We study annotation projection in text classification problems where source documents are published in multiple languages and may not be an exact translation of one another. In particular, we focus on the detection of unfair clauses in privacy policies and terms of service. We present the first English-German parallel asymmetric corpus for the task at hand. We study and compare several language-agnostic sentence-level projection methods. Our results indicate that a combination of word embeddings and dynamic time warping performs best.

This article explores the potential of artificial intelligence for identifying cases where digital vendors fail to comply with legal obligations, an endeavour that can generate insights about business practices. While heated regulatory debates about online platforms and AI are currently ongoing, we can look to existing horizontal norms, especially concerning the fairness of standard terms, which can serve as a benchmark against which to assess business-to-consumer practices in light of European Union law. We argue that such an assessment can to a certain extent be automated; we thus present an AI system for the automatic detection of unfair terms in business-to-consumer contracts, a system developed as part of the CLAUDETTE project. On the basis of the dataset prepared in this project, we lay out the landscape of contract terms used in different digital consumer markets and theorize their categories, with a focus on five categories of clauses concerning (i) the limitation of liability, (ii) unilateral changes to the contract and/or service, (iii) unilateral termination of the contract, (iv) content removal, and (v) arbitration. In so doing, the paper provides empirical support for the broader claim that AI systems for the automated analysis of textual documents can offer valuable insights into the practices of online vendors and can also provide valuable help in their legal qualification. We argue that the role of technology in protecting consumers in the digital economy is critical and not sufficiently reflected in EU legislative debates.