remarx

Summary

A text-reuse detection tool built on semantic embeddings (the numerical representation of language that powers large language models), remarx enables scholars to surface and connect related texts across documents, regardless of language.

Research Challenge

Older tools for tracking citation or reuse rely on literal comparison of characters, making reuse across language or in paraphrase invisible and limiting the questions that can be asked and the texts that can be compared. needed a way to trace reuse across languages, and so developed a method using semantic similarity to find quotes across texts that previously could not be compared.

CDH’s Technical Innovation

Semantic embeddings, a foundational component of large language models, make linking texts with similar meanings possible, even when the characters or words differ entirely. The CDH is developing remarx, a software pipeline with a user-friendly interface that combines semantic search with rule-based approaches to work across languages and with non-exact usage. This enables scholars to find examples of reuse between two corpora that find quotes that were invisible to the existing tools.

Collaborative Origins

Edward Baring (Princeton, History) approached the CDH to see if computational tools could assist him in tracing the variety of ways Marxist scholars have used the language of Marx and how those practices diverge among different schools of Marxist thought. The reuse data from the remarx tool will inform his “Citing Marx” project.

How you can use this tool

Beyond the Citing Marx project, remarx enables scholars to find examples of reuse between two texts or collections of texts, regardless of content or language, within an interface designed for a non-technical user. We have released the tool on GitHub, making it available to all researchers.

Tracing the movement of words and ideas over time is a key task within humanities research and especially within the digital humanities, as exemplified by projects such as Viral Texts (passim) and “The Spine of American Law” (textreuse), which trace the movement of words and ideas in the press and in the legal code.