pith. sign in

arxiv: 1906.09000 · v1 · pith:Z5OD35ZEnew · submitted 2019-06-21 · 💻 cs.CL

Demonstration of a Neural Machine Translation System with Online Learning for Translators

Miguel Domingo , Mercedes Garc\'ia-Mart\'inez , Amando Estela , Laurent Bi\'e , Alexandre Helle , \'Alvaro Peris , Francisco Casacuberta , Manuerl Herranz This is my paper

Pith reviewed 2026-05-25 19:04 UTC · model grok-4.3

classification 💻 cs.CL
keywords neural machine translationonline learningpost-editingcomputer-aided translationmodel adaptationproduction environmenttranslator workflow
0
0 comments X

The pith

A neural machine translation system with online learning integrated into professional translation software adapts continuously from user corrections to reduce post-editing effort.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper presents a working demonstration of online learning applied to neural machine translation inside a production workflow. The system receives corrections from translators and uses them to update its models on the fly. The integration connects the translation engine directly to an existing editor so that each edit becomes new training data for future sentences. A reader would care because repeated use in one domain or by one translator could steadily lower the amount of manual work required after the first machine output.

Core claim

The paper demonstrates an end-to-end platform that links neural machine translation servers to SDL Trados Studio and applies online learning so the models update from each translator correction, adapting the output to a specific domain or individual style and thereby saving post-editing effort.

What carries the argument

Online learning updates triggered by translator post-edits inside the integrated CAT environment.

If this is right

  • Models become more accurate for the current domain as translators continue working.
  • Individual translator preferences can be captured without separate fine-tuning runs.
  • The same correction data improves future sentences within the same document or project.
  • Integration keeps the workflow inside familiar editing software rather than requiring new interfaces.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same online update loop could be tested with other computer-aided translation tools.
  • Long-term use might produce measurable divergence between general-domain and user-adapted models.
  • If updates accumulate without periodic resets, drift from the original training distribution could appear.
  • The approach opens a path to measuring adaptation speed as a function of correction volume.

Load-bearing premise

That repeated updates from human corrections will produce steady reductions in post-editing effort without destabilizing the models or demanding impractical computing resources during live use.

What would settle it

A side-by-side measurement of total post-editing time or keystrokes on the same documents before and after several rounds of online updates, showing no net decrease.

Figures

Figures reproduced from arXiv: 1906.09000 by Alexandre Helle, \'Alvaro Peris, Amando Estela, Francisco Casacuberta, Laurent Bi\'e, Manuerl Herranz, Mercedes Garc\'ia-Mart\'inez, Miguel Domingo.

Figure 1
Figure 1. Figure 1: Architecture of our in-house OL framework. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: User Interface from Trados Studio SDL. have to be enabled in the translation provider plu￾gin (see [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Machine translation plugin configuration. [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Example of Qualitivity’s logging file. time and effort. Acknowledgments The research leading to these results has received funding from the Spanish Centre for Technolog￾ical and Industrial Development (Centro para el Desarrollo Tecnologico Industrial) (CDTI) and ´ the European Union through Programa Oper￾ativo de Crecimiento Inteligente (Project IDI￾20170964). We gratefully acknowledge the sup￾port of NVID… view at source ↗
read the original abstract

We introduce a demonstration of our system, which implements online learning for neural machine translation in a production environment. These techniques allow the system to continuously learn from the corrections provided by the translators. We implemented an end-to-end platform integrating our machine translation servers to one of the most common user interfaces for professional translators: SDL Trados Studio. Our objective was to save post-editing effort as the machine is continuously learning from human choices and adapting the models to a specific domain or user style.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 0 minor

Summary. The paper presents a demonstration of an end-to-end platform that integrates online learning for neural machine translation servers with SDL Trados Studio. Translators' post-edits are used to continuously update the models so that the system adapts to a specific domain or user style, with the stated objective of reducing post-editing effort.

Significance. A production-ready integration of incremental NMT adaptation inside a widely used CAT tool would be of practical interest to the translation industry. However, because the manuscript contains no quantitative measurements of effort reduction, translation quality, update stability, or resource cost, it is not possible to determine whether the claimed benefit is realized.

major comments (1)
  1. [Abstract] Abstract: the manuscript states that the objective is to save post-editing effort through continuous learning from translator corrections, yet supplies no before/after metrics (e.g., TER, time per segment, keystroke counts), no stability analysis of incremental updates, and no resource profiling of the live SDL Trados integration.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the review. This is a demonstration paper focused on the technical integration of online NMT adaptation with SDL Trados Studio; we address the comment on metrics below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the manuscript states that the objective is to save post-editing effort through continuous learning from translator corrections, yet supplies no before/after metrics (e.g., TER, time per segment, keystroke counts), no stability analysis of incremental updates, and no resource profiling of the live SDL Trados integration.

    Authors: The manuscript is a system demonstration describing the end-to-end platform and its integration. The objective statement describes the intended use case and motivation for the work, but the paper does not present quantitative evaluations of post-editing effort, translation quality, update stability, or resource usage. Such measurements would require a separate experimental study with controlled conditions, which falls outside the scope of a demonstration paper. We therefore do not claim empirical results on effort reduction in this work. revision: no

Circularity Check

0 steps flagged

No circularity: system description with no derivations or fitted quantities

full rationale

The paper is a demonstration and integration description of an online-learning NMT system with SDL Trados Studio. It states an objective (saving post-editing effort via continuous adaptation) but contains no equations, no parameter-fitting steps, no uniqueness theorems, and no derivation chain that could reduce to its own inputs. No load-bearing claims are justified by self-citation or by renaming fitted results as predictions. The manuscript is self-contained as an engineering report; absence of quantitative evaluation is a separate correctness issue, not circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No free parameters, axioms, or invented entities are introduced because the paper is a descriptive systems demonstration without mathematical modeling or new theoretical constructs.

pith-pipeline@v0.9.0 · 5631 in / 963 out tokens · 24699 ms · 2026-05-25T19:04:21.014057+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

17 extracted references · 17 canonical work pages · 2 internal anchors

  1. [1]

    URL: " 'urlintro :=

    ENTRY address author booktitle chapter edition editor howpublished institution journal key month note number organization pages publisher school series title type volume year eprint doi pubmed url lastchecked label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block STRINGS urlintro eprinturl eprintpr...

    work page

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

    work page

  3. [3]

    Leiva, Bartolom \'e Mesa-Lao, Daniel Ortiz-Mart \'i nez, Herv \'e Saint-Amand, Germ \'a n Sanchis-Trilles, and Chara Tsoukala

    Vicent Alabau, Ragnar Bonk, Christian Buck, Michael Carl, Francisco Casacuberta, Mercedes Garc \'i a-Mart \'i nez, Jes \'u s Gonz \'a lez-Rubio, Philipp Koehn, Luis A. Leiva, Bartolom \'e Mesa-Lao, Daniel Ortiz-Mart \'i nez, Herv \'e Saint-Amand, Germ \'a n Sanchis-Trilles, and Chara Tsoukala. 2013. CASMACAT : An open source workbench for advanced compute...

    work page 2013

  4. [4]

    Ana Guerberof Arenas. 2008. Productivity and quality in the post-editing of outputs from translation memories and machine translation. Localisation Focus, 7(1):11--21

    work page 2008

  5. [5]

    Robert Dale. 2016. How to make money in the translation business. Natural Language Engineering, 22(2):321--325

    work page 2016

  6. [6]

    Miguel Domingo, Mercedes Garc \'i a-Mart \'i nez, \'A lvaro Peris, Alexandre Helle, Amando Estela, Laurent Bi \'e , Francisco Casacuberta, and Manuel Herranz. 2019. Incremental adaptation of NMT for professional post-editors: A user study. In Proceedings of the Machine Translation Summit. Under publication

    work page 2019

  7. [7]

    Marcello Federico, Nicola Bertoldi, Mauro Cettolo, Matteo Negri, Marco Turchi, Marco Trombetti, Alessandro Cattelan, Antonio Farina, Domenico Lupinetti, Andrea Martines, Alberto Massidda, Holger Schwenk, Lo\" i c Barrault, Frederic Blain, Philipp Koehn, Christian Buck, and Ulrich Germann. 2014. The matecat tool. In Proceedings of the 25th International Co...

    work page 2014

  8. [8]

    Hany Hassan, Anthony Aue, Chang Chen, Vishal Chowdhary, Jonathan Clark, Christian Federmann, Xuedong Huang, Marcin Junczys-Dowmunt, William Lewis, Mu Li, et al. 2018. Achieving human parity on automatic chinese to english news translation

    work page 2018

  9. [9]

    Ke Hu and Patrick Cadwell. 2016. A comparative study of post-editing guidelines. In Proceedings of the 19th Annual Conference of the European Association for Machine Translation, pages 34206--353

    work page 2016

  10. [10]

    Guillaume Klein, Yoon Kim, Yuntian Deng, Jean Senellart, and Alexander M. Rush. 2017. Open NMT : Open-source toolkit for neural machine translation. In Proceedings of the Association for the Computational Linguistics, pages 67--72

    work page 2017

  11. [11]

    Philipp Koehn and Rebecca Knowles. 2017. Six challenges for neural machine translation. In Proceedings of the First Workshop on Neural Machine Translation, pages 28--39

    work page 2017

  12. [12]

    Sachith Sri Ram Kothur, Rebecca Knowles, and Philipp Koehn. 2018. Document-level adaptation for neural machine translation. In Proceedings of the 2nd Workshop on Neural Machine Translation and Generation, pages 64--73

    work page 2018

  13. [13]

    \'A lvaro Peris and Francisco Casacuberta. 2018. Online learning for effort reduction in interactive neural machine translation. Accepted in Computer Speech & Language

    work page 2018

  14. [14]

    \'A lvaro Peris, Luis Cebri \'a n, and Francisco Casacuberta. 2017. Online learning for neural machine translation post-editing. arXiv:1706.03196

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  15. [15]

    Marco Turchi, Matteo Negri, M Amin Farajian, and Marcello Federico. 2017. Continuous learning from human post-edits for neural machine translation. The Prague Bulletin of Mathematical Linguistics, 108(1):233--244

    work page 2017

  16. [16]

    Y. Wu , M. Schuster , Z. Chen , Q. V. Le , M. Norouzi , W. Macherey , M. Krikun , Y. Cao , Q. Gao , K. Macherey , J. Klingner , A. Shah , M. Johnson , X. Liu , . Kaiser , S. Gouws , Y. Kato , T. Kudo , H. Kazawa , K. Stevens , G. Kurian , N. Patil , W. Wang , C. Young , J. Smith , J. Riesa , A. Rudnick , O. Vinyals , G. Corrado , M. Hughes , and J. Dean ....

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  17. [17]

    Joern Wuebker, Patrick Simianer, and John DeNero. 2018. Compact personalized models for neural machine translation. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 881--886

    work page 2018