analysis of
  translations false friends, most common
  mistakes/errors, etc.