Lettria

Changed
- New document format ! Our API now features a brand new format that hold information for a whole document.
- Document coreference resolution: Coreference spans and clusters are available on your whole document.
- New document method: replace_coreference() will allow you to replace spans by the head reference of their cluster.
- New global attributes: spans and clusters
- New document attributes : str_doc, original_text_doc, emotion_doc, sentiment_doc, domain_doc, type_doc, original_text_doc, emoticon_doc
- New tokens attributes : auxiliary, gender, plural, infinitive, mode, conjugate

Added:
- to_annotation_format : Exports the data to the input format of our annotation plateform.
- reformat_data: Reformat old data results to the new document format.

Fixed
- subsentence str property is now working correctly.

Added
- add_documents() method used to batch together different documents for faster inference

Fixed
- subsentence is now available again when 'sentiment' module is not activated.

Added
- NLP.reset_data() allows to reset the nlp object and erase all documents data.
Changed
- Multiple NLP instances can be used at the same time, documents ids are not common anymore accross all NLP instances.
- Changes to calculation of average for word_emotion, meaning_emotion, word_sentiment, meaning_sentiment.
- emotion and emotion_ml attributes now return different values. 'emotion_ml' stays the same while 'emotion' is now a mix of our ML model with another algorithm.
- Fix for subsentences tokens.

Added
- New argument 'skip_errors' for match_pattern.
- New attribute 'original_text' for retrieving input text before tokenization.