Machine Translation at the EPO Removing language barriers...
21
19/09/2013 Machine Translation at the EPO Removing language barriers from patent documentation Paul Schwander European Patent Office The 5th Workshop on Patent Translation, MT Summit 2013
Transcript of Machine Translation at the EPO Removing language barriers...
19/09/2013
Machine Translation at the EPO
Removing language barriers from patent documentation
Paul Schwander
European Patent Office
The 5th Workshop on Patent Translation, MT Summit 2013
Roadmap
The context: why is MT strategic?
Machine Translation @ the EPO:
state of play and future plans
Why Machine Translation?
- Reducing the language
barrier in the European
context: Unitary Patent
system supported by MT.
- Access to global patent
information for prior art
searches.
Global patent filings rising continuously, especially Chinese applications
IP5= Europe, USA, Japan, China and Korea
Accessing Asian languages patent for EPO
examiners
Addressing the Chinese language wall
MT Full-text acquired, ca. 5 milion documents.
An on demand manual translation service offered to examiner
5 million patents: manual translation -> 1 day a patent -> 22 years of work for a team of 1000 translators.
Search in
MTed text
Relevant patents
Order a manual
translation
Detect
Understand
Patent Translate launched on 29 February 2012
System integrated in Espacenet,
the EPO Publication Server and EPOQUE
Patent Translate: how does it work?
Result of a collaboration between the EPO and Google
Patent data represent a huge source of corpora.
Patent documents and their translation/corresponding documents are prepared and stored in a corpora repository.
Translation system is trained using this corpora.
Translation quality assessed before launch: test fit for purpose level.
Translation
Memory
Translation
(Google)
Patent Translate: Architecture
Patent Translate Service
EPO
API (REST)
Corpora
Repository
Espacenet Bulk GPI Examiner tools Publication Server
EPO
«Functionblock»
Business Services::ApplicationBE
EPO«Functionblock»
SysManag.::Monitoring
«Functionblock»
SysManag.::Installation
«file»
ConfigurationFile
«Component»
Superinstaller
uses/provides
«file»
Monitoring::Checks
«document»
InstallationGuideline
Installation-GUI
Actors::ITAdministartor
«Component»
LiveUpdate
Update Plugin
«Component»
EPO::EPO update server
«file»
ExternalSoftware
«file»
OLF Software build
«file»
Installation Packages
provides
provides
Update-GUI
Actors::PatentApplicant
soap via http
«file»
OLF & NO Plugins
CLI
«Component»
ApplicationBE::FileManagerBE
provides
install
National
Patent
Offices
Quality level ranking
Current achievements
Patent Translate now covers translations between English and
21 other languages: Bulgarian, Chinese, Czech, Danish, Dutch,
Finnish, French, German, Greek, Hungarian, Icelandic, Italian,
Japanese, Norwegian, Polish, Portuguese, Romanian, Slovak,
Slovenian, Spanish and Swedish
305 million different machine translations of complete patent
documents can be accessed 'on the fly', using the current
language pairs offered = 1500 years of work for 1000 translators
if done manually
Patent Translate usage
Plans
Project to be completed end of 2014:
32 languages European and Asian.
2013-2014
Turkish, Estonian, Croatian, Latvian,
Lithuanian, Albanian, Macedonian, Serbian,
Russian and Korean
Patent Translate : illustrative example (Description)
MT Example
MT Example
Topics around Patent MT
CLIR or not?
OCR combined with MT for
non-digitised collections
Fit for purpose quality
assessment
Perception of the quality
Conclusion
MT is more than ever a must in the context of
the global patent documentation:
– The size of patent collections to search is
increasing and systematic manual translation
is not an option.
– MT has proven to be fit for purpose
– Quality will continue improving
