Text-Fabric

Overview

Text-Fabric is machinery for processing such corpora as annotated graphs. It treats corpora and annotations as data, much like big tables, but without loosing the rich structure of text, such as embedding and multiple representations. It deals with text in a state where all markup is gone, but where the complete logical structure still sits in the data.

Whether a corpus comes from plain texts, OCR output, databases, XML, TEI: Text-Fabric has support to convert it to single column files, where each file corresponds with a feature of the text.

The Python library tf can be used to collect a bunch of features and display it as an annotated text. What ties the features together are natural numbers, that serve to anchor the elementary positions in the text as well as the relevant structures within the text.

When Text-Fabric loads a dataset of features, you can instruct it to get the features from anywhere. That means it supports workflows where annotations are produced by third parties and can be used against the original corpus, without additional work. It also facilitates mappings between ongoing versions of the corpus, so that annotations made on older versions can be ported to newer versions without redoing the annotation creation.