Presentazione di PowerPoint

1
using corpora learning to translate and the web
era
2
We have to stop pretending we do not know what
should go into the training of translators.
Obviously, we are supposed to train people to
perform clearly identified functions in clearly
identified environments where they will be using
clearly identified tools and systems. The
problem may be one of defining the architecture
and implementations of the curriculum, not of
defining the curriculum itself.  No serious
translator training programme can be dreamt of
unless the training environment emulates the work
station of professional translators.   the
curriculum should concentrate on emulating
the actual work conditions of language services
providers. Gouadec 2002
3

• would be translators must have the
  physiological and psychological qualities to
  withstand pressure and stress and to surf over
  and manage crises. They must be both born leaders
  and fully cooperative team-mates
• Gouadec 2002

4

• Know-how
• Get the information and knowledge required
• Find the terminology
• Find the phraseology
• Translate
• Proofread
• Rewrite
• Manage their task(s)
• Manage a project (and other people)

• Skills
• Fully understand material to be translated
• Detect, interpret cope with cultural gaps btw S
  and T cultures
• Transfer information, facts, concepts, arguments,
  lines of reasoning, into a different culture, for
  a different set of users
• Write and rewrite
• Proofread
• Control and assess quality

5

• Get into the habit Montserrat Monastery
• Latin Primers Latin dance
• Getting high Castells
• Death on the mountain Montjuïc
• A place in the sun Beaches
• On the tiles Josep Maria Jujol
• All that glitters New architecture
• From Time Out Barcelona Guide 2003 Penguin

6

• all that glitters
• non è tuttoro
• lit. not is all gold

• isnt gold
• quel che luccica
• that that glitters

7

• (1) iron and glass modern design, shiny newness
• (2) fig. disappointing, a fraud
• (1) modernist mosaics/buildings traditional
  design
• (2) fig. disappointing, a fraud

• All that glitters
• Non è tutto oro

8

• So what are the general abilities to be taught at
  school? They are the abilities which take a long
  time to learn text interpretation, composition
  of a coherent, readable and audience-tailored
  draft translation, research and checking,
  correcting. But nowadays one constantly hears
  that what students really need are skills in
  document management, software localization,
  desktop publishing and the like. I say nonsense.
  If you cannot translate with pencil and paper,
  then you cant translate with the latest
  information technology.
• (Mossop 1999 online)

9
Estimates of Web size in words, as indexed by
AltaVista (from Kilgarriff and Grefenstette 2003)
10
What is it? Harry asked shakily. This? It is
called Pensieve, said Dumbledore. I sometimes
find, and Im sure you know the feeling, that I
simply have too many thoughts and memories
crammed into my mind. Er, said Harry, who
couldnt truthfully say that he had ever felt
anything of the sort. At these times, said
Dumbledore, indicating the stone basin I use the
Pensieve. One simply siphons the excess thoughts
from ones mind, pours them into the basin, and
examines them at ones leisure. It becomes easier
to spot patterns and links, you understand, when
they are in this form. Dumbledore placed his
long hands on either side of the Pensieve and
swirled it, rather as a gold prospector would
swirl for fragments of gold
11

• Although a growing number of professional
  translators considers the World Wide Web the
  ultimate reference tool (mostly Americans for
  whom Internet time represents less of an
  investment), a considerable number of obstacles
  have to be overcome before the Web's riches can
  be harvested expediently and efficaciously. It is
  mainly the inability to collectively order and
  sort the search words and their contexts and the
  lack of reliability discussed above that
  considerably limit the profitability of Web
  searches.
• Ingrid and Michael Friedbichler at CULT 1997

12
(No Transcript)
13

• average SVM performance
• accuracy 87
• precision 89
• recall 83
• F 86
• chance level 50 for each
• Info on SVMlight
• http//svmlight.joachims.org/

14
fold 11 SVM vs humans

• SVM
• accuracy 87
• precision 87
• recall 87
• F 87

• Humans
• accuracy 77
• precision 84
• recall 74
• F 79

15
Silvia Bernardini School for Interpreters and
Translators University of Bologna,
Italy silvia_at_sslmit.unibo.it
CULT BCN, 22-24 January 2004 Universìtat
Autònoma de Barcelona