Building an Automated Movie Transcription and Machine Translation Platform, by Takeyoshi Nakayama, Human Science

A lot of companies have training and support materials in movie format. And the number of the movies keeps growing. Making movies is easy. Even a smartphone can record and edit a FullHD video. But how about translating movies? Although YouTube has all the functions we need to auto-transcribe and machine-translate movies, we often need to complete the task internally in the organization due to security and batch-process requirements. Find out how we built our own automatic movie transcription/machine translation system and what challenges we faced.

1. Building an Automated Movie
Transcription and Machine Translation
Platform
TakeyoshiNakayama
Human Science Co., Ltd.
WWW.SCIENCE.CO.JP

2. .
WWW.SCIENCE.CO.JP
.
Topics
Demo
Custom Movie Translation Platform
Advantage / Disadvantage
Statistics
Future Challenges
Q&A

3. .
WWW.SCIENCE.CO.JP
.

4. .
WWW.SCIENCE.CO.JP
.

5. .
WWW.SCIENCE.CO.JP
.
Components
Transcription
Audio extraction
Audio conversion
Transcription
Manual correction
Translation
Text conversion
Tag/Entity/Terminology
Machine Translation
Post-edit
Subtitle /
Voice Over
Time stamp creation
SRT file generation
Voice generation
Video/Audio merging

6. .
WWW.SCIENCE.CO.JP
.

7. .
WWW.SCIENCE.CO.JP
.
Advantage / Disadvantage
・Lower cost
・Shorter schedule
・Can handle many movies at once
・Glossary and TM can be applied
・Customizability
・Setup and maintenance cost

8. .
WWW.SCIENCE.CO.JP
.

9. .
WWW.SCIENCE.CO.JP
.
Statistics: Transcription
HS is better than YouTube
Both have errorsand manual correction is needed
0
5
10
15
HS YouTube
Edit Distance
(The lower the score, the higher the accuracy)
Edit Distance

10. .
WWW.SCIENCE.CO.JP
.
Statistics: Transcription Time
Schedule Comparison
42
100
0 20 40 60 80100
With Platform
Without Platform
%
58% reduction

11. .
WWW.SCIENCE.CO.JP
.
Transcription Samples
Correct HS YouTube
You can modify dashboardsby
editing or deleting content.
You can modify dashboardsby
editing or deleting content.
you can modify dashboardby
editing or deleting content
You can choose from:
* Forms, which includes pre-
defined form types
* Charts, which includes pre-
defined chart types
* Links to external websites
that automatically update to
display current data, and
custom text you can use to
elaborate on your data.
you can choose from forms
which includes predefined form
types chartswhich includes
predefined chart types linksto
external websites that
automatically update to display
current data and custom text.
You can useto elaborate on
your data.
you can choose from forms
which includes pre-defined
form types chargewhich
includes pre-defined chart
types linksto external website
that automatically update to
display current data and
custom text you use to
elaborate on your data

12. .
WWW.SCIENCE.CO.JP
.
Statistics: Translation
0%
20%
40%
60%
80%
100%
HS SMT RBMT
1. Perfectly
Understandable
2. Fully
Understandable
3. Barely
Understandable
4. Not
Understandable

13. .
WWW.SCIENCE.CO.JP
.
Statistics: Translation Time
Schedule Comparison
90
100
0 20 40 60 80100
With Platform
Without Platform
%
10% reduction

14. .
WWW.SCIENCE.CO.JP
.
Wrapping up results
Custom movie translation platform can reduce time
and cost
Best used for support contents or procedure
documents

15. .
WWW.SCIENCE.CO.JP
.

16. .
WWW.SCIENCE.CO.JP
.
Future Challenges
Using movie translation platform for other types of
documents than procedures
Collecting of speaking bilingual corpus

17. .
WWW.SCIENCE.CO.JP
.
17
www.science.co.jp/english
[phone]
[email protected]