# Third-party input notices These notices apply to source-derived material where present in an experiment. Original software and prose licences do not replace upstream terms. The experiment README identifies the required inputs and transformations; its settings and manifest identify the exact snapshots. Model weights and raw response datasets are obtained separately from the original providers. ## WikiText and Wikipedia excerpts Fixed natural-text prefixes and token selections are drawn from WikiText, compiled by Stephen Merity, Caiming Xiong, James Bradbury and Richard Socher, from text by Wikipedia contributors. Source: https://huggingface.co/datasets/Salesforce/wikitext The excerpts retain Creative Commons Attribution-ShareAlike 3.0 terms: https://creativecommons.org/licenses/by-sa/3.0/ The upstream dataset also lists the GNU Free Documentation License. Selections may be truncated, normalised or tokenised as specified by the runner; this does not apply the companion's CC BY 4.0 licence to the original text. ## LAMBADA selections The motion-bounds input includes tokenised selections from the LAMBADA dataset by Denis Paperno and colleagues. Original dataset and CC BY 4.0 terms: https://zenodo.org/records/2630551 https://creativecommons.org/licenses/by/4.0/ The OpenAI-formatted variant used by the runner is provided at https://huggingface.co/datasets/EleutherAI/lambada_openai Its card links the modified MIT notice reproduced below. Both the original source attribution and the variant's notice are retained. ## Sentence-prediction norms Sentence prefixes selected from the resource maintained by Jonathan Peelle: https://github.com/jpeelle/sentence-prediction The resource's MIT notice is reproduced below. The original responses are obtained separately; the fixed selection and order are described in the README. ## CounterFact The fixed-edit selection is derived from CounterFact by Kevin Meng, David Bau, Alex Andonian and Yonatan Belinkov: https://rome.baulab.info/ https://github.com/kmeng01/rome The ROME repository's MIT notice is reproduced below. The original CounterFact file is obtained separately and checked against its expected digest. ## Human prediction resource and UCL sentence material Source: Andrea Gregor de Varda, Marco Marelli and Simona Amenta, https://github.com/Andrea-de-Varda/prediction-resource https://doi.org/10.3758/s13428-023-02261-8 The sentence material originates in the UCL reading corpus (Frank et al., 2013): https://doi.org/10.3758/s13428-012-0313-y The original sources retain their rights. This project grants no additional licence over their sentence material or response data. ## DERCo Fixed narrative prefixes are selected from DERCo by Boi Mai Quach, Cathal Gurrin and Graham Healy. The dataset is distributed under CC BY 4.0: https://osf.io/rkqbu/ https://doi.org/10.1038/s41597-024-03915-8 https://creativecommons.org/licenses/by/4.0/ The selection preserves context identities and is normalised as specified in the experiment. Participant response files are obtained separately. ## Model and tokenizer material Weights are not included. Fixed tokenizer spellings, token indices and model-derived numerical inputs retain any applicable upstream terms; these notices do not assert ownership over vocabulary words or numerical facts. The exact model repositories and revisions are listed with each experiment. ## Upstream notices ### Sentence-prediction MIT License Copyright (c) 2018 Jonathan Peelle Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. ### ROME MIT License Copyright (c) 2022 Kevin Meng Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. ### OpenAI-formatted LAMBADA source notice Modified MIT License Software Copyright (c) 2019 OpenAI We don’t claim ownership of the content you create with GPT-2, so it is yours to do with as you please. We only ask that you use GPT-2 responsibly and clearly indicate your content was created using GPT-2. Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. The above copyright notice and this permission notice need not be included with content created by the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.