Like a notebook or sandbox, ScratchPaper is a place to jot down ideas which do not have their own place on the WikiTtone. These are all works in progress and subject to heavy revision, so should not be cited in any meaningful sense (this is a personal Wiki after all, what are you doing citing something as unstable as this?). Content may be merged into different pages or spun off entirely should they become developed enough.
How are indexical associations with sociolinguistic variables inferred from a distribution of utterances? If a computational solution is found, would we be able to infer indexical fields from datasets?
Kleinschmidt (2018) describes a method of reasoning under uncertainty for the inference of the proper number of phonemes from a distribution of voice onset times (see Voice onset time). But what if the distributions were more rich? For example, each token was a 2-dimensional vector that contains its VOT and duration. The model should be able to infer the correct number of categories, and also the correct parameter estimates for each dimension. That is, Category A would be segments of a particular VOT and duration, and Category B would be segments of meaningfully different VOT or duration; the model should be able to infer that some cues are more informative than others, e.g. VOT is more important for stop identification (in English) than length. What if we included social information? Would it overgenerate and produce too many categories? That is, because of the salient social difference between apical and velar "-ing" would it infer two progressive morphemes? Perhaps that is desirable?
If the model were to be fed a great deal of social and linguistic information, would its inferences about overlapping social meanings point to an indexical field? If a particular agent were to attempt bricolage, how would these stylistic moves affect the wider system? Could the model predict how legible particular stylistic moves would be based on the prior meaning (in the sense of Silverstein 2003)? If say, the model finds t/d deletion to be associated with coolness but not with librarians, would it be able to predict whether a librarian eliding their t's and d's is perceived as cool or not? Would this be a function of which is the stronger cue and would it be analogous to the way in which acoustic cues are weighted?
2 October 2019
How would Kleinschmidt's (2018) category inference work in conjunction with that paper by Andy Wedel (2008? don't remember) where he manages to describe a model that manages to derive an organized phonological system?
2 October 2019
What if, instead of starting from the variable and determining associate meanings, we started from an ideology and observed the linguistic forms people use to enact them? 2 Oct 2019
Rob: "You do a persona, you don't do a character type" Character type is an abstraction, persona is its particular invocation. How might this relate to the generative concept of phoneme and phone?
Refactor Gumball experiment but for accommodating social meaning? /s/ COG in maximally ambiguous phrase manipulated? Top down effects? Tell participants "women are more optimistic" but have statistical distribution of women vs men using probably and definitely be uniform? 12 December 2019
Devyani Sharma, switches and leaks relation to task dynamic phonetics?
Notes for Tulsa photo restoration
- It seems like the panorama was in a book at some point.
- Most of the vertical damage is from folding. It seems that
- The most damaged section has been repaired twice
- The first repair seems to be a single piece of clear tape horizontally across the tear.
- The second repair seems to be two pieces of clear tape vertically across the top and bottom detachments
- The second repair seems to be in response to continued wear.
- The other detachments appear to have been cut by a handheld blade to prevent further damage
- Cuts tend to follow creases, but many inaccuracies exist