How does textract determine when to segment text vertically or horizontally?

As of the moment, I am unaware of this parameter. However if you are mostly expecting horizontal text, you might want to test the image above to AWS Rekognition text detection instead https://docs.aws.amazon.com/rekognition/latest/dg/text-detection.html

As of now, Textract does not support vertical text alignment. Please refer to : https://docs.aws.amazon.com/textract/latest/dg/limits.html for more information. The Textract algorithm will perform better on sparser text use cases so we recommend padding the image to improve performance.