Browse Prior Art Database

System and Method to Determine Malicious Crowd Translations

IP.com Disclosure Number: IPCOM000235985D
Publication Date: 2014-Apr-01
Document File: 3 page(s) / 49K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed are a system and method to combat the submission of malicious translations within crowd translation services to improve the overall quality and accuracy of translations.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 47% of the total text.

Page 01 of 3

System and Method to Determine Malicious Crowd Translations

Companies translate owned documentation and web content using known techniques such as automatic machine translation. One of the shortcomings of using a machine translation service is that the machine does not understand, as a human can, the context of the information.

Crowd translation services help with this problem. Crowd translation projects are often led by an individual that speaks one or two languages, but not all of the languages that are being translated. Consequently, crowd translation projects must heavily rely on the crowd to help determine whether the translated portions are malicious. For example, rather than translate the content, a user might input entirely unrelated or otherwise disturbing content. Suppose the sentence to translate is: "This is an example"; a malicious user might translate that sentence (in one language) to, "Try out this new drug free at somewebsite.com." Unless the administrators on the project are reviewing every translation or someone else who speaks the language catches the malicious translation, that new sentence can go live on the company's website. A method is needed to monitor crowd translation

services for incorrect or malicious input.

The solution is a system and method to combat the submission of malicious translations within crowd translation services to improve the overall quality and accuracy of translations.

In the preferred embodiment, a crowd translation service is used per current practices to configure, import, control quality, and publish content for translation. The novel solution modifies this service as follows:

1. For each translator in the crowd, the project administrator(s) establishes a translator trust level threshold. This threshold is used to determine when to accept a translation submitted by the translator. For example:

A. Set a default trust level threshold of 20% for new, unknown translators.

If a translation suggested by this unknown translator has more than 20% of the words different from compared sources, the translation is flagged for further action and review.


B. As this unknown translator gains the confidence and trust of the

administrator (and crowd), increase the trust level threshold for this translator (e.g., to 30%), so that the translator is allowed to make more substantial changes without unnecessary flagging for review.


C. A fully vetted and trusted translator is assigned a trust level threshold

of 100%

2. For each suggested translation of a clause, benchmark translations are determined, including:

A. Previous and accepted translations of the clause

B. If no previous trusted translations are available, then use one or more machine translation services. This is done by running a machine translation (MT) service against the clause.

3. A suggested translation is compared against benchmark translations (e.g., translations previously accepted or MT translations).

A. The system counts each word...