PhDniX's blog

Segolate Plurals and North-West Semitic

Yesterday, I wrote a little thread on the productive segolate plurals in Arabic, and how this challenges some of the notions of this as a North-West-Semitic isogloss. I received some pushback, several people felt I misrepresented Huehnergard's position.

I still don't feel I quite did that. Huehnergard clearly not only sees the spread of the *malk pl. *malakīm as against broken plural options as the innovation (which is defensible, and more clearly formulated here), but also as the regularization of a-insertion in sound plurals (the only way to make sense of the cited examples of laʕnah pl. laʕanāt versus ʔahl pl. ʔahlūn).

The latter point I really think we ought to throw out completely. ʔahl pl. ʔahlūn indeed lacks the expected a-insertion, but also in Arabic this is the exception and not the rule. I honestly only know this word to have this plural (and ʔahālī- is the much more typical plural, but ʔahl-ū- is the Quranic form!). If the innovation is that this was generalized, we are talking about an innovation in a single noun *ʔahl- where this happened. That's not significant.

But I think the whole discussion, by focusing on these segolate plurals is in fact a red herring. Arabic's plural system cannot be simply compared to the North-West Semitic plural system, and by assuming that they get equated important details are lost. I think if we take a more subtle approach, we can actually come to see a much more pervasive innovation in North-West Semitic, but it has nothing to do with segolate plurals.

The plural system of Arabic, maximally has three semantically distinct "plurals". This is best represented with the noun baqarah 'cow'. This noun has 1. a paucal plural 2. a collect and 3. a normal plural: sg. baqar-at pauc. baqar-āt coll. baqar pl. ʔabqār.

Not all nouns have collectives, and especially post-Quranic Classical Arabic the paucal versus plural distinction is not always so well-kept. But it is prescribed by the Arab grammarians, and the Quran and pre-Islamic poetry quite clearly has the distinction (paucal being 3-9 things; plural being more than that). Clear traces of this system actually continue in Modern Arabic dialects.

North-West Semitic has a much simpler system, essentially just singular and plural. There are traces of collectives, but they have basically been integrated into the plural system. So now the question becomes: is the Arabic paucal/plural distinction an expansion of the system we see in North-West-Semitic, or is NWS a simplification of the Arabic system? My inclination is to think of the Arabic as original, but people do not talk about this enough, if you want to argue for some special significance of the segolate plurals, it's nothing to do with those plurals specifically, it has to do with the loss of the true plurals in favour of paucal plurals!

For feminine segolate nouns in Arabic, the formation of paucal plurals besides true plurals is regular. So laʕnah/niʕmah/ẓulmah and ANY other CVCC-at- noun always and regularly have the segolate plural, i.e. laʕanāt, niʕa/imāt, ẓula/umāt. However, these plurals are always paucal. Besides this such nouns also have true plurals, which are typically broken plurals liʕān, niʕam, ẓulam. NWS lacks broken plurals of this type completely.

However, these kinds of broken plurals have clear parallels in other Semitic languages (notably Ethio-Semitic) as well as in Afro-Asiatic languages, which certainly suggests that they are archaic (this is definitely convincing for liʕān type plurals, but one actually has to wonder whether niʕam and ẓulam aren't just clipped versions of the paucal plural with the -āt removed).

So for CVCC-at- nouns in NWS, if there is an innovation at all, it is the spread of the paucal plural to come to function as the only plural, and thus the loss of the broken plurals. There is nothing special about the CVCC-at- plurals here in Arabic (or NWS) for that matter. This is the generalizable rule for feminine nouns in general: What becomes the regular plural in NWS is simply the regular paucal of Arabic, whereas what is the true plural of Arabic disappears without a trace in NWS.

This basically resolves the feminine side of the picture. The masculine (or maybe more precise: the nouns that lack the -at- ending) is more complex. Where CVCC-at- pauc. CVCaC-āt- is regular, and perfectly parallel between Arabic and NWS (with the caveat that in Arabic it is paucal and in NWS it is plural), it is more difficult with CVCC nouns. Here, finding compelling parallels to the NWS *malk pl.*malakīm system is significantly harder. There are a couple of standard examples that get cited, e.g. ʔarḍ pauc. ʔarad-ūn 'earth', ʔahl pauc. ʔahal-āt "family". But this kind of plural is legitimately rare in Arabic. So are we wrong to say CVCaC-āt- plurals are NWS, but right that CVCaC-ū/ī- are a NWS innovation? Maybe, but this really needs to be qualified.

Masculine nouns, by definition cannot have collectives, but otherwise have the same system as the feminine, e.g. sg. kalb 'dog' pauc. ʔaklub (not **kalab-ūn) pl, kilāb. The notable difference here is therefore that the masculine nouns use a broken plural pattern (rather than a suffixed pattern) to makes the paucal (feminine nouns can actually do this too niʕmah pauc niʕa/imāt, ʔanʕum). Now, you might conclude that since broken plurals are archaic, clearly NWS is the innovator and Arabic the archaic one, but here I would urge to point out that not all broken plurals are created equal. While the true plural pattern kilāb has excellent Afro-Asiatic comparanda, and must certainly be old, ʔaklub is in fact extremely isolated, so isolated that it only occurs in Arabic (the Gəʕəz hägär pl. ʔähgur looks superficially similar, but would be equivalent to *ʔaCCūC). Should we really conclude here that Arabic is the only one to retain this system of paucal broken plurals? OR alternatively, should we take seriously the possibility that Arabic is in fact innovative here. I think the latter. While there are a number of broken plural patterns that convincingly be reconstructed for Proto-Semitic and even earlier, I don't think any of the ʔa- prefixed ones (which are strictly all paucal plurals according to the Arab grammarians) are necessarily that, and seem to rather be a typical areal innovation shared between Arabic, Ethio-Semitic, Ancient South Arabian and perhaps Modern South Arabian. If the ʔa- plurals are an innovative paucalisation system we can Imagine a Proto-Semitic system as follows:

	Masculine	Feminine
Singular	*kalb-	*kalb-at-
Paucal	*kalab-ū	*kalab-āt-
Plural	*kilāb-	*kilāb-

Not the gender neutralization of *kalb-/*kalb-at- in the plural. The innovation of NWS would then be the generalization of the paucal plural for all plurals (which has as an added bonus that it annuls the gender neutralization).

	Masculine	Feminine
Singular	*kalb-	*kalb-at-
Paucal	*kalab-ū	*kalab-āt-
Plural	*kilāb- > *kalab-ū	*kilāb- > *kalab-āt-

The Arabic (and perhaps a more general "South Semitic" areal development) would be the introduction of a new paucal plural pattern, which completely ousted the original paucal in the masculine, and occasionally competes with the original feminine paucal.

	Masculine	Feminine
Singular	*kalb-	*kalb-at-
Paucal	*kalab-ū > ʔaklub-	*kalab-āt- (*ʔaklub-)
Plural	*kilāb-	*kilāb-

This is just one way of explaining it, but I think it should be clear that "the *malk pl. *malakīm plural was generalized in NWS" clearly does not capture what has actually happened and reconfigured, and it gives the mistaken impression that kalbah pl. kalabāt style plurals are not actually the regular paucal pluralisation strategy of Arabic.

One may want to argue that the ʔa- prefixed broken plural patterns are in fact archaic, and they form the old productive paucalisation strategy in Semitic. If this is true, I think that can only be reasonably be argued for the masculine paucal. I see no way to account for the regular use of the -āt- feminine paucals in Arabic if this is not an archaism. This would make for an unusually asymmetric system, which is not necessarily an argument against it, but not exactly in favour of it either.

Laysa r-rasmu bi-qirāʾah!

Me and my dear friend and colleague Hythem Sidky keep on telling each other that every time we're at a conference, we should get pins that read ليس الرسم بقراءة "The consonantal skeleton is not a reading!", because this motto highlights an extremely common misunderstanding within the field. I'm attempting the description of this mistake a bit, so it can be written down somewhere and cited from that point onward. This is a first draft at me putting these thoughts together into writing.

Readings follow the consonantal skeleton.

One of the main requirements for a reading to be considered valid is that it needs to follow the standard consonantal skeleton, as standardized by ʿUṯmān. As a result, all ten canonical reading traditions today are in agreement with the Uthmanic Text.

But, what is commonly misunderstood is that this is not some kind of feature unique to the seven or the ten canonical readers. The Islamic tradition records many other readers that have perfectly Uthmanic readings, that is, readings that agree with the Uthmanic text, for example: al-Yazīdī, Sallām ʾAbū al-Munḏir, ʾAbū Ḥātim al-Siǧistānī, ʾAbū ʿUbayd al-Qāsim b. Sallām, just to name a few. Moreover, if we turn to vocalised manuscripts from the early Islamic period, they are all perfectly Uthmanic, and so necessarily the reading that is marked with these vowels have to agree with the Uthmanic text. But vocalised manuscripts very frequently reflect readings that fall well outside anything covered by the Islamic tradition, but they are nevertheless perfectly Uthmanic. There are, therefore, countless readings that agree with the rasm.

Sometimes the rasm can contain hints of the reading that the scribe had in mind. Perhaps they placed some consonantal dots in place where there is actually disagreement among the readers on the consonantal dot. But usually there is so much ambiguity in early, barely dotted, manuscripts that even if every single case you check fits one of the readings you know, this still does not necessarily mean that the text is written according to that reading. All this means is that the text is consistent with that reading. Especially with readers from similar regions, it is actually very likely that multiple readers would be consistent with a certain manuscript. This is a good indication that the person who wrote the text was also from that region, and was familiar with readings from that region, but it does not mean we can with certainty identify the reader intended by the scribe.

The Regional rasm dictates the regional reading, not the other way around

The regional codices of Medina, Basra, Kufa and Syria are not perfectly identical to one another in terms of consonantal skeleton. Here and there a regional codex has a letter less or more, or a slightly different spelling. In one notable case there is a word absent in the Medinan and Syrian codices, which is present in the Kufan and Basran codices, namely in Q57:15 فان الله [هو] الغنى الحميد. As a result the Syrian and Medinan readers read fa-ʾinna ḷḷāha l-ġaniyyu l-ḥamīd, whereas the Kufan and Basran readers read fa-ʾinna ḷḷāha huwa l-ġaniyyu l-ḥamīd. Save a couple of small disagreements among the canonical readers, all readers in a certain region in principle follow their regional rasm. In other words, the rasm dictates what they read, and not the other way around. This is clear from the fact that these regional rasm variants make a stemma, even if you use just the manuscripts we have.

All too often, however, in discussions of early manuscripts such points are presented as the rasm being adjusted to accomodate the reader. This clearly is not what is happening, as if it were simply up to the whims of the reader on how to read, and the scribe slavishly followed that, rather than some existing regional norm of the written text, we would not at all expect a stemma too appear.

All Readers postdate the consonantal skeleton

The previous two points already point out that there are good reasons to think of the rasm as primary, and the readers simply following it. This is not to say that scribes would never at any point be persuaded to let their reading of a verse influence how they would write it down. They probably would, but that is clearly not the general direction of things. Another reason why the rasm is clearly primary is for the rather basic fact that the rasm is much much earlier than most readers for whom we have a full description, and certainly those that belong to the canon. When the rasm was established, during the reign of Uthman (23-35 AH), none of the canonical readers had yet been born (except perhaps Ibn ʿĀmir, if you believe he really did live until the ripe age of 110!). Even when most of our earliest manuscripts were written, most of the readers were either still infants or had not even been born.

Just for reference these are the death years in order of earlierst to latest of the 10 canonical readers, their main transmitters from whom we actually receive the readings are later: ibn ʿĀmir (d. 118/736); ibn Kaṯīr (d. 120/738); ʿĀṣim (d. 127/745); ʾAbū Ǧaʿfar (d. 130/747); ʾAbū ʿAmr (d. 154/770); Ḥamzah (d. 156/773); Nāfiʿ (d. 169/785); al-Kisāʾī (d. 189/804); Yaʿqūb (d. 205/820); Ḫalaf (d. 229/844).

Therefore saying that any manuscript that predates these readers, which many of them do, is written "in" their reading is simply anachronistic. (On the other hand, if you do find a vocalised manuscript that follows a canonical reader or transmitter perfectly, and the section of the manuscript displays enough distinctive features, then you can certainly date such a manuscript to their lifetime at the earliest, but most likely after their lifetime).

A manuscript is not written according to a "Mix of Readings"

As most of the readers we have access to postdate our earliest manuscripts, it should not surprise us that when we try to extract some reading information from the sparse consonantal dotting in such a manuscript, this will often yield a list of specific variants that do not correspond to any known reader. I frequently see this inverted, discussions well say that a manuscript X shows the readings of reader A, B and C. Which is then construed as a scribe grabbing readings from all over the place and putting them in their manuscript. But this is obviously not what is happening. The scribe of manuscript X simply chose to represent a single reading -- presumably the reading he would recite himself -- and the variants that they have, happen to sometimes correspond to readings that reader A adopted, and other times what reader B adopted and other times what reader C adopted. He is not "mixing" their readings (For earlier manuscript since they hadn't been born, that doesn't make sense. But even for later manuscripts this is not the obvious explanation), he is simply reflecting a reader.

You could do the same time by looking at canonical readers. If you would look at the specific choices Ḫalaf makes, sometimes he has the reading of Ḥamzah and other times the reading of al-Kisāʾī and rarely a reading that neither of them have. Ḫalaf really did combine the readers from both of these Kufan readers, but the result is his OWN choice of variants. He is not trying to represent a "mix" of readings, but what he considered a (personal) right way to recite the Quran. This is how one should think of the "cocktail of readings" in any manuscript as well. By looking closely at what regions a manuscript most often aligns with you might be able to get a sense in what milieu it was written (e.g. if it has a lot of readings that are typical for Basra, it's more likely to be a Basran manuscript), but it doesn't mean that the intention is to reflect multiple readings at once. It is reflecting a single reading.

Readings have nothing to say about the consonantal skeleton

A final point that is worth making within the context the rasm not being a reading, is also that a reading is not the rasm. Because the modern print editions such as the Medina Quran are printed according to the reading of Ḥafṣ from ʿĀṣim, many scholars and lay people alike assume that all that is in such a print edition is part of the reading of Ḥafṣ, even specific orthographic idiosyncrasies. But this is wrong. The information of a reading is purely the phonetic layer of the Quran: how exactly the Quran is pronounced according to this reader. But it contains no information about how words are spelled. So, for example, the beginning of Q18:23 is read by Ḥafṣ as wa-lā taqūlanna li-šayʾin "And never say of anything". This is the information carried by his reading. But in the modern print editions it is spelled ولا تقولن لشاى with the irregular and unusually spelling of šayʾ with an extra ʾalif, rather than the typical spelling without found all throughout the rest of the Quran (e.g. لشى in Q16:40). This is a question of rasm and it is stypulated in rasm works, and has absolutely nothing to do with reading, and certainly is not something communicated in the data of Ḥafṣ' reading.

Tawātur al-Qirāʾāt according to ibn al-Jazarī

One of the interesting parts of modern Islamic orthodoxy is the belief that the ten canonical reading traditions are transmitted by tawātur, mass transmission so massive that it is unthinkable that people could have agreed upon a lie. The requirement of tawātur doesn't really enter the Islamic discourse until rather late, and while it has become the dominant position, as late as the 15th century people were still giving pushback against it. One of the people who pushed back against this is Šams al-Dīn Ibn al-Jazarī (d. 833 AH/1429 CE), the very same person who wrote a definitive description of the three reading traditions after the seven which are now considered canonical. He is seen as the canonizer of the ten. Ironically the ten are now widely held to be transmitted by tawātur even though he himself emphatically (and cogently) denied this, not just for his three after the ten but even for the seven!

Since this is not widely known, or outright denied, I thought it'd be useful to write up a translation. I've based this off the text of the beautiful edition by Ayman Suwayd.

Before we get into the meat of the argument, it is useful to first know start with the "Pillars of a Valid Recitation"  (pg. 117):

The Pillars of Valid Recitation

Each Reading:

1. Follows the (grammatical rules of) Arabic, even if just in some aspect.
2. And follows one of the Uthmanic codices, even if it only permits it.
3. Is sound in its chain of transmission.

This is valid recitation which is not allowed to refute, and not valid to be denied. Indee,d it is from among the seven ʾaḥruf which the Quran was revealed by, and it is acceptable for the people to follow it, whether it is from the seven ʾImāms, or the ten or some other of them from among the accepted ʾImāms.

And when one of these three pillars is lacking, it is dispatched of as weak (ḍaʿīf), unusual (šāḏḏ) or invalid (bāṭil), regardless of whether it is from the seven or from among who is the greatest of them.

A lot is happening in such a short section. What is notable is that these three requirements are interdependent. No matter how good your sound chain of transmission is, even if it's back to one of the seven canonical readings, if said reading doesn't follow proper Arabic grammar or doesn't follow the Uthmanic text, it is still rejected!

Ibn al-Jazarī is channeling a view that was clearly around with earlier authorities, including the canonizer of the seven, Ibn Mujāhid (d. 324 AH/936 CE) who would frequently dismiss canonical readings that he described themselves and had a perfectly valid chain of transmission for, because he considered the grammar to be unacceptable, for example.

This view, by Ibn al-Jazarī's time had clearly come to be under significant pressure. Authorities such as ʾAbū Ḥayyān al-Ġarnāṭī (d. 745/1344 CE) would frequently challenge such dismissals because after all: the reader was a true Arab, so how could he not know what proper Arabic was, and the Isnad was fine. Who are you to say that the language is bad, you can't reject it! With such a line of argumentation, you make two of the three pillars (1 and 2) essentially powerless, and chain of transmission becomes tantamount. It is therefore no surprise that after having weakened two of the three pillars, ʾAbū Ḥayyān invested heavily on the doctrine of tawātur. These readings were mass-transmitted. That was the only way you could plausibly neuter pillar 1 and 2, by making pillar 3 unassailable. Ibn al-Jazarī is implicitly denying these by explicitly saying that things are to be rejected even if they do come from the seven.

What is also worth noting is that Ibn al-Jazarī clearly holds open the door for the possibility that not just the seven, or his ten would fit these requirements. Clearly there are other readings with perfectly valid chains of transmission, which follow one of the Uthmanic codices, and have proper Arabic grammar. And this indeed is clearly the case. Al-Yazīdī's (one of the 4 after then 10) clearly is valid in all those ways, and several; other complete transmissions not covered by the seven or the ten have come down to us in the literary tradition. People often think that anything beyond the ten is šaḏḏ "unusual/non-canonical". But this is not the case as far as Ibn al-Jazarī is concerned.

He then continues to explain precisely what he means with "to be in agreement with Arabic grammar" and "to follow one of the Uthmanic codices". Discussions that are interesting in their own right, and maybe I'll sit down to translate those as well one day but I will skip over it for now. Eventually he comes to the section where he explains his third requirement "Soundness in chain of transmission" (pg. 126).

As for our words: "Is sound in transmission" we mean by it that and honest and trustworthy person transmitted this reading tradition from someone who was likewise, continuing like that until you arrive (at the authority), and in addition it is well-known among the ʾImāms of this matter, the ones who  have mastered it, without being considered by them to have fault, or anything that is considered unusual by some of them.

Some modern authors  (al-mutaʾaḫḫirīn) required tawātur for this pillar and it is not enough for it to just have a sound chain of transmission. They claim that the Quran can only be established by tawātur and that which comes in a single strand transmission (ʾāḥād) does not establish the Quran.

It is obvious what this implies: If tawātur would establish soundness, then we would not need the other two requirements, that is, adherence to the Uthmanic codices and the other one (adherence to proper grammar). When the words of disagreement were established to come from the prophet by tawātur, we would be required to follow it, as it would be certain that it would be Qurʾān regardless of whether it followed the Uthmanic text or differed from it.

And if we would require tawātur of each word among the words of disagreement that are established from these seven ʾImāms (i.e. the seven canonical readers) and others besides them would be rejected.

Previously, I used to incline towards this opinion, but then its wrongness became clear to me. Following the ʾImāms of the first generations (salaf) and those that followed (ḫalaf) is the better option.

He then continues to cite examples of other authorities who have likewise not required tawātur. Here again we are confronted with an extraordinary example of lucid reasoning. The laconic dismissal of the requirement of tawātur leads to what Ibn al-Jazarī considers a logical absurdity: If it is really true that transmission needs to be so massive that nobody could disagree on its truth, then the other two pillars become absurd. 1. The Uthmanic text postdates the prophet, if there were a mass transmission of a reading that didn't follow the Uthmanic codex, you could obviously not dismiss such a (purely hypothetical!) reading just because it doesn't follow a written text that did not even exist at the time of the prophet. That's the whole point of tawātur.

Even more implicit (Ibn al-Jazarī was perhaps embarrassed to have to point out just how heretical the claim was): if there really is tawātur, the requirement of sound grammar becomes nonsensical. If there was tawātur, as a Muslim, one would have to accept it as certainly from the prophet -- and by extension that it is the literal word of God. And then this requirement is supposed to mean that this reading be subjected to the question whether or not it agrees with the grammar of Arabic? Are Muslims to be the school teachers of God checking His Arabic?

But after the hypothetical example of a tawātur reading that doesn't follow the Uthmanic text (something Ibn al-Jazarī clearly does not actually believe exists), he then turns to the concrete: For the vast majority of the words of the Quran everyone is simply in agreement how to recite it. Not just among the ten canonical readers, but even beyond it. Clearly such words, which make up the majority of the text, do meet the requirement of tawātur. But it is the words of disagreement, i.e. words where not all readers agree on how to read it where stipulating tawātur becomes difficult.

This becomes especially difficult in the case of unique readings. Among the twenty canonical transmitters of the ten canonical readers only the now dominant Ḥafṣ (d. 180 AH/796 CE) reads kufuwan in Q112:4, all others read kufuʾan (and Ḥamzah kufʾan). This doesn't change massively if we start including chains of transmission outside of the twenty canonical transmitters. In other words: this reading is isolated, and the transmission of Ḥafṣ was not such a prolific teacher, and his chain of transmission does not reach back to a huge number of companions of the prophet. Logically, Tawātur cannot be claimed for such a point of disagreement. Ibn al-Jazarī understood this.

Even in practical terms, this is clear: the great exegete al-Ṭabarī (d. 310 AH/923 CE) does not seem to show any awareness of readings that are unique to Ḥafṣ, while he is well aware of different readings, and cites them extensively in his exegesis. Is that what a reading that is so massively transmitted that nobody could not know about it looks like? No, of course not.

But of course, one may be forgiven for being swept up in the zeal of declaring absolute certainty about the recitation of the Quran, after all Ibn al-Jazarī, obviously a razor sharp thinker himself used to think that tawātur was necessary and possible for the readings as well, and it took time for the wrongness to become clear to him. Luckily for us, he shared with us his excellent logic.

Medio-Passives of the Primae Waw verbs

A final conundrum among a whole slew of conundrums among these among the primae waw verbs is the behaviour of the mediopassive derivation.

In general, I don't think we are quite ready to reconstruct the Proto-Berber at all yet, a bunch of Berber languages show vastly different formations, which do not seem obviously reconcilable, but even sticking to the Tuareg forms, the mediopassive of the primae waw verbs behaves really weird.

First the normal verb, here we see that the mediopassive is created through an m prefix and added the |Ə-A-->| vowel pattern to the aorist stem. Thus we get: ikrəs (< *yăkres) 'to tie', and yămmăkrăs (< *yəmmăkrăs) 'to be tied together'. So if the primae waw verbs like *ăkkəs 'to remove' should behave like 'normal' triradical verbs, one would expect either something like *yəmăkkăs or if in this stem a *u should appear (maybe from *ăw?) one might predict something liek yəmukăs.  Since the presence of *u seems to triger a all-Ə vocalism in the stem, one would probably expect *yəmukəs.

Instead, we basically get a combination of the two. Indeed an u appears, and this turns the aorist voalism into all-Ə, but... the geminate doesn't disappear giving us Tuareg yəmmukkəs 'to be removed'. I don't know what to make of that.

*vcc verbs act as expected. yakər 'to steal' has a mediopassive yămmakăr (< *yəmmakăr), keeping the low-vowel aorist, and simply using the long vowel as a regular long vowel slot.

So with the *wcc roots, you end up with both u and gemination being in a kind of quantum superposition, one time one comes out, and the other time the other, and we don't fully understand the conditioning, but then the mediopassive comes along and both come out!

More Proto-Berber Primae Waw

After my previous blogpost, Maarten Kossmann pushed back in the comments on Twitter on the idea that *əw really yields *u, which is something that would be needed to derive *əssuɣəd from *əssəwɣəd. There are, namely several cases where u and əw really seem to remain distinct, especially in the *CəCC-an plural pattern. Thus *a-ḍăwwal pl. *i-ḍəwl-an 'brother-in-law' where the later does not surface as **iḍulan in several dialects but as iḍəwlan, for example in Figuig. Now that could perhaps be explained as the result of analogy, but Prof. Kossmann pointed out two other words where that absolutely cannot be the case, as there (surprisingly!) is no *w in the singular:

• Riffian ḏ̣aḏ̣ pl. i-ḏ̣əwḏ̣an 'finger' (though surprisingly not Figuig ḍaḍ pl. i-ḍuḍan)
• Riffian aẓwar pl. iẓəwran 'root' besides e.g. Kabyle a-ẓar pl. i-ẓuran which suggests a regularized but originally irregular alternation between singuler a-ẓar pl. i-ẓəwran.

Not only do we see here that *əw apparently does not yield *u in all dialects (but causative əssuɣəd always has u), but we also see yet another example of an a corresponding to the presence of a w, as in the instrument nouns like a-satəl from the root ăttəl.

With noun formations, we have a certain amount of formations that we can explore, which would generate some different possible formations. Nouns typically have the shape CăCVC or CCVC (from syncopated *CəCVC?) where V can be any vowel. The middle radical can also be geminated, but we can pretty clearly count that out as an option since we know the outcome of that from *a-ḍăwwal, there *w is just retained.

So could it be that the a comes from the loss of an intervocalic *w or, at least a pre-vocalic *w? Assuming that *t-addar-t pl. *t-udr-en 'house' are indeed from the root *wdr 'to live', we'd reconstruct these as *ta-wăddar-t and *ti-wədr-en. Depending on the relative chronology of the univerbation of the noun prefix to the noun (it was surely at one point a definite article or something like that) that either suggests the sound laws: *awă > a and *iwə > u or *wă > a and *wə > u.

Words like *urəɣ 'gold' (cf. iwriɣ 'to be yellow') make it clear that *a-wrəɣ or *a-wərəɣ yields *urəɣ, so that for the vocalisation of *u from *w we can probably posit *Vwə > *u and possibly *aw > u. So What could 'finger' and 'root' be similarly from something like *a-ḍăwad or *a-ḍăwăd (or perhaps even *a-ḍ(ə)wăd/*a-ḍ(ə)wad)?

If we assume this, this has implications for our causatives of Primae waw verbs, and revives an idea I have had kicking around in my head before, one would have to assume that *əssuɣəd goes back to *əssəwəɣəd. This runs into some trouble with especially the perfect in light of how I reconstruct heavy verbs with four vowels, but this is something that can be resolved with some shuffling of relative chronology.

One reason why it's attractive to posit *əssəwəɣəd with a vowel between the w and the rest of the root, is because it would resolve a bit of irregular behaviour that is widespread in Berber dialects. The verb *ănkər 'to get up' in many dialects (irregularly) assimilate the *n to k, thus yielding kkər, but in a number of dialects this *n reappears again in the causative, e.g. Kabyle kkər caus. ssənkər.  That is beautifully explained if there was a *ə in between the first and the second radical of the root, blocking the assimilation in the causative that took place in the underived root.

To make this work, the instrumental formation (*a-satəl) would also have to historically have had an intervocalic *w that become a *a. Not unsurprising, I would suspect instrumental formation is ultimately a kind of nominalization from the causative 'a thing which you make/cause to wrap').

However, this could not be reconstructed as *a-s-ăwătəl as that would predict that the normal instrumental formation would be **a-săɣănəs, but comparative evidence makes it pretty clear it's *a-săɣnəs. I think we can sneak in a ə between the second and third root consonant and posit a syncope of some kind (as we would need to do in the causative *əss-əlməd < əssələməd), but there is very little reason to assume that *ă ever syncopated. So... *a-s-ăwətəl? That would suggest that *ăwə yields yields *a. This would mean 'finger' could also be from *a-ḍăwəd. This solution is pretty ad hoc and only really solves this specific problem, which isn't great.

It does, however, resolve our issue why the agent noun doesn't show the same behaviour. We saw that for the agent noun we expect *m-|ă...ă/a|, but with instrumentals of Primae waw stems it shows up with an initial *u! So *a-n-uḍăf could come from *a-n-ăwḍăf, with yet another shift: *ăw > u. Again this is fairly ad hoc, but it doesn't run into big trouble I think.

So to sum up (assuming vocalisation of *w happened before the noun prefix became part of the root):

• *(ə)wə > *u (əssuɣəd, urəɣ)
• *ăwə > *a (a-ḍad?, a-satəl)
• *əwC > *əw
• *ăwC > *u

There is a final issue that I don't think is fully resolved yet, these Primae waw verbs are also an important source of words with word-initial *e that do not come from an earlier *a. Thus forms like *eḍəs 'sleep' (verb *ăṭṭəs) and eləs 'fleece' (verb *ălləs), *eɣəd 'embers' (verb ăqqəd 'to burn') but also a fairly widespread pattern *ec̄ac of verbal nouns which seems reconstructible, e.g. in Tuareg eqqad 'cauterizing' from əqqəd 'to cauterize', cf. Fig. iqqad from qqəd 'to burn'.

Prasse derives these from *a-Hĭɣĭd and *a-Hĭqqad. *a-Hĭ quite nicely yields *e, in a pretty similar way as it does in word-final position in verbs, əmda + 1sg -əɣ yields əmdeɣ which in Prasse's reconstruction would be something like *əmdăH-ĭɣ. We could perhaps rewrite this rule with *w so that *awĭ yields *e and *awŭ yields *u, but that would require us to assume that these contractions only happened once the prefix had become part of nominal stem and we have the assume that Pre-Proto-Berber had a distinction between *ĭ and *ŭ  (which is not unreasonable, but there's not that much evidence for it).

The reason why this is necessary is because stems with a long vowel as stem initial vowel (verbal stems of the type *agəm) do not yield e once they come in contact with the prefix but just i. e.g. Tuareg abər 'to take a handful' VN ibbur not **ebbur. Prasse explains the *e in eqqad as lowering of *i to e under influence of a, That words for those forms, but it doesn't explain the *e vowel in *eḍəs, *eləs, *eɣəd.

It is fine of course to assume that the contraction only happened once the prefix had univerbated with the rest of the noun stem, but it yields a rather ugly reconstruction where *ăwə yields *a whereas *awə yields u or e (presumably depending on whether it is *awŭ or *awĭ), which is... phonetically not so plausible, even if it doesn't  technically get me in trouble. Historical linguistic explanations should not only work (which this one does) but also be plausible (not so sure if this one is)... Rewriting the rules with prefix univerbation in mind:

• *awŭ > *u
• *awĭ > *e
• *ĭwĭ > *u (*ĭ is probably requires here to explain why it is *əssigəm from *agəm but *əssuɣəd from *ăqqəd)
• *ăwə > *a
• *əwC > *əw
• *ăwC > *u

So, err, who knows. I don't know if it's any good yet. I've solved all the problems, but it requires a lot of ad hoc sound changes which are moreover not phonetically particularly plausible... I wonder if there is a possibility to think of a suprasegmental labialization which is used to explain difficult things in Chadic... but not entirely sure where to start with that.

Proto-Berber Primae Waw: Is it really a /w/?

Last time I talked a bit about an interesting possible connection of how Berber and Semitic treat verbs with initial w as the first root consonant, both of them notably seem to lack the w in the underived prefix conjugation, but reappears on other derivations. Thus yaqid but causative yūqid in Semitic, and *yăqqəd but causative *yəssuɣəd in Berber.

But as I was looking into the morphology of the Berber Primae-Waw, I noticed (and Prasse noticed with me) something strange: It is true that something which becomes a *u appears in the Verb and in verbal derivations of the group of verbs like *ăqqəd, but Berber-internally it is not actually all that obvious that it's in fact a *w element. While it is plausible that *yəssəwɣəd would become *yəssuɣəd, Berber internally there is little to stop you from simply saying that it is just a plain vowel that is added to the root.

The plain vowel that is added, however, is qualitatively different from the plain vowel that is added in verbs that in the underived form have a plain vowel already. As in those kinds of verbs like *agəm 'to draw water' the plain vowel becomes *i:

	Aorist	Perfect	Causative Aorist
v̄cc stems	agəm	ugăm	əssigəm
c̄c stems	ăqqəd	əqqăd	əssuɣəd

This difference in behaviour is not at all implausibly explained with the presence of an ancient *w (Pre-Proto-Berber *issīgim or *issiHgim versus *issiwɣid)

The strange thing, however, and this is something Prasse noticed already, the plain vowel that appears in these roots has surprisingly v̄ like behaviour (v̄ is a "plain vowel slot" which depending on whether the vowel pattern requires low vowels or high vowels surfaces as *a or *i,*u respectively. Thus a v̄cc verb in the Aorist (which has a |low-high| pattern surfaces as agəm but in the Perfect (which has a low-high pattern) surfaces as ugăm.

This is most notable in in the nouns of instrument. Nouns of instrument in Proto-Berber are formed by a prefix *s- followed by a |low-high| pattern, where the last high vowel show up as both long and short. So:

• *ăɣnəs 'to put on a veil' -> *a-s-ăgnəs 'veil'
• *ălsəʔ 'to get dressed' -> *a-s-ălsuʔ 'garment', 
• *aləy 'to hang' -> *a-saləy 'hook'
• *arəʔ 'to open' > *ta-s-aruʔ-t 'key'

However things get weird with our c̄c verbs, here the instrumental noun behaves as if it is a v̄cc stem, thus:

• *ăttəl 'to wrap' gives us *a-s-atəl 'envelope of cloth'

So how should we understand this if the plain vowel that appears is supposed to come from *w? Is this *a-s-ăwtəl, and do we have to assume that *ăw > *a /_C? *wC clusters are certainly not so common in Berber, but also not so uncommon that this etymology works without a hitch. The most obvious examples are *a-wtul  pl. *i-wətl-an 'rabbit' and *a-wtəm pl. i-wətm-an 'male'. These can be resolved in a number of ways:

• The VwC cluster occurs accross a morpheme boundary, so that may cause different behaviour.
• The preceding vowel (at least in the free state) is *a and not *ă, which may have cause different behaviour.
• It is possible that these words originally had a short vowel in the stem (this is how Prasse would perhaps reconstruct them), i.e. *a-wətul and *a-wətəm.

There are not many other places where the presumably *w yields *a, but one example might be the word *t-addar-t pl. t-udr-en 'house', which is presumably derived from the verb idir pf. əddăr 'to live'. But, if we must derive this from a dropped *w, we would have to reconstruct *ta-wăddar-t pl. *ti-wədr-en. So now we have an example of not only *ăw shifting to *a but also (a)wă shifting to *a!

Another place where we would predict a *ăw in pre-Proto-Berber is in the agentive formations, which are formed with a prefix -m/n- followed by a |low-low| vowel pattern, where the last vowel may either be short or long. Thus, we see:

• ccc stems: *a-m-ărwal 'a flee-er' and *e-m-ărwăl 'id.' <- *ărwəl 'to flee'
• v̄cc stems: *a-m-araw 'a parent' <- *arəw 'to give birth'
• cc stems: *a-m-ădad 'biter' and *e-m-ădăd 'id.' <- *ădəd 'to bite'

But here instead of getting the expected a as was the case with the instrumentals, we instead find an u!

• c̄c stems: a-n-uḍăf < ?a-n-ăwḍăf 'someone who holds' <- ăṭṭəf 'to hold'.

The instrumental and the agentive therefore cannot easily be reconciled, they give two different answers of what *ăw should yield.

Moreover, reconstructing *t-addar-t 'house' as *ta-wăddar-t also runs into problems. At least in Ghadamsi, in a not atypical te-CăCC-e verbal noun formation, c̄c stems do show an actual *wă, e.g. ta-waɛn-e 'load, burden' < *te-wăɣn-e <- *ăqqən 'to tie up', some varieties of Tuareg seem to show the same behaviour. In verbal nouns *w does show up in other verbs as well, in Central Moroccan Berber, for example the verbal noune of ddu < *əddu 'to go' is tawada < *ta-wad-a (plain vowel *a in the stem not fully understood).

But it is also clear that *w in certain contexts does really yield plain vowels, at the very least *u and *e. For example, the stative verb iwriɣ 'to be yellow' with adjective *wăraɣ- (cf. Semitic *w-r-ḳ!) has an extremely well-attested derivation *urəɣ 'gold'.

There are also clear examples of c̄c verbs with verbal nouns of this type: Central moroccan Berber lləf 'to divorce', verbal noun uluf. And also *idir pf. *əddăr 'to love' has a well-attested derivation *t-udər-t 'life'. 

Prasse would like to derive those *u's from *ta-Hədər-t, where *Hə yields u but we might be able to rewrite the rule where *wə yields u. This would get us in trouble. This gets us in trouble with i-wətm-an 'males', but that can definitely be an analogical restoration from the singular. And while we don't understand the singular t-addar-t 'house', we could explain nicely the plural *t-udr-en as coming from *ti-wədr-en then.

Sometimes *e shows up instead, such as in ăṭṭəs 'to sleep' VN eḍəs 'sleep'. In almost all cases *e is a secondary vowel harmony result of *a, but that cannot be the case for words like this. This is one of the few places in Berber where *e seems to be 'real'. It is tempting to derive it from something like *a-yəḍəs  > eḍəs or *a-wiḍis > eḍəs (and then *a-wuruɣ > urəɣ).

So summarizing all these we end up with a number of conflicting outcomes for what would be assumed to be a *w. Sometimes just behaviing like a consonant *w, sometimes behaving like a plain vowel slot that can just take an low-vowel pattern (instrumentals), sometimes with as an exceptionally *u-colouring plain vowel...

*wă	*wă (*te-wăɣn-e), *a (*t-addar-t)
*ăwC	*a (*a-s-atəl), *u (*a-n-uḍăf)
*V-wə	*u (*urəɣ), *e (*eḍəs), *wə (*i-wətm-an)
*əwC	*u (əssuɣəd)

I'm not really sure how to resolve these competing outcomes. Some I think could be gotten rid of as secondary, especilly the *wă/*wə outcomes, but others... I don't know. Thoughts?

Proto-Berbero-Semitic Primae Waw Verbs

So after our Summerschool course Benjamin Suchard and I were excited to finally get working on a paper together, something we've always wanted to do but somehow the stars never aligned. We had a cool idea about a parallel between the Proto-Semitic and Proto-Berber morphological behavior of roots that probably had an initial *w radical.

Turns out Prasse's Law of Berber Historical Linguistics reared its head ones again. Prasse's Law goes as follows:

Whenever you find a new insight into Proto-Berber historical linguistics, Karl-G. Prasse already saw it but wrote it in such an oblique way that you only understand what he meant after you've reinvented the wheel.

So since someone else saw it, on its own there's not much to publish I think, so let me write up the basic idea here. I might write about some more confounding features of Proto-Berber Primae Waw stem morphology in a next post.

Proto-Berber has a series of biradical verbs which in their underived form have an initial geminate, e.g. *ăffəɣ 'to go out', *ăqqəd 'to burn', *əddu 'to go'. In derived forms, most notably the causative these forms degeminate their geminate and a vowel u appears before it: *əssufəɣ, *əssuɣəd, *əssudu. This looks like whatever caused the geminate in the underived forms is something u-like. So people have imagined that we're looking at some kind of assimilation of *wc > c̄. There is also at least one pretty compelling Berber-Semitic parallel that might suport that *aqqəd 'to burn' sure looks like Sem. *w-ḳ-d 'to burn' (e.g. Ar. waqada).

However, we have a pretty good sense how both normal forms and causative forms are supposed to look and this runs into trouble, let's compare the causatives of this type to the regular triradical type:

	Underived		Causative
	Proto-Berber	Pre-Proto-Berber	Proto-Berber	Pre-Proto-Berber
Regular	ălməd	ălməd	əssəlməd	əssəlməd
Primae-Waw	ăqqəd	ăwɣəd	əssuɣəd	əssəwɣəd

The problem is that, if we assume that *wc yielded a geminate, then we would expect it to have yielded a geminate in the causative as well, but there it vocalises to u.

This is where Benjamin noted: Well in Semitic Primae Waw verbs behave quite weird too, sometimes the *w (notably in underived verbs!) disappears altogether. You can still see this quite nicely in Classical Arabic: walada impf. yalidu not **yawlidu, the causative however, does mainting the *w in the prefix conjugation. ʔawlada impf. yūlidu (ultimatly from *yusawlid-). Thus Benjamin's idea was: what if the geminate found in the underived form isn't due to assimlation at all, but just lengthening of the initial radical of an original biradical -CvC- stem. that would resolve our problem!

	Underived		Causative
		Berb-Sem		Berb-Sem
Semitic	aḳid	< *aḳid	usawḳid	< *usv̆wḳid
Berber	ăqqəd		əssuɣəd

And it's pretty attractive too. You could understand the gemination in Proto-Berber either as simply being there for the verb to fill up the triradical shape needed for the light verb.

But there is in fact some very similar geminative behaviour in heavy verbs (verbs with three vowels), there too we see that the consonant that comes after the Person-Number-Gender prefixes are geminated, which yields alternations between the aorist and imperfect (e.g. əddukəl 'to walk together' impf. ətt-ədukul).

The comparison is note quite perfect because the imperfect of the primae waw verbs does not degeminate the stem geminate (ăkkəs impf. ətt-ăkkăs not *ətt-ăkăs), but that's something that could quite easily be explained with some analogy I think.

Anyway, this explanation reveals a highly specific and unusual morphophonological behaviour of Primae-waw verbs that is shared between Semitic and Berber, a nice point in favour of Proto-Berbero-Semitic I think. But as I said, we got Prasse'd, he said it already in 1973, albeit much more succinctly and much more briefly:

La semivoyelle tombe, mais la 2" subit un gémination compensative. Nous preférons nouns exprimer ainsi, au lieu de dire qu'il y a assimilation des 1" et 2" parce que d'une part le sémitique et l'égyptien connaisent aussi la perte d'une semivoyelle initiale, mais sans gémination compensative, et que d'autre par le berbère lui-même ne connaît pas ailleur des assimilations de cette espèce (p. ex. au milieu des quadrilitères).

Or to put it in English:

The semivowel (of the primae wāw verbs) drops, but the second radical undergoes compensatory gemination. We prefer to express ourselves as such, instead of saying that it is an assimilation of the first to the second radical on the one hand because Semitic and Egyptian also have the dropping of the initial semivowel, but without compensatory gemination, and on the other hand because Berber itself doesn't have any out assimilation of this kind (for example in the middle of quadriradical stems).

Our comment that the loss of the wāw in the underived stem but reappearance in the causative, just like in Semitic is not spelled out by Prasse specifically, although I have little doubt that such behaviour is what he had in mind. I'm curious about Prasse's comment about Egyptian. I'm aware there is some weirdness with the initial w- but it is no doubt conditioned differently on account of Egyptian not having a prefix conjugation, but I'd love to hear how it does function there, if anyone knows, drop a comment!

Issues with Lexical Comparison of Berber within Afroasiatic

I have just finished a really inspiring and interesting course at the Leiden Summer School named "Reconstructing Afro-Asiatic" together with Benjamin Suchard, Marwan Kilani, Ahmed Sosal and Maarten Mous. A big aim of the course was to introduce an audience (and, to be honest, each other) to the specific challenges we see in the reconstruction of our respective Afro-asiatic language families of specialty, Berber (Me), Semitic (Suchard), Egyptian (Kilani) and Cushitic (Sosal and Mous).

In the past years a contingent of Leiden scholars have made steady headway into pre-Protolanguage reconstructions of especially Berber and Semitic, and have found a number of really interesting and strikingly specific morphological commonalities especially between Berber and Semitic, but on occasion have also lead to notable comparisons with Egyptian (Kossmann & Suchard 2018; Van Putten 2018; Suchard & Groen 2021; Suchard 2021). This is an exciting development, and has made me much more optimistic about the prospects of Afro-Asiatic reconstruction. And there is much more than what is presented just in these papers especially with Semitic, but also ni other languages. We can reconstruct quite specific plural formations in common between Berber and Semitic (e.g. Ar. fāris pl. fursān vs. Proto-Berber *a-ǵăllid pl. *i-ǵəldan; Ar. kalb pl. kilāb versus. Proto-Berber *a-ǵălzim pl. *i-ǵəlzam), the prefix conjugation, the suffix conjugation and adjective formation (article on this forthcoming by me very soon!), the shape of the infinitive PSem. *parās³-  PBerb. *a-lămad, and of course the feminine marker *-t.

Still, we are stuck with an issue: even though we can reconstruct really significant amounts of specific morphology, but vocabulary has lagged behind. In my fairly strict estimation, I've so far found about 30 lexical comparisons between Semitic and Berber (my strongest language families of the bunch) that I actually consider compelling. Yet, Gábor Takács who by now has published 15 articles on "Some Berber Etymologies" speaks of an "immense Afro-Asiatic heritage in the Berber lexical stock" and has collected no less than 609 (!) comparisons. That's 20x more than what I have found! Militariev has found 518 entries in his records. Did I not look hard enough?

No, the issue is that I find almost none of them convincing. Militariev's dictionary is, as far as I'm concerned, completely useless. As Lameen Souag has shown years ago, Militariev really lacks the required knowledge of Berber to even be able to recognize obvious cognates within the language family. Takács' work is significantly less problematic in this regard. As a whole he typically correctly identifies cognates within Berber. But Takács has refused to engage with Berberists that have worked on Proto-Berber reconstruction, despite important treatises on the topic having existed for several decades now (they, in fact, typically go completely uncited).

Even disregarding typical methodological issues:

• Comparing isolated words in one language family with isolated words in another;
• allowing for absurd amounts of semantic leeway
• Assuming ad hoc "root extensions" without explanation whenever you have a consonant too many or too few, etc.

the etymologies simply fail to connect with anything we know about the proto-language.

Length alternation between go-on and kan-on readings

Japanese doesn't have many long vowels in native vocabulary, but in Sino-Japanese it has many. But as with the native ones, these are almost invariably the result of contractions of bi-moraic diphthongs.

• 登 Middle Chinese toŋ > Middle Japanese toũ > tō
• 當 MC taŋ > MJ taũ > tɔ̄ > tō
• 通 MC tuwŋ > MJ tuũ > tū
• 刀 MC *taw > MJ tau > tɔ̄ > tō
• 答 MC *top > MJ topu > tofu > tou > tō
• 蹋 MC *dap > MJ dapu > dafu > dau > dɔ̄ > dō
• 丁 MC *teŋ > MJ teũ > tyō
• 鳥 MC *tew > MJ teu > tyō
• 笘 MC *tep > MJ tepu > tefu > teu > tyō
• 中 MC *ṭjuwŋ > MJ tiũ > tyū
• 丁 MC *teŋ > MJ (kan-on) teĩ > tē

For all of these, the length thus stems from an approximation of heavy syllables (not known to Japanese at the time) by making a bimoraic syllable.

But this is not the only origin for vowel length in Sino-Japanese. In a couple of cases fairly simply syllables in Middle Chinese take on a long vowel in Sino-Japanese in the kan-on layer but not in the go-on layer. Such cases fascinate me, because the difference is not easily explained Japanese-internally, and thus seems to suggest a real (and somewhat unexpected) phonetic difference between the go-on and kan-on layers.

The main differences between go-on and kan-on (most of which must be attributed to the underlying Chinese dialects) can be summarized as follows:

• Voicing distinction:
•     go-on: voicing distinction is maintained MC t, d > MJ t, d;
•     kan-on: MC t ,d > MJ t, t
• Initial nasals:
•     go-on: Initial nasals are maintained MC m, n > MJ m, n;
•     kan-on: become prenasalized stops MC m, n > MJ b, d
• Vowel breaking:
•     go-on: MC e, æ and ɛ break before velar finals to ~ia MC *mek > *miak > MJ myaku; MC *bæk > MJ byaku; MC *tṣɛŋ > MJ syaũ > syō;
•     kan-on: doesn't break (and reflects e  (= e) differently from æ/ɛ (= a)): MC *mek > MJ beki; MC *bæk > MJ haku; MC tṣɛŋ > MJ saũ > sō
• Palatalization of Division III:
•    go-on: Division I was still pharyngealized/ Division III non-pharyngealized initials, are not distinguished in this layer: MC 五 *ŋo [ŋˁə] > J go;  MC 魚 *ŋjo [ŋə] > J go; 
•     kan-on: Division I has become unpharyngealized/Division III has become palatalized, they are distinguished in this layer: MC 五 *ŋo [ŋˁə] > J go;  MC 魚 *ŋjo [ŋʲə] > J gyo

But there is a group of syllables that has short vowels in the go-on layer, and a long vowel in the kan-on layer. This happens in the Division I -uw and its Division III equivalent -juw rhymes.

Compare:

• MC 口 *kʰuw; go-on: ku; kan-on kou > kō
• MC 頭 *duw; go-on: du; kan-on tou > tō
• MC 母 *muw ; go-on mu, mo; kan-on bou > bō
• MC 留 *ljuw; go-on ru; kan-on riu > ryū
• MC 酒 *tśjuw; go-on syu; kan-on siu > syū
• MC守 *śjuw; go-on syu, su; kan-on siu > syū
• MC 九 *kjuw; go-on ku; kan-on kiu > kyū

As we have already seen, long vowels seem to stem from diphthongs. It therefore stands to reason that in the kan-on layer both of these finals must have been diphthongs, whereas in the go-on layer they were monophthongs. I am not able to reconstruct why Baxter & Sagart originally decided to represent these rhymes with a final w and if sino-japanese played a role in that. I would assume so, but I'm just reinventing the wheel here for my own satisfaction.

Knowing that Japanese *o can come from ə (transcribed in Middle Chinese as <o> typically) it seems attractive to reconstruct the kan-on reflex of the division I rhyme as *-əw which then naturally gives rhis to kan-on ou > ō. But the Middle Chinese that fed into the go-on layer, this must have been a monophthong (or at least a diphthong that did not allow approximation with <ou> = [əu]) so something like *-u.

口 MC kʰuw ; MC Go-on dialect: kʰˁu > go-on ku; MC kan-on dialect kʰəw > kan-on kou

As for the -juw rhyme, it is more difficult to reconstruct what the rhyme would have looked like in the Kan-on layer of Middle Chinese. But it can't have been *-jiw, as it behaves distinct from it (-jiw becomes -yū in the go-on layer). Here too -jəw seems like an option. A -jow rhyme does not exist so, we could slot it in there. But other rhymes that have MC o in it, show up with -yo- rhymes in kan-on, so presumably the vowel of the diphthong was higher, e.g. ɨ. Again, in the go-on dialect these must have monophthongized.

九 MC kjuw; MC go-on dialect ku > go-on ku; MC kan-on kʲɨw > kan-on kiu

It's interesting to note that this reconstruction (əw besides ɨw) perfectly mirrors the middle Chinese reconstruction of Zhengzhang Shangfang. Probably not a coincidence (did not check what that reconstruction is based on), but an independent conclusion on my end!

I say that in the MC go-on dialect, these diphthongs were monophthongized. But it is actually likely that we are rather dealing with a lack of of diphthongization. The -uw and -juw rhymes are reconstructed as coming originally Old Chinese monophthongs -u and -uw.

OC *-o > MC Div I -uw; Div III -ju
OC *-u > MC Div I -aw, -uw; Div III -juw

So one might in fact imagine that the go-on layer, which also in its lack of palatalization in the Division III appears appears to predate Middle Chinese proper, is actually continuing the Old Chinese vowel system in this regard.

Japanese o can come from both ə and o, so OC *-o being borrowed with Japanese -u is a bit surprising. But there are good reasons to think that the Old Japanese o was actually a rising diphthong wo, which possibly did not feel like a good match for approximating Old Chinese *o. Thus we might tentatively reconstruct the Chinese dialect that feeds into the go-on layer with this distinction maintained:

口 *kʰˁo > Go-on ku
九 *ku > Go-on ku

Whereas the Chinese dialect that feeds into the kan-on layer has all the typical Middle Chinese breaking of these vowels:

口 *kʰˁo > *kʰəw > Go-on kou
九 *ku > *kʲɨw > Go-on kiu

Some musings on Japanese onbin changes

Japanese is a weird language. In its modern form it is surprisingly regular and nice, but to compensate its historically attested linguistic history seems to be utterly filled with irregular sound laws. It's not always clear to me whether these sounds laws really are irregular, or whether the spelling of Japanese (and tendency towards historical spellings) just renders these developments extremely opaque. My sense is that the culture of Japanese historical linguistics seems less concerned with regular sound change than is typical from researchers in the indo-europeanist tradition. Whether that lack of concern is truly because it is concerningly irregular, or just not figured out well enough, I do not really know. But in this blogpost I'm going to attempt to present some thoughts on it that will hopefully make it look more regular...

Disclaimer: what I know about Middle Japanese is based entirely on textbook descriptions like Frellesvig's A History of the Japanese Language, and some searches on Wiktionary. Both are thoroughly in the camp of "not so bothered by irregular sound laws", so take all I say with a grain of salt, but perhaps it'll be resonating with someone, perhaps I'll be reinventing the wheel, and all this has been figured out but is just badly described in English literature (My Japanese is definitely not good enough to read Japanese historical linguistics in Japanese).

So my issue today is with the so-called onbin 音便 "euphonic" sound changes. These are reductions of certain consonants with the vowel u/i. that happens at some point early in the Middle Japanese period. It seems that historical spellings continue to represent much more archaic phases of these words.

What is worth knowing his that what today in Japanese are voiced consonants, were originally pre-nasalized consonants. The onbin sound changes often come in two types vocalic and consonantal, and while to some extent vocalic versus consonantal seems to be a Western versus Eastern dialects distinction, Standard Japanese has a rather eclectic mix of both (presumably because Standard Japanese is based on the Eastern Tokyo dialect, but for centuries before that the Western Kyoto dialect was the central prestige dialect). For now I will assume that the irregularity in terms of consonantal versus vocalic outcomes is purely the result of dialect mixing.

The Onbin changes are as follows:

• pi  > Vocalic u; Consonantal C (nipi-ta 'new field'> niuta > nyūta; > niCta > nitta)
• pu > Vocalic u; Consonantal C (taputwo- 'exalted' > tauto > tōto; > taCto > tatto)
• bi > Vocalic ũ (ũ causes pre-nasalization of the next word); Consonantal N (N likewise causes prenasalization of the next word) (ywobite 'to call and' > yoũte > yōde ; > yoNte > yonde)
• bu> Vocalic ũ; no known  consonantal outcome. (*sabu-sabu-si 'lonely'> saũsaũ > sōzō)
• mi > Vocalic ũ; not sure if there is an attestecd consonantal outcome,
• Frellesvig says N, but there I disagree and that's why I'm writing this blogpost (ko-miti 'small road' > koũti > kōji)
• mu > Vocalic ũ; not sure if there is an attested consonantal outcome, Frellesvig says N. *pi-muka 'personal name' < east) > piũka > hyūga).
• ki > Vocalic i; Consonantal C. (okite > oite; tuki-tati > tsuitachi; ikite 'to go and' > iCte > itte [only example of consonantal outcome?]).
• ku > Vocalic u; no known consonantal outcome. (yoku 'well' > you > yō (not typically reduced in standard Japanese but seen in yōkoso 'welcome')
• gi > Vocalic ĩ; no known consonantal outcome. (tugite > tuĩde > tsuide)
• gu > Vocalic ũ; Consonantal N (kagupasi- 'fragrant' > kaũpasi > kōbashī; > kaNpasi > kanbashī; But also a non-reduced form (a loan from the literary language?) survives: kagupasi > kaguwashī.
• ni > Consonantal N; no known vocalic outcome. sinite 'to die and' > siNte > shinde.

However, these are not the only reductions of such syllables attested in the language. Especially the nasal syllables, i.e. mi, mu and ni leave no syllabic trace whatsoever, but instead purely leave a trace as pre-nasalization.

Thus pi-muka(si) 'turning to the sun; east', same etymology as hyūga about in Modern Japanese becomes higashi 'east'.

ko-miti 'small road' shows up as a surname kodi > koji こぢ 【小路】

uri-nipa 'selling courtyard' > uriba(but nipa uncompounded has no reduction and becomes niwa 'garden' these days ba can also be used on its own outside of compounds with a specialized meaning "place")

pumi-te 'writing+hand' > fude  not funde

This last one is especially puzzling because a phonetically identical verbal form pumite 'to step and' does have a Consonantal outcome: funde.

Frellesvig considers these non-moraic outcomes to be part of the same shift, and thus neglects to explain how pumite "brush" and pumite "to step and" would have different outcomes. As I understand it, the reason why these two shifts are considered part of the same shift is because they continued to be spelled historically so it's hard to see any historical layering that might be going on. Still I would be inclined to see this in two different layers.

What is striking about the forms that have only pre-nasalization as a trace is that they are all nominal compounds, whereas the moraic onbin changes occur both before morphological suffixes and in nominal compounds. This is a weird distribution, since if anything 'suffixes' are more part of the preceding word (and thuse suspectible to reductions sooner) than compounds, which may still be treated as somewhat separate words.

It is perhaps attractive to think of forms such as fumite 'to step and' as infinitive fumi 'stepping' + an auxiliary verb te, and that these were still considered separate words at the time that pumite 'brush' was a compound. And at this stage a first reduction of mi, mu and ni took place yielding only to pre-nasalization. And then in a later stage the full set of true onbin changes take place. But it seems to me that this is not so easy. The -te gerund suffix has been around from the Old Japanese period, and there is nothing to suggest that it was considered a separate word except the difference in behaviour that we see here.

But taking the difference in behaviour as the reason to analyse this as a separate word, is of course circular. It has to be a separate word because it behaves differently, and because it behaves differently we conclude it was a separate word. This may not even be wrong just unfalsifiable and unprovable.

Then there is of course the issue of pairs like higashi and hyūga which both seem to come from the same compound *pi-muka-, where the latter avoided the proposed early compound reduction while the form did not. This could be understood perhaps as the same compound being re-compounded at a later stage, after *pi-mukasi turned into pigasi and the form was no longer understood to be such a compound.

There is no satisfying solution to this for me yet, just trying out some thoughts... It's difficult to find the examples of compounds where only pre-nasalization is left, in part because authors tend to just consider them part of onbin changes, and not a separate sub-group (which it should be, whatever solution you come to). fude, -ba, koji and higashi are the only ones I'm aware of. Are they the only once around? Are there more? Would love to hear from anyone who knows of a list of such forms. Might be easier to to make sense of the patterns that way.

====

There is still the other issue that even the Moraic onbin changes are by no means regular. Where *nipi-ta gives us nyūta, the normal non-onbun outcome of nipi is well-attested in place names and some words too:

*nipi-duma > nihizuma > nīzuma 'new wife', not **nyūzuma

*nipi-gata > nīgata (a placename), not **nyūgata

*nipi-pada > nīhada 'skin where two lovers first touch'

This last one is clearly a recent compound as hada did not even undergo the expected shift to wada. The regular outcome should have been nyūwada. And this might the the "solution" to understanding these compounds, they might be compounds that all post-date the onbin changes. In many cases that seems defensible, and even in the cases where the compound does predate the onbin changes, they were probably morphologically transparent enough for them to be analogically re-compounded without the onbin changes at a later stage.

Anyway, those are my thoughts for now. Again: happy to hear of other examples of compounds where *mi, *mu and *ni only yield pre-nasalization, without a moraic N or ũ.