the role of translators in mt: eu 2010
DESCRIPTION
What role can human translators play in today\’s machine translation scenarios?TRANSCRIPT
The Importance of Human Translators in a Machine
Translation Process
Lori Thicke
www.lexcelera.com
www.lexcelera.com
. A short history of MT
. Different approaches to MT
. Factors driving MT today
. The changing role of translators
. What is post-editing?
. A final word about quality
A Short History of MT
Machine Translation has been around since the 50s.
.A short history of MT
But expectations were unrealistic
And the human dimension was missing
Different Approaches to MT
Machine Translation is not one program …but different programs taking different approaches to get to the same place
.Different approaches to MT
Rules-Based MT (RMBT)e.g. Systran, ProMT
Comes with linguistic rules, trained by domain-specific user dictionaries built by linguists
.Different approaches to MT
Statistical MT (SMT)
e.g. Moses (open source), Google, Language Weaver, Asia Online
May have domain built in, trained by engineers on millions of segments
.Different approaches to MT
Advantages of RBMT
▪ Customization requires trained linguists
▪ Only one language pair at a time
▪ Not available for every language
▪ Labour intensive to scale/extend to new domains (4-6 weeks)
▪ Respects grammatical rules: Le chat vert
▪ User dictionaries control terms
▪ System can be trained/corrected in real time
▪ Large corpora not necessary
▪ Most user cases today are Systran
Disadvantages of RBMT
.Different approaches to MT
Advantages of SMT
▪ Requires millions of segments of clean training text (TMs, bilingual and monolingual corpora)
▪ Unpredictable: terms may be correct in one sentence and not in another▪ System makes its own decisions based on probability▪ Corrections not always easy to integrate into engine▪ Memory and processor intensive▪ No grammatical rules to govern text it hasn’t seen: le vert chat
▪ Learns automatically ▪ Same process for virtually any language▪ Customization can be done with limited
human input▪ Extending to new languages/domains takes
up processing capacity, not linguistic resources
Disadvantages of SMT
.Different approaches to MT
www.lexcelera.com
Factors Driving MT Today
.Factors Driving MT Today
Content is Exploding.
A week’s worth of The New York Times contains more information than an 18th century person
encountered in a lifetime
More unique information will be generated this year
than in the previous 5000 years.
More internet messages are sent in one day than the population of the world
Symantec created more content last year than in the previous decade
.Factors Driving MT Today
And that’s not to mention UGC.
Hundreds of millions of blogs
54% of bloggers post content or tweet daily
5 billion videos are being viewed on YouTube per month
25% of searches link to user-generated content
.Factors Driving MT Today
Many languages to translate.
.Factors Driving MT Today
More platforms for distributing information.
.Factors Driving MT Today
Increasing “Glocalization”.
.Factors Driving MT Today
Users want the information they want when they want it.
.Factors Driving MT Today
And sometimes Good Enough is Good Enough.
.Factors Driving MT Today
Human Translation Can’t Meet the Needs to:
Translate more content
Into more
languages
For more platfor
ms
Coming from multipl
e source
s
FasterCheap
er
.Factors Driving MT Today
“…we contend that many companies and government agencies will consider automated translation as a way to maximize the amount of information available to customers and constituencies who speak other languages.” (Common Sense Advisory Report:“Automated Translation Technology”, 2006)
Machine translation can help us meet these new challenges
.Factors Driving MT Today
Machine Translation has become just another productivity tool.
. Fewer translators are graduating, so we need to optimize their time
. It’s twice as fast to post-edit MT output than to translate from scratch
. MT tools are integrated with TM tools for even better leveraging
. It’s also faster to post-edit MT than fuzzy matches
. After training, 7000+ words per day for human quality, 10,000+ for “comprehensibility”
.Factors Driving MT Today
The changing role of translators
.The changing role of translators
The Scribe.
.The changing role of translators
Part of the Production Line.
.The changing role of translators
Subject Matter Expert.
What is post-editing?
. Correcting machine translation output
. Making global changes that once entered in the MT engine will ensure that those same errors will not recur
.What is Post-Editing?
How does post-editing differ from traditional translation revision?
. One question that may be asked at the outset is what quality is required to determine the post-editing effort. One-off errors are corrected but attention is paid to systematic changes that can be made to the engine to correct repetitive errors
.What is Post-Editing?
Skill set of a post-editor
Like a translator:. Excellent source language knowledge. Writing ability in target language. Specialization. Computer fluency. DiscriminationPlus. Ability to detect patterns and make global corrections to propagate changes. An open mind to MT
.What is Post-Editing?
Light Post-Editing.
. Understandable quality
. Message should be accurate and complete
. Terminology generally managed by MT engine
. May contain stylistic errors, awkward sentences
. Make changes only when absolutely necessary for comprehension
. Preferential changes should be avoided
. Quality goals: Can it be understood without the original? Does it increase access to information?
.What is Post-Editing?
Full Post-Editing.
. Publishable quality
. Message should be accurate and complete
. Terminology generally managed by MT engine
. Editing effort to make more fluent sentences
. Quality goals: same as with a traditional translation!
.What is Post-Editing?
Post-Editing rules
. If it ain’t broke, don’t fix it
. Keep in mind the goal of the document (e.g. timely information)
. Don’t spend too long on any one issue
. Style issues are not important unless they get in the way of understanding
.What is Post-Editing?
Why don’t translators like post-editing?
.What is Post-Editing?
. Another new tool to learn (like CAT tools)
. Unfamiliar, not much training available
. Fears that they will become obsolete, or make less money
. A different way of working
. Less room for creativity
. An activity requiring lesser skills
. Impression of correcting the same errors, and not those a human would make
What are some positive aspects?
.What is Post-Editing?
. Another new tool to add to your skills
. A chance to become expert in a specialization that will be in greater demand. A chance to vary translation activities. Less routine work. Participating in improving the machine by correcting systematic errors. Improving the MT output over time. Using other skills. Possibility of adding other tasks such as coding dictionaries, writing guidelines, creating regular expression to automatically correct errors
A Final Word about Quality
Once terminology is defined, a trained engine will provide consistent terminology across all documents
MT erases stylistic and terminology differences among multiple translators
Quality can be high because post-editors perform a full edit (source visible)
Post-editors miss more mistakes when editing fuzzy matches
Virtuous circle of feedback improves the engine in real-time
.A final word about quality
Evaluations of MT plus human post-editing
“Contrary to all expectations, using MT in Bentley has improved the translation quality in the pilot projects”
French OLH reviewer: “I give a 9…I find this translation very good…I found it better than the translations I used to see before”
German courseware reviewer: “It was the best translation of courseware I ever read.”
. A final word about quality