Two notes on Proto-Ersuic

This paper looks at the history of Tosu using 'forward reconstruction'. It concludes that Proto-Ersuic changed *-im to *-am already before its breakup as a unity, but the ‘brightening’ of *-a- to -i- took place independently in Tosu and Lizu-Ersu. In Tosu this brightening did not target labial (or velar) initial words lacking an inherited medial *-j-. A number of changes in the history of Tosu probably preceded brightening, namely *-um, *-ak > -o and *-u, *-it, *-at, *-ra > -e. In contrast, the change *-e- > -i- in Tosu, of unclear conditioning, appears to be quite late. A dissimilation *CeCe > CeCa is potentially also a recent change.

Since Yu's initial study, more data on these three languages have become available. A deposit at the Endangered Languages Archive (mpi655457) contains material from all three languages. Zhang (2013) devoted a PhD dissertation to Ersu. Katia Chirkova published new data on Tosu (2014), including contributing to a grammar (Han et al. 2019), and also published studies on the phonology of all three languages (Lizu: Chirkova and Chen 2013, Ersu: Chirkova et al. 2015, Tosu: Chirkova 2015. In terms of reconstruction, Chirkova and Handel propose a series of voiceless nasals (2013), explore the role of inherited high vowels and glides in conditioning spirantization (2013), and compare the Tosu tone categories to those in Bradley's (1979) Proto-Loloish reconstructions (2016).2 Yu (2019) offers a variety of modifications to his previous work on the basis of the newly available Tosu data, incorporating some of the proposals of Chirkova and Handel. Most recently, Chirkova and Handel (2019) argue that the change *-a > -i seen in these three languages (conventionally called 'brightening'), occurred independently in their histories; I confirm this finding below.
My brief remarks here do not systematically revise the reconstructed system of Yu (2012), not least because the author himself is partway through such an enterprise (Yu 2019). Instead, I make use of only that data that Yu himself employs in his two studies and I confine myself primarily to the historical phonology of Tosu. I begin with a look at 'son' and 'daughter' and turn from there to the related questions of Yu's reconstruction of *e and *i. I rely on 'inverted reconstruction' (Hockett 1958, 512-516, Anttila 1972) also called 'reconstructing forward in time' (Watkins 1962, 97) and 'reconstruction from the top down' (Blust 1972, 1), a method known to Trans-Himalayan linguistics (Jacques and Michaud 2011), but still poorly known, as evinced by an anonymous referee's admonition that reconstruction "proceeds from the bottom up". In keeping with this method, I intentionally compare Tosu forms to cognates in distantly related languages.

2
The reconstruction of 'son' and 'daughter' Yu (2012, 96)  factory solution; one should avoid positing competing similar forms in a protolanguage with the same meaning (Fellner and Hill 2019). I regard *zijo² as innovative, analogically renewed on the basis of 'son' and the diminutive suffix seen in *ŋuijo 'calf' and *gojo¹ 'mouse' , among other forms; only *zjeji requires explanation. Yu (2019) revisits these reconstructions in light of the newly available Tosu data. The relevant Tosu forms are ʑi³² 'son' and za⁴⁴-mi⁴⁴ 'daughter' . Although Tosu has the three distinct correspondences -a-, -i-, and -e-for his 2012 *-i-(see Table 1), Yu refrains from adding further reconstructions to account for these divergent correspondences. He etymologizes Tosu ʑi³² to Proto-Ersuic *zi². In contrast, on the basis of new Tosu data Yu "rather formulaically" updates his 2012 reconstructions of the rimes *je and *jẽ to *-an and *-am (see Table 2).4 Yu is not explicit in 2019 about how he proposes to handle the variation in the word for daughter that he posited in 2012. Presumably his new reconstruction *zan² is only meant to replace the first syllable of *zjeji, so we should reconstruct *zanji² as the form ancestral to Zeluo Ersu zi³³ji⁵⁵ and Kala Lizu-C 'ʑeje. By implication, he regads this *-ji as unrelated to the -mi⁴⁴ of Tosu.
If we instead permit ourselves to rewrite Yu's (2019) *-an and *am (his 2012 *-je and *jẽ) mechanically as *-am and *-Am, the solution to 'daughter' falls into place.5 Before 'brightening' we had *za² 'son' and *zami² 'daughter' . The condoculect reflected in Huáng (1992). Note that the raised 1 and 2 in Proto-Ersuic forms index tonal categories and do not indicates phonetic pitch values. 4 The 2012 distinction between *-je-and *-jẽ-, seems to have something to do with distinct outcomes in Mianning Lizu, but I am unable to locate a clear statement of its motivation in Yu's works. 5 A draft version of this article followed Yu's (2002) notational convention by writing *-am and *-ãm, but a referee mistook this as a positive proposal, so I now go with this more explicitly arbitrary solution. The difference between *-am and *-Am is in any event here irrelevant (see note 4). In sum, Proto-Ersuic 'daughter' should be reconstructed *zami² and Proto-Ersuic 'sister' should be reconstructed *hnAmae¹. We may also note in passing that both 'cloud' and 'house' (respectively Tosu tɕa³⁴ < *tɕam¹, Bur. တိ မ် tim, and Tosu ja⁴⁴ < *jAm¹, Bur. အိ မ် im, my reconstructions) point to a change *-im > *-am, unambiguously shared by the three Ersuic languages and therefore an isogloss for this family (see Table 2). Reconstructing *i and *e In our discussion of 'daughter' we saw that Yu's (2012) reconstruction of Proto-Ersuic *-i-corresponds to -a-, -i-, and -e-in Tosu. Yu notes that "[p]eeking at the ptb roots, we notice that many in the -i set have open syllables, whereas a number in the -e set have closed syllables" and considers reconstructing distinct origins in Proto-Ersuic, but decides to "leave this as an exercise for the future" (2019, 32). Instead, he derives the Tosu from the Proto-Ersuic forms, for example proposing the change *ri > e-to account for 'star' and 'hear' . The conditioning environments necessary to explain these data are more complex that the open versus closed syllable that Yu toys with. On the one hand, 'bamboo' with inherited *-i-is a closed syllable (if it is not a Wanderwort); on the other hand 'meat' and 'hear' inherited open syllables but have *e. My suggestion, agreeing with Chirkova and Handel (2019), is that brightening had not yet occurred at the Proto-Ersuic level and that when it did occur it affected Tosu differently than Ersu and Lizu. The words 'frog' and 'thin' , which Yu refers to as "somewhat aberrant" (2019, 32), have impeccable Trans-Himalayan etymologies (see Table 1). Yu's current reconstruction suggests that Burmo-Qiangic *a became Proto-Ersuic *i, which then reverted to Tosu *a, with the most recent change lexically conditioned. To propose that Tosu maintains the inherited vowel is a much simpler explanation. But, if Tosu generally changes inherited *a to -i, then either the two words 'frog' and 'thin' inherited a vowel, call it *a₂, that was distinct from the vowel inherited in 'eat' , 'salt' , etc. (in which case Japhug, Burmese, etc. merged *a₁ and *a₂) or the differing results are due to different phonetic environments. To suggest that in the proto-language 'frog' and 'thin' had distinct vowels from 'eat' , 'salt' , etc. is less parsimonious than to propose that Tosu brightening did not target labial initial words. Those labial initial words with the vowel -i-in Tosu that might appear to be exceptions ('bee' and 'monkey') inherited medial *-j-(Bur. ပျား pyāḥ and ေမျာက် myok), which is what triggered the brightening.
The sound changes affecting Tosu appear to include *at > e for 'kill' and 'eight' , and *ra > e for 'star' and 'hear' (as suggested after a manner by Yu). The -i-in 'bamboo' and the -e-in 'near' and 'name' are likely inherited. As for brightening, it can now be stated more precisely as *-a-> -i-after coronals or medial *-j-. The most problematic forms are 'gold' and 'meat' . In the case of 'gold' , it is quite possible that the Burmese comparandum is simply incorrect. The most obvious analysis of 'meat' is that the inherited rime should be treated as *-jarather than *-a-, but even so, there is a problem because we then expect -i-as we see in 'bee' . The treatment of *-ju-is problematic, with -i-as expected in 'monkey' but -e-for 'swallow'; I have no solution to offer.
Having looked at the Tosu correspondences to Yu's (2012) Proto-Ersuic *-i-, we turn to the Tosu correspondences to his 2012 Proto-Ersuic *-e- (Table 3). Yu sees the -o-vowel of the Tosu forms for 'hand ' , 'deep' , 'breath' , and 'three' as unexplained; the fact that Burmese has the rime -ak for three out of four of these words points to a change *-ak > -o, as Chirkova and Handel (2016, p. 30) propose.7 The change *-um > -o is warranted on the basis of 'three' and 'give' . Some examples of Tosu -i-are probably inherited ('two' , both syllable of 'eye'). The changes *-at > -e, and *-u > -e, already proposed above, account for 'root' , 'who' , 'nine' , 'sky' , 'insect' and the first syllable of 'phlegm/spittle' . Both 'goat' and 'seven' point to a change *-it > -e. If *-i-is indeed inherited, it is surprising to see -e-in 'wind' . The etymon 'fire' appears with -i-as an independent lexical item, but with the vowel -e-in compound with 'smoke' . However, when we note that the first syllable for 'this year' optionally appears with either -e-or -i-, it seems likely that we are not seeing here a phenomenon of any great age. The -a-seen in 'spittle' and 'smoke' suggests a change *-u-> -a-after velars, but if so, 'nine' somehow escaped this change. Perhaps 'spittle' and 'smoke' instead point to a dissimulation of *CeCe to *CeCa.8 If Proto-Ersuic *-a > -i occurred independently in Tosu and Lizu-Ersu, two questions arise, namely (1) how to now reconstruct in Proto-Ersuic those words that Yu (2012) reconstructs with *-i and (2) how to now reconstruct in Proto-Ersuic those words that he reconstructed with *-a. Since all proposals made here relate to Tosu, these questions we can safely kick into the long grass if we understand Yu's (2012) reconstructions as pertaining to Proto-Ersu-Lizu rather than proto-Ersuic.

Conclusions
Proto-Ersuic changed *-im to *-am already before its breakup as a unity, but the 'brightening' of *-a-to -i-took place independently in Tosu and Lizu-Ersu. In Tosu this brightening did not target labial (or velar) initial words lacking an inherited medial *-j-. A number of changes in the history of Tosu probably preceded brightening, namely *-um, *-ak > -o and *-u, *-it, *-at, *-ra > -e. In contrast, the change *-e-> -i-in Tosu, of unclear conditioning, appears to be quite late. A dissimilation *CeCe > CeCa is potentially also a recent change.