Annotations can be generated automatically using one of our annotators. You can create tasks to annotate either individual words or full projects with labels of semantic proximity. If you create a project task with the same project or if you create a word task with the same project and the same word, the old annotations will be overridden by the new annotations. You can always check the status of the task on the task overview page. If your tasks are failing repeatedly, don't hesitate to contact us. It is known that large projects can lead to problems.
Random: samples a random integer between 1 and 4 with uniform probability.
XL-Lexeme: XL-Lexeme is a bi-encoder that vectorizes the input sequences using a XLMR-based Siamese Network. It is trained to minimize the contrastive loss with cosine distance on several WiC datasets. There are two versions of XL-Lexeme: XL-Lexeme-Cosine returns the cosine similarity between word vectors, while XL-Lexeme-Binary predicts either value 1 or 4 based on thresholding cosine similarity between vectors. XL-Lexeme-Multi-Threshold takes three threshold values to distinguish all values of the DURel scale (1-4).