Month: June 2020

Asymmetries in Latin glide formation

Let us assume, as I have in the past, that the Classical Latin glides [j, w] are allophones of the short high monophthongs /i, u/. Then, any analysis of this allophony must address the following four asymmetries between [j] and [w]:

  1. Intervocalical /i/ is [j.j], as in peior [pej.jor] ‘worse’; intervocalic /u/ is simple.
  2. Intervocalically, /iu/ is realized as [jw], as in laeua [laj.wa] ‘left, leftwards’ (fem. nom.sg.), but /ui/ is realized as [wi], as in pauiō [pa.wi.oː] ‘I beat’.
  3. /u/ preceded by a liquid and followed by a vowel is also realized as [w], as in ceruus [ker.wus] and silua [sil.wa] ‘forest’, but /i/ is never realized as a glide in this position.
  4. There are two cases in which [u] alternates with [w] (the deadjectival suffix /-u-/ is realized as /-w-/ when preceded by a liquid, as in caluus [cal.wus] ‘bald’, and the perfect suffix /-u-/ is realized as /-w-/ in “thematic” stems like cupīuī [ku.piː.wiː] ‘I desired’); there are no alternations between [i] and [j].

What rules gives rise to these asymmetries?

Results of the SIGMORPHON 2020 shared task on multilingual grapheme-to-phoneme conversion

The results of the SIGMORPHON 2020 shared task on multilingual grapheme-to-phoneme conversion are now in, and are summarized in our task paper. A couple bullet points:

  • Unsurprisingly, the best systems all used some form of ensembling.
  • Many of the best teams performed self-training and/or data augmentation experiments, but most of these experiments were performance-negative except in simulated low-resource conditions. Maybe we’ll do a low-resource challenge in a future year.
  • LSTMs and transformers are roughly neck-and-neck; one strong submission used a variant of hard monotonic attention.
  • Many of the best teams used some kind of pre-processing romanization strategy for Korean, the language with the worst baseline accuracy. We speculate why this helps in the task paper.
  • There were some concerns about data quality for three languages (Bulgarian, Georgian, and Lithuanian). We know how to fix them and will do so this summer, if time allows. We may also “re-issue” the challenge data with these fixes.