Tekstiviesti: Scalable syntactic inductive biases for neural language models