We have set up this page as a hub to all interesting prompts, that is, those that are part of a larger analysis or those that show an AI model providing impressive correct or incorrect responses. So far, prompts have been organized into [[#Construction Prompts]] and [[#FrameNet Prompts]]. However, this is not a closed discussion and pages can be reorganized to provide a better experience. In some cases, prompts will contain a "follow-up response" made by authors with the goal of pointing out mistakes made by the Chatty LLM or ask for related information. The answers to these extra prompts are also relevant because they show how well can the models adapt to these situations. These follow-up prompts are presented along with their original ones. To visualize more complete analyses using prompts, you can check [[Sample Analyses]] and [[LLMs on the English Comparative Correlative]]. If you want to submit your own prompts and discussion, go to [[🚃 Join the Convo]]. ## Construction Prompts These are prompts design to test the ability of chatty LLMs to analyze and provide instances of constructions (from any language) informed by the user. - [[Apology construction]] - [[Intransitive verbs in caused-motion clausal patterns]] - [[Knock Knock jokes]] - [[Split argument structure in Brazilian Portuguese]] - [[un-VERB construction]] - [[Wedding Toast construction]] - [[X-safe construction]] - [[XYZ construction]] ## FrameNet Prompts Prompts in this section are slightly different, as they are a product of our [[Prompt Engineering]] method. These are based on templates and can be reused for different pieces of data, specifically frame information. Their most important use is to evaluate in a quantitative manner how well chatty AI models respond to requests that have a varying degree of information quantity and quality. All of that while trying to control as much as possible of other textual characteristics, such as word-choice. - [[Proposing new lexical units]] - [[Proposing new frames of the same kind]] - [[Proposing new frames based on inheritance]]