University of Kentucky

I just saw a really interesting paper by Scott Nelson and Jeffrey Heinz (Nelson and Heinz 2025) that proposes a model of phonology and phonetics as complex function application that maintains a discrete phonology while also alowing for things like incomplete neutralization. I myself am always

 sort of

able to follow formal notation, but get a better understanding if I try rewriting it in a programming language of some sort.

Blueprint Model of Production, Pythonically

I had a lot of fun working with the XKCD color survey data, and I think I’ll keep messing around with it here and there in the future. But I also think I’ll want access to the full data. The tidytuesday data set was necessarily boiled down. The answers data frame contained just one hex code associated with one color label, not every color label given to every hex code in the survey.

Getting the xkcd color survey sqlite database

I’ve been playing a

 lot

of Baldur’s Gate 3 lately. My party has gotten to max level, and we’re running around, casting spells and taking names. One of the spells my wizard has is called

 Disintegrate

, which gets summarized like this


Disintegrate Shoot a thin green ray from your finger. If the spell reduces the target to 0 hit points, it disintegrates into a crumbly ash.

Rolling for Damage with the Central Limit Theorem

I’ve been doing a

 lot

of python development recently, and really leaning into object oriented programming for my projects. For example, in the aligned-textgrid package, I started off by defining a class to represent intervals, which at their core have start and end times, and a label.

Experimenting with Object Oriented Programming in R

Today in my stats class, my students saw me realize, in real-time, that you

 can

include random intercepts in poisson models that you couldn’t in ordinary gaussian models,

 and

this might be a nicer way to deal with overdispersion than moving to a negative binomial model.

Random Effects and Overdispersion

When I saw that the TidyTuesday dataset was the the XKCD color survey this week, I had to jump in! source(here::here("_defaults.R")) library(tidyverse) library(tidytuesdayR) library(tinytable) library(mgcv) library(marginaleffects) library(ggdist) library(ggdensity) library(geomtextpath) set.seed(2025-07-08) # eval: false # downloading &amp;

tidytuesday color survey

I’m still thinking about priors, distributions, and logistic regressions. The fact that a fairly broad normal distribution in logit space turns into a bimodal distribution in probability space has got me thinking about the standard deviation of random effects in logistic regression.

Thinking About Hierarchical Variance Parameters

Weinreich, Labov, and Herzog (1968) is a foundational text in my subfield of linguistics. Entitled “Empirical foundations for a theory of language change,” it’s both a comprehensive review of the field at the time, and a programmatic outlook for the future, laying down problems that researchers are still grappling with today.

In Remembrance

I’ve been working a lot with the Discrete Cosine Transform in python, specifically as it’s implemented in scipy. But, I really prefer doing my data munging and stats in R.

 1

What to do! I knew the answer rested in using the reticulate package, which lets you communicate back and forth between python and R, but I hadn’t appreciated

 how

cool reticulate was, which is why I’m making this blog post.

Working with the Discrete Cosine Transform in R

I’m currently teaching a computational linguistics course, and for just the second time, I’m using Github Classroom to manage assignments. I might make a few posts about what I’ve learned in the process, but for now, I think I’ll focus on what’s gone right, and what’s been more challenging in using Github Codespaces. Github Codespaces   A Github Codespace is basically a virtual machine hosted by Github with VS Code as the default interface.

Some lessons in using Github Classroom and Codespaces

I’m not sure when I first came across his stuff, but I’ve really admired Andrew Heiss’ public scholarship, including his blog, which is full of handy stuff. At some point I noticed he had post-level DOIs minted. I knew you could mint repository-level DOIs with GitHub and Zenodo, but maintaining each post as a repository and adding them as submodules to a blog is… a lot.

Setting Up Rogue Scholar

This tidytuesday dataset of colors labels is like the perfect confluence of interests for me! I’ve started learning how to do digital art to illustrate characters for a D&amp;D campaign: Which means I’ve been looking a

 lot

at a color picker that uses Hue, Saturation and Lightness sliders (even though they’re not labelled that way). But I’ve had an interest in colors and color theory for a while.

Visualizing the Color Sphere

Recently, for class notes on probability/the central limit theorem, I wanted to recreate the table of 2d6 values that I made here. A really cool thing I found between that blog post and now is that gt has afmt_icon() operation that will replace text with its fontawesome icon.

Building up a complex {gt} table.

By default in a quarto document, the code and output look something like this: set.seed(2025) rnorm(10) [1] 0.62075674 0.03564140 0.77315448 1.27248909 0.37097543 -0.16285434 [7] 0.39711189 -0.07998932 -0.34496518 0.70215136 Maybe this is just me not wanting my peas to touch my mashed potatos, but I don’t like how

 close

the output is to the text of the document.

Custom Code Chunk css

For me, teaching stats this semester has turned into a journey of discovering what the distributional and ggdist packages can do for me. The way I make illustrative figures will never be the same. So I thought I’d revisit my post about hierarchical variance priors, this time implementing the figures using these two packages.

Random effect priors, redo

I’m teaching a class on quantitative methods in linguistics, and after getting tired, over the years, of debugging everyone’s computer before we could get to the

 content

of learning R (much less quantitative methods), I decided to seize control. I’m running the course through GitHub Classroom which comes with an educational allowance for using GitHub codespaces.

Troubleshooting RStudio for GitHub Classrom and Codespaces

So now I can finally get to visualizing the effect of “light” and other modifiers on colors! When I eventually get to the plotly code, there’s nothing tidy going on, so I’ll be code-folding most of this stuff.

Light vs Dark 

Yesterday I posted about the normalization functions in the tidynorm R package. In order to implement formant track normalization, I had to also put together code for working with the Discrete Cosine Transform (DCT), which in and of itself can be handy to work with.

Doing cool things with the Discrete Cosine Transform in tidynorm

Last semester, I spent time co-developing some python packages:  aligned-textgrid with Christian Brickhouse  fasttrackpy with Santiago Barreda  So, I thought I’d share a little walkthrough of a cool way to use them. They can both be installed with pip. bash pip install fastttrackpy pip install aligned-textgrid FastTrackPy   fasttrackpy (Fruehwald and Barreda 2023) is a python implementation of Santiago Barreda’s Praat plugin (Barreda 2021).

Using FastTrackPy and aligned-textgrid

The upshot The tidynorm package has convenience functions for normalizing Point measurements norm_barkz() norm_deltaF() norm_lobanov() norm_nearey() norm_wattfab() Formant Tracks norm_track_barkz() norm_track_deltaF() norm_track_lobanov() norm_track_nearey() norm_track_wattfab() DCT coefficients norm_dct_barkz() norm_dct_deltaF() norm_dct_lobanov() norm_dct_nearey() norm_dct_wattfab() As well as generic functions to implement your

Introducing tidynorm

This summer, I’ve been spending some time getting more familiar with the nitty-gritty of Bayesian models, and I’m working on some modelling in Stan for logistic regression. Setting up the basic model formula in Stan for logistic regression is pretty straightforward: mu = intercept + slope * x; y ~ bernoulli_logit(mu);  And then you “just” need to set a prior for for the intercept and slope terms.

Getting a sense of priors for logistic regression

This might be a “everyone else already knew about this” thing, but I’ve finally gotten to a place of understanding about setting default colors scales for ggplot2, so I thought I’d share.

Væl Space

Thinking About Hierarchical Variance Parameters

Getting a sense of priors for logistic regression