Can AWS Textract handle Greek letters?

Question

Hello,
I'm using Textract for extracting information from academic papers. It's working great except for when Greek letters are used, like α, β, σ, etc. It will often translate α as *a*, and β will be completely dropped. I know AWS supports only five or so languages, none of which are Greek, but I wanted to confirm if AWS Textract does have difficulty with Greek letters, and if there's an easy workaround.

Answer

Hello,

It seems like you're encountering issues with Amazon Textract's handling of Greek letters in academic papers. Textract currently only supports a limited number of languages, and these ML models often have difficulties recognizing certain characters. In Textract, there is a feature known as Custom queries, which might help resolve the problem.

Custom queries allow us to customize the output of the text extraction using adapters. With adapters, you can train Amazon Textract's deep learning model on domain-specific documents and queries to improve accuracy.

Here is a link to a guide on how to implement these custom queries

https://aws.amazon.com/blogs/machine-learning/customize-amazon-textract-with-business-specific-documents-using-custom-queries/

Answer

Hi There

Textract currently supports English, Spanish, German, Italian, French, and Portuguese.  Handwriting, Invoices and Receipts, Identity documents and Queries processing are in English only

See https://aws.amazon.com/textract/faqs/

Can AWS Textract handle Greek letters?

Conteúdo relevante