Determining optical character recognition parameters

A non-transitory computer-readable storage device having instructions stored thereon that, when executed by a computing device, cause the computing device to perform operations comprising:

• performing a first optical character recognition process with an optical character recognition engine on one or more pages of a document to generate a first output response for each of multiple writing systems that each comprise a character set that is associated with one or more natural languages, wherein the first optical character recognition process uses a first configuration of the optical character recognition engine, and wherein the first configuration of the optical character recognition engine configures the optical character recognition engine to recognize, for each of the writing systems, a limited subset of characters from the character set of the writing system;
• using a trained classifier to identify, from among the multiple writing systems, a dominant writing system of the document based on the first output responses generated by the first optical character recognition process;
• after identifying the dominant writing system of the document and before selecting a dominant natural language of the document, reconfiguring the optical character recognition engine from the first configuration to a second configuration that is different from the first configuration, wherein the second configuration of the optical character recognition engine configures the optical character recognition engine to recognize the full character set that is associated with the one or more natural languages associated with the dominant writing system of the document;
• performing a second optical character recognition process with the reconfigured optical character recognition engine on one or more pages of the document to generate a second output response, wherein the second optical character recognition process uses the second configuration of the optical character recognition engine;
• and
• selecting the dominant natural language from the one or more natural languages associated with the dominant writing system by applying one or more statistical language models to the second output response.

A system comprising:

• one or more data processing apparatus;
• and
• a computer-readable storage device having stored thereon instructions that, when executed by the one or more data processing apparatus, cause the one or more data processing apparatus to perform operations comprising:
• performing a first optical character recognition process with an optical character recognition engine on one or more pages of a document to generate a first output response for each of multiple writing systems that each comprise a character set that is associated with one or more natural languages, wherein the first optical character recognition process uses a first configuration of the optical character recognition engine, and wherein the first optical character recognition configuration of the optical character recognition engine configures the optical character recognition engine to recognize, for each of the writing systems, a limited subset of characters from the character set of the writing system;
• using a trained classifier to identify, from among the multiple writing systems, a dominant writing system of the document based on the first output responses generated by the first optical character recognition process;
• after identifying the dominant writing system of the document and before selecting a dominant natural language of the document, reconfiguring the optical character recognition engine from the first configuration to a second configuration that is different from the first configuration, wherein the second configuration of the optical character recognition engine configures the optical character recognition engine to recognize the full character set that is associated with the one or more natural languages associated with the dominant writing system of the document;
• performing a second optical character recognition process with the reconfigured optical character recognition engine on one or more pages of the document to generate a second output response, wherein the second optical character recognition process uses the second configuration of the optical character recognition engine;
• and
• selecting the dominant natural language from the one or more natural languages associated with the dominant writing system by applying one or more statistical language models to the second output response.