Topic 4: The Magicians Hat: Turning Data into Business Intelligence (2)

Turning Data into Business Intelligence

Konstantin DranchMemsource

Building BI in a translation toolOur approach

1. Separate language data from metadata

visualisation tools API 3rd party BI

Memsourcetranslation platform

Elastic Searchdata warehouse

2. DIY metrics & dashboard builders

3. Embedding data into process

4. Higher data resolution over time

1. Project data

2. Job data

3. Segment data

4. Time tracking

5. Subsegment data

Thank you!

konstantin@memsource.com

Cloud tools lead to Big Data

Server tools: private data silos Cloud tools: centralized data

And the clouds are getting bigger…

In May alone, users processed 0.8 billion words in Memsource

Benchmarks: translation memory

Data for jobs where post-editing analysis has been performed, December 2015 - May 2016

Sample9 bnwords

Benchmarks: machine translation

Sample size 20 million words, December 2015 - May 2016

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

en:es pt:en en:pt es:en en:ru ru:en en:de pt:es en:fr es:pt

% O

F W

OR

DS

IN S

EGEM

ENTS

FR

OM

MT

SAMPLE LANGUAGE PAIRS

EDIT DISTANCE FOR MAJOR LANGUAGE PAIRS

mt.match100 mt.match95 mt.match85 mt.match75 mt.match50 mt.match0

MT not used

Raw MT

Moderate edits

Heavily edited

Benchmarks: rising languages% increase Language Volume67,481.6 Marathi *3,087.7 Hindi *1,006.0 Latvian **784.3 Kazakh *689.1 Serbian *659.5 Finnish **657.5 Romanian *613.9 Vietnamese *592.5 Indonesian *585.7 Turkish **543.6 Hungarian *467.4 Italian ***457.2 Ukrainian *448.9 Chinese ****440.7 Swedish **428.8 Dutch **390.0 Norwegian *378.9 English *****335.2 Danish **

% increase Language Volume331.3 Estonian *319.2 Polish **318.5 Norwegian Bokmål *307.8 German ***306.9 Thai *299.7 Overall increase in volume295.8 Korean **269.4 Czech **268.3 Russian ****245.5 French ****241.4 Slovak *237.8 Portuguese ***237.4 Bulgarian *230.0 Arabic *200.9 Japanese ****194.8 Spanish ****137.8 Bulgarian *121.4 Greek ***

-