DReAM

This space allows you to test a set of LLMs tuned to perform different tasks over dream reports. Three main tasks are available:

• Name Entity Recognition (NER), with an English-only model that generates the identified characters.

• Sentiment Analysis (SA), with two English-only models (one for multi-label classification, and one for generation) and a large multilingual model for multi-label classification.

• Relation Extraction (RE), with an English-only model that identifies relevant characters and existing relations between them following the Activity feature of the Hall and Van de Castle framework.

All models have been tuned on the Hall and Van de Castle framework. More details are on the page for each model. For more techincal and theoretical details, see the Bertolini et al., 2024, and Bertolini et al., 2024.

Use the current interface to check if a language is included in the multilingual SA model, using language acronyms (e.g. it for Italian). the tabs above will direct you to each model to query.

If you want to use the models outside the space, you can easily do so via DReAMy

Anonymise

This space allows you to test a set of LLMs tuned to perform different tasks over dream reports. Three main tasks are available:

• Name Entity Recognition (NER), with an English-only model that generates the identified characters.

• Sentiment Analysis (SA), with two English-only models (one for multi-label classification, and one for generation) and a large multilingual model for multi-label classification.

• Relation Extraction (RE), with an English-only model that identifies relevant characters and existing relations between them following the Activity feature of the Hall and Van de Castle framework.

All models have been tuned on the Hall and Van de Castle framework. More details are on the page for each model. For more techincal and theoretical details, see the Bertolini et al., 2024, and Bertolini et al., 2024.

Use the current interface to check if a language is included in the multilingual SA model, using language acronyms (e.g. it for Italian). the tabs above will direct you to each model to query.

If you want to use the models outside the space, you can easily do so via DReAMy

NER

A T5 model tuned to perform text generation, and predict which characters are present in the report. Note that, in the Hall and Van de Castle, the character lists never includes the dreamer. Hence, if you (willingly or not) enter a report that does not contain another character reference, the model will/should (correctly) produce an empty string. Moreover, it is likely that the produced list of CHAR could be longer than the one produced by the SA model, as not all CHAR might be associated with emotions.

SA Large Multilingual

This model is an XLM-R tuned model, pre-trained with 94 languages available, and tuned on emotion-annotated DreamBank English data. (see original model card to see which are available)

SA Base English-Only

A BERT-base-cased model pre-trained on English-only text and tuned on annotated DreamBank English data.

DReAM

A T5 model tuned to perform text generation and predicts the characters and the (activity) relation between them.