Márton Sóskuthy

Linguist at @UBCLinguistics. Phonetics, sound change, the lexicon, computers & #rstats.

[Eleanor Chodroff @echodroff]

Workshop announcement and call for abstracts!

Join us for CorpusPhon, a workshop on Corpus Phonetics, co-located with LabPhon19 –– June 26 –– in Seoul and online! 🗣️👂👩‍💻

https://t.co/BIIkwXis2l

Abstracts due Wed. March 6

Co-organized w/ @masonderegger @ctdicanio @msoskuthy

[Márton Sóskuthy @msoskuthy]

tagging some GAM folks here @ucfagls @noamross @martijnwieling

[Márton Sóskuthy @msoskuthy]

Question for stats Twitter: model comparison on fixed effects for models fitted with REML is not kosher. BUT: what if we fit the models with REML, extract the *conventional ML criteria* from the final models, and use those for model comparison? Example in screenshot.

[Márton Sóskuthy @msoskuthy]

... this is kind of theoretical for linear mixed effects models, where ML / REML are not that different in terms of computational time, BUT it could make a huge difference for GAMs where fREML with the discrete=T + multiple threads can be two orders of magnitude faster than ML.

[jen hay @jbhay]

NZILBB Postdoc opening alert!! Te reo Māori contains many long vowel sequences. We'll be studying the acoustics of these sequences, especially in the context of an intervening morpheme boundary. Come and work with us. Application deadline May 28 (1/2) https://t.co/KGSRjlX81t

[Timo Roettger @TimoRoettger]

🚨New preprint🚨 (in press at AMPPS):

"Multidimensional signals and analytic flexibility: Estimating degrees of freedom in human speech analyses."

With the most awesome co-PIs @StefanoCoretta & @jvcasill

https://t.co/YlcojJlfVV

🧵1/

[Márton Sóskuthy @msoskuthy]

@joey_stan @ucfagls thanks for this – a public service indeed! will need to update that tutorial soon...

[Sean Roberts @aucksAdventures]

Come work with me! I'm hiring a research assistant to help with a project on language evolution (full time, 15 months, starts in August/September) https://t.co/oIGY8SyD75

[Márton Sóskuthy @msoskuthy]

@carignan_c ... though, I guess a ZOIB is also a mixture model of a sort! But with fewer parameters & more constrained.

[Márton Sóskuthy @msoskuthy]

@carignan_c Is the data bounded between 0 and 1? Given the peaks at 0/1, I'd first try a zero-one-inflated beta regression. Mixture modelling may be a bit of an overkill, as ZOIB may well be able to capture the distribution you're showing. There's a great tutorial: https://t.co/xgu6DNb6hR

[Márton Sóskuthy @msoskuthy]

@ctdicanio @DingemanseMark @BodoWinter @SciReports @MarcusPerlman So, in the map, your example would actually be shown as no /r/ in rough (even though one of the forms does have an /r/), because it is not the case that the majority of rough words have an /r/.

[Márton Sóskuthy @msoskuthy]

@ctdicanio @DingemanseMark @BodoWinter @SciReports @MarcusPerlman Cool examples! Our statistical models include all forms from a given language – so, in this case, they would have two rough words but only one smooth word. The map simplifies a little: the colour is based on a majority rule.

[UBC Language Sciences Institute @UBCLangScis]

Congratulations @msoskuthy for the newly published article, 'Trilled /r/ is associated with roughness, linking sound and touch across spoken languages,' in @SciReports. Read the full article here! https://t.co/h2m2rQ6YCM

[Márton Sóskuthy @msoskuthy]

I am *so* thrilled (trilled?) that this paper is finally out!!!

[Bodo Winter @BodoWinter]

Our "r is rough" paper is finally out in @SciReports! It's called "Trilled /r/ is associated with roughness, linking sound and touch across spoken languages".

Joint work w/ @msoskuthy @MarcusPerlman & @DingemanseMark

Link: https://t.co/OdaqxW6TpD

Thread ⬇️

[timelyportfolio @timelyportfolio]

this #rstats ggplot2 geom is really nice https://t.co/URYoZEW8uc. I have been waiting for something like for a very long time.

[Márton Sóskuthy @msoskuthy]

@StefanoCoretta @joey_stan Yup, we did this with Sarah Tasker, looking at whether instances of schwa with different spelling (<i> vs. <e> vs. <a>) come from different underlying distributions. Poster attached! We're just returning to this data & analysis, so would love to chat!

[Richard McElreath 🐈‍⬛ @rlmcelreath]

It has been 951 days since Bill Gates gifted every stats teacher with this finely distilled tweet. It's so good, because Gates is not dumb. There is nothing dumb about not understanding conditional probability. It's only human.

[Márton Sóskuthy @msoskuthy]

@ctdicanio Looks familiar – I'm sure I've seen that error message before! Possibly sth to do with how categorical variables are coded, i.e. character / factor / ordered variable. I'm happy to help debug things if you DM me the code / data.