Corpus Linguistics Project on Question Syntax in the Map Task Corpus:

I analyzed the Map Task corpus from the University of Edinburgh to explore the relationship between questions and responses. The corpus consisted of 128 text-file transcripts of conversation between pairs of participants using slightly different schematic maps. The participant without a route on their map had to recreate it by talking to their partner with the route. Using Python in the JupyterLab IDE, I applied the spaCy package to tag parts of speech in the dialogue, so that I could identify questions by detecting subject-auxiliary inversion. I then recorded key information such as the question and response, the speaker role (direction-giver or follower), shared lemmas between the question and response, and the presence of affirmative or negative responses. I grouped the data by file labels and calculated metrics like the number of questions, shared lemmas, and yes/no responses. These were mapped to the metadata like number of lines in the conversation, route deviation, familiarity between participants, and whether eye contact was permitted, to create a comprehensive data frame. Finally, I performed correlation analysis to identify relationships between questioning behavior and these variables. I found that a negative correlation (-0.214) between "nos" and route deviation suggests that more frequent "no" responses correspond to better clarity and communication, leading to less deviation. A slight negative correlation (-0.113) between the number of questions and deviation suggests that asking fewer questions may hinder understanding and lead to greater errors. I also identified limitations in the methodology, such as missing questions expressed through tone or non-standard structures (e.g., "you have carved stones?"), and challenges with counting multi-word affirmatives or negatives.