Related Links

 

Related Projects

CrackerCracker
QT21 is collaborating wirth CRACKER on shared resources. CRACKER pushes towards an improvement of MT research in terms of efficiency and effectiveness by implementing the successful example of other disciplines where massively collaborative research on shared resources – guided by interoperability, standardisation, agreed major challenges and comprehensive success metrics – has led to breakthroughs that would have been impossible otherwise.
from
01/2015
to
12/2017
EuroMatrixPlusEuroMatrixPlus
QT21 is using Moses PBSMT and HPB-SMT systems originally developed under EM+ (and later MosesCore – see below) as baseline systems in QT21 WPs 1-4. Early EM+ research on rich tree-based SMT in EM+ informs some of the research tasks in QT21 WP1. EM+ research on continuous space language models informs the use of RNNs and deep learning in both translation and language models in QT21 WPs 1-3.
from
2009
to
2012
META-NETMETA-NET
QT21 uses key findings from the META-NET Language White Papers to concentrate on “difficult” languages and language pairs that exhibit complex morphology and strong syntactic variation and/or lack of sufficient training or language technology resources. The concept underlying QT21 initially grew out of the MT part of the META-NET Strategic Research Agenda (SRA) Planning process including the Technology Council and Vision Groups.
from
02/2010
to
01/2013
QTLaunchpadQTLaunchPad
In more than one way QT21 is the brain-child of the QTLaunchPad Planning Panels: these panels consist of a wide-ranging cross section of leading MT researchers and MT stake-holders including industrial users and providers of MT technologies as well as professional organisations. QTLaunchPad is tasked to continue the MT part of the META-NET SRA Planning process. From QTLaunchPad, the QT21 concept “inherits” its strong focus on High-Quality Outbound Translation, the guiding principle of systematic identification and sub-sequent removal of quality barriers and the close involvement of human professional translators in defining novel error typologies (MQM and DQF), evaluation and continuous learning from post-editing in WP3. QTLaunchPad research on QE (the Quest system) will provide the base-line for QT21 QE in WP3.
from
07/2012
to
06/2014
MosesCoreMosesCore
MosesCore continues the development of the open source Moses SMT toolkit, ensuring that advances in MT technologies are reflected in the Moses system as early as possible. QT21 will use the latest Moses System technologies as strong baselines in WPs 1-4.
from
02/2012
to
01/2015
matecat
CASMACAT
matecat and CASMACAT
Early research on harvesting post-edits and using them to incrementally adapt the MT model will inform parts of the QT21 WP3 research. MateCat baselines will be used in WP3.
from
11/2011
to
10/2014
EU-BridgeEU-Bridge
QT21 will use already existing EU-Bridge base-lines in language portability and data dependency reduction research in WP2.
from
02/2012
to
01/2015
QTLeapQTLeap
QT21 and QTLeap are complementary: where QTLeap research concentrates on the use of deep linguistic analyses (based in particular on deep parallel treebanks) and linked open data resources to improve machine translation, QT21 is data-driven and machine-learning focused, concentrating on model improvements and applications to morphologically rich, syntactically diverse and resource scarce environments. QT21 will follow, compete and explore synergies with ongoing QTLeap research.
from
11/2013
to
10/2016