Eurobarometer
Flash 2.0

EUROBAROMETER_FLASH_2.0
├── 01_SOURCE
│   └── Flash-XXX-TEST
│       ├── DSE20T01_EU.xlsm
│       └── Flash-XXX-TEST_TAVF_xxx-XXX_TEMPLATE.xlsx
├── ...
├── 08_FLASH_PROJECTS
│   └── Flash-XXX-TEST
│       ├── 01_TRA
│       │   ├── esp-ESP
│       │   │   ├── 01_To_Translator
│       │   │   ├── ...
│       │   │   └── 05_Machine_Translation
│       │   │       ├── Eurobarometer_Flash-XXX-TEST_esp-ESP_OMT
│       │   │       │   ├── script_output
│       │   │       │   │   └── Eurobarometer_FLASH-XXX-TEST_esp-ESP_OMT (en - es).xls
│       │   │       │   ├── source
│       │   │       │   │   └── DSE20T01_EU.xlsm.html
│       │   │       │   ├── target
│       │   │       │   │   ├── DSE20T01_EU.xlsm.html
│       │   │       │   └── tm
│       │   │       │       └── auto
│       │   │       │           └── EB_fr-BE_TM.tmx
│       │   │       ├── Eurobarometer_Flash-XXX-TEST_esp-ESP_OMT.omt
│       │   │       └── Flash-XXX-TEST_TAVF_esp-ESP_TEMPLATE.xlsx
│       │   ├── ...
│       └── 02_ADA
│           ├── ...
├── 10_DELIVERABLE
│   └── Flash-XXX-TEST
│       └── DSE20T01_EU.xlsm
└── _tech
    ├── DSE20T01_EU.xlsm.html
    ├── Eurobarometer_Flash-XXX-TEST_xxx-XXX_en_OMT.omt
    ├── Eurobarometer_Flash-XXX-TEST_xxx-XXX_fr_OMT.omt
    ├── _assets
    │   └── FL419_TM
    │       ├── EB_es-ES_TM.tmx
    │       ├── ...
    └── config.ods

Why is this project challenging?

  • Multilingual source, with only some parts on scope for translation
  • Many target languages
  • HTML content embedded in Excel file
  • Fast turnaround time (3 working day delivery)
  • Frequent requests
  • Cryptic language codes
  • The client provides translations and expects consistency

* docs & preferences

* workflow initiation

* file preparation

* xdiff reports

* post-processing

* language assets

contents

* docs & preferences

* workflow initiation

* file preparation

* xdiff reports

* post-processing

* language assets

docs & preferences

* docs & preferences

* workflow initiation

* file preparation

* xdiff reports

* post-processing

* language assets

contents

workflow initiation

unzip

backup

* docs & preferences

* workflow initiation

* file preparation

* xdiff reports

* post-processing

* language assets

contents

file/project prepping

 Welcome to this survey!

 Select the right answer.

 Strongly agree.

 Click here to continue.

 Select the right answer.

extraction.

 Welcome to this survey!

 Select the right answer.

 Strongly agree.

 Click here to continue.

 Select the right answer.

 Welcome to this survey!

 Select the right answer.

 Strongly agree.

 Click here to continue.

 Select the right answer.

extraction

extraction..

 Welcome to this survey!

 Select the right answer.

 Strongly agree.

 Click here to continue.

 Select the right answer.

 Welcome to this survey!

 Select the right answer.

 Strongly agree.

 Click here to continue.

 Select the right answer.

extraction...

to XLIFF or not to XLIFF

Pros of XLIFF

  • User cannot modify files
  • User has freedom to use their tool of choice

OmegaT (unlike OLT and other tools) is not just an XLIFF editor, but a fully-fledged CAT tool.

That means one has the choice whether to prepare files as XLIFF or not (provided there's a filter in OmegaT for that file type).

Cons of XLIFF

  • No preview
  • One more dependency (e.g. Okapi)

MRT format

  • Excel file, HTML content
  • One column per language
  • Translation needed?

Solution

  • Extract source text column
  • Transfer text to HTML file:
    • cell -> div
    • cell row number -> div's id
    • translation needed -> translate=yes, class=hide
    • css: layout for preview
    • js: hides untranslatable rows, direction for RTL
  • For each version, back convert translated HTML to Excel column

What we need to translate

What the filter sees

Text

MRT (Excel)
the original file

HTML
(OmegaT's source file)

* docs & preferences

* workflow initiation

* file preparation

* xdiff reports

* post-processing

* language assets

contents

xdiff reports

* docs & preferences

* workflow initiation

* file preparation

* xdiff reports

* post-processing

* language assets

contents

post-processing

 Welcome to this survey!

 Select the right answer.

 Strongly agree.

 Click here to continue.

 Select the right answer.

 Welcome to this survey!

 Select the right answer.

 Strongly agree.

 Click here to continue.

 Select the right answer.

 Tout à fait d'accord.

 Bienvenue à cette enquête!

 Sélectionnez votre réponse.

 Sélectionnez votre réponse.

 Click here to continue.

 Tout à fait d'accord.

 Bienvenue à cette enquête!

 Sélectionnez votre réponse.

 Sélectionnez votre réponse.

 Click here to continue.

merge

merge...

Back-translation

  • Generate translation in OmegaT
  • Send translation to MT engine to produce back-translation
  • Put translation and back-translation in TAVF
  • Done with initial and final translation (steps 3 and 5)

* docs & preferences

* workflow initiation

* file preparation

* xdiff reports

* post-processing

* language assets

contents

language assets

on initiation

on post-processing

Google’s privacy policy

When people use our products they trust us with their information, and it’s our job to do right by them. This means always being thoughtful about what data we use, how we use it, and how we protect it.

  • GT stores inputs to build its vocabulary base
  • GT does not display users' data as search results to Google search engine users

results: is

it working out?

Benefits

  • Speed
  • Precision (less errors to fix)
  • Durability

 

Costs

  • Rigid steps (lack of freedom)
  • Strict adherence to certain patterns
  • Fine-tuning takes some time

In a nutshell:
Free people up for more meaningful pursuits

Learning new behaviours

When does automation make sense?

To automate repetitive actions, some actions need to be repeated exactly many times, all through the project (2-3 years long). We automate them and forget about them. But is this premise realistic?

Risks and caveats

What is supposed to be a means of human emancipation or higher productivity (for PMs to be freer to focus on non-repetitive tasks) might end up being an slavering factor for the engineer(s)   if requirements are unclear or erratic.

other models

FLASH

  • Release-based
    • many independent workflows
    • triggered by a new init bundle

PaRiS

  • Rolling-based
    • one permanent workflow that is updated over time
    • triggered by a new XLIFF

tech