**Project outline for 2009_6**
**Keywords:** EEG; ERP; scalar implicature; entailment
**Overview:** Exp. 1 from Hartshorne, Snedeker, et al., 2015. An ERP investigation of scalar implicature. See paper for details.
------------------------------------
**Publications:**
1. Hartshorne, Joshua K., Jesse Snedeker, Stephanie Yun-Mun Liem Azar, and Albert Kim. (2015). The neural computation of scalar implicature. Language, Cognition, & Neuroscience, 30(5), 620-634. [link](http://www.tandfonline.com/doi/abs/10.1080/23273798.2014.981195#.VxYsFZMrKV4) Experiment 1
2. Hartshorne, Joshua K., Jesse Snedeker & Albert Kim. (2012). The neural computation of scalar implicature. Architectures and Mechanisms in Language Processing (AMLaP), Riva del Garda, Italy.
**Team:**
1. Joshua Hartshorne
2. Al Kim
3. Stephanie Liem Azar
4. Miki Uruwashi
**Data Collection:**
Cambridge
**Data Notes**
Because EEG data files are large and binary, they are not in the repo. If you would like the data, you can find the files [here](http://l3atbcdata.s3-website-us-east-1.amazonaws.com/?prefix=SI_EEGData/). The complete archive is 21 gigs (sorry), and so it has been broken up into individual 1 gig pieces in order to make downloading "easier".
Once you have downloaded the files, you can reassemble them in unix/Mac command line with
cat myfiles_split.tgz_* | tar xz
cat EEGData_split.tar.gz.part-* | tar xz
For windows you can download ported versions of the same commands or use cygwin.
Many thanks to [this poster](http://stackoverflow.com/questions/1120095/split-files-using-tar-gz-zip-or-bzip2/1121070#1121070).
**Method Details**
30 declarative sentences
30 conditional sentences
30 declarative controls (non-rest continuations)
30 conditional controls (non-rest continuations)
35 all controls
38 garden-path controls (19 of each type)
4 additional fillers to begin the experiment
61 sentences were followed by questions
4 lists, crossing order (forwards/backwards, except the first 4 filler trials) and context (declarative/conditional). Note that the fillers were the same in all lists.
2 experiments: experimental and control (control was same except every "some" was changed to "only some")
Presentation was broken up into 8 blocks.
350 stimulus
followed by 250 blank
ITI = 800 + (800,1200)
**Analysis Notes**
Analysis of this experiment was complex, partly because of the novelty of the design, and partly because it was JKH's first ERP experiment and AK had never worked with the EGI system before either. A reasonably complete discussion of the various attempts at analysis is included in AnalysisNotes.pdf. (This document is from the lab notebook JKH kept for this experiment.)
Ultimately, 9 different ways of processing the data were tried. These are largely different attempts to deal with artifact. The EGI system is a high-impedence system, and so appropriately cleaning noise is more important than for other systems. Also, the face electrodes on our caps tended to not stay put, and so those electrodes produced very noisy data. This means that automatic methods of cleaning the data tended not to work. It also affected re-referencing. It took some time to find a method that was robust to this noise.
For these reasons, the results of the study were replicated using AK's EEG lab, for which data-cleaning was much simpler (see 2011_9).
Also see 2011_9 for description of the the cluster analyses and relevant code. That code is not included in this repository.
