magpie3-causal-implicature

This repository contains several Magpie experiments design to investigate "clausal implicatures", i.e., cases where interpreters infer causal information where none was explicitly conveyed.

There are several kinds of (pilot) experiments here (see experiments folder). Most interesting / interpretable: 1a/b & 4. Pilot 3 is actually interesting but has a "null result" feel to it, as all different formulations were treated the exact same. Pilot 2 attempted iterated narration chains but failed (technical reasons, participants didn't understand what to do).

• Pilot 1: forced choice + justification of that forced-choice decision
  • pilots 1a and 1b differ only in the formulation of the background scenario (pilot 1b contains information that 'ralocrop' is costly to cultivate)
• Pilot 2: forced choice + justification + reproduction (2a only reproduction; 2b/c with actual iteration)
  • pilot 2a: similar to 1b, but with slightly rephrased instructions; also includes additional reproduction task
    • N=33, more reproduction with 'causal language' when participants chose to cultivate both crops
  • pilot 2b & c: similar to 2a, but with actual iterated narration chains; unfortunately, flawed by insufficient understanding of participants regarding what to reproduce
• Pilot 3: background description, choice and justification like in Pilot 1b; N = 161
  • presented three different prompts:
    1. [association] "a high yield of xeliherb is associated with the presence of another plant called ralocrop"
    2. [intervention] "a high yield of xeliherb was observed whenever another plant called ralocrop has been cultivated as well"
    3. [observation] "a high yield of xeliherb was observed whenever another plant called ralocrop was observed as well"
  • adds two more trials:
    • forced-choice select true statement about ralocrop
    • forced-choice select what the science team said
  • choice rates (single crop vs both) for all three prompts are indistinguishable (!?)
• Pilot 4: instructions like 1b, but with free reproduction of /all/ information from the background vignette
  • similar to 2a in structure, but different instructions for reproductions
  • used only the "association" vignette
  • N=50
  • hand-coded reproduction -> shows a abundance of "causal language" especially for "causal responders"

Live version of the experiment here.