BNP Paribas Case Study

bnp-squareBNP Paribas was ranked by Forbes as the world’s third largest bank in 2012. The Group has operations in nearly 80 countries and some 200,000 employees.

The Challenge

BNP Paribas manages many thousands of publications in a whole host of different languages every year. In order to present a homogenous image throughout the world, BNP Paribas has made a significant investment in machine translation, which is integrated into its content-management process. Available via intranet as a self-service model for its worldwide employees, their machine translation engine allows BNP Paribas to offer confidential and secure MT access behind their firewall.

In order to use their MT engine to the fullest extent, BNP Paribas was interested in improving its processes in order to increase translation quality and accuracy and reduce the time spent reworking translations.

BNP Paribas selected LexWorks (Lexcelera-Eurotexte Group) to field a test program to determine if better machine translation could help them meet certain business goals. Specifically, BNP Paribas wanted to determine if training and customization by the expert team from LexWorks could improve upon the speed and quality of machine translation used at the bank.

LexWorks was asked to customize the existing MT engine based on terminology validated by BNP Paribas. This included terminology specific to the BNP’s different business activities such as their French retail banking group (BDDF).

The Solution

LexWorks designed a global architecture and dictionaries using Systran 7 Enterprise Server to integrate optimal use of machine translation within BNP Paribas. LexWorks also created a translation profile tailored to domain-specific terminology for BDDF’s retail activities.

Created on the basis of a corpus of 426,787 words and set up in just 6 weeks, the Systran engine was trained by LexWorks using 3 types of resources:

  • Standard dictionaries and bilingual BNP Paribas business sector dictionaries
  • Normalization dictionaries to correct the source or target text
  • A statistical language model

LexWorks also set up a system to maintain these dictionaries so that the translation engine continually improved as new projects were fed into it by in-house users and/or external post-editors. This turned out to be critical to the successful deployment of machine translation within BNP Paribus.

The Result

The test exceeded its goals with:

  • 44% of sentences translated automatically with no need for the services of an editor
  • Only 33% of machine-translated sentences needed to be proofread by a third-party
  • Only 23% of sentences required more in-depth correction

Furthermore, the new system for machine translation resulted in significant time-savings – an increase of productivity of nearly 50%.

This solution enables BNP Paribas to manage machine translation of many confidential documents in-house, along with internal corrections, safely and securely behind their own firewall – and at substantially reduced translation costs.