Tweets – Page 249 – Trochaisms – Letters make words; sentences make paragraphs

@mmitchell_ai @BayesForDays @haldaume3 in particul…

@mmitchell_ai @BayesForDays @haldaume3 in particular, seems like modified interp KN learns word orderings well. Explainable? Not so much

@mmitchell_ai IIUC ngram win is a victory for general purpose LMs over task specific models. Seem right to you? @BayesForDays @haldaume3

@mmitchell_ai of COURSE you’ve already worked on this. Because Meg. @BayesForDays @haldaume3

@jhamby oh I misunderstood I thought you were saying you were 31% done with a port, but you’re saying “of a build”

Aargh the perplexing grand old contorted brash ancient sturdy linguistic rules that bedevil us non-native speakers. https://t.co/QQHq36QEbQ

@quominus I’m a wrong-half, tyvm. But tell me I’M wrong https://t.co/ZW9F9SXJDP

@BayesForDays @haldaume3 available double-modified NP is small. Wonder how counts of given eg’s sub orderings compare to other sub orderings

@quominus you laugh but your profile pic bears a striking resemblance to my brother (dkg)

@BayesForDays @haldaume3 I see, NNN compounds are sometimes ((NN)N). But doesn’t that speak to the role order too?