Glottal, stop

A gentle refutation of glottalic theory and a less gentle assertion that you are not capable of thinking for yourself

Large parts of philology are still the preserve of received wisdom rather than actual enquiry. The best illustration of this is our understanding of the Proto-Indo-European laryngeals.

For some reason, everyone in the field of Comparative Mostly-Peaceful Steppe-Migrantese is obsessed with the idea that “largnyeals” were, by nature, some kind of guttural fricative sound. In what I can only presume constitutes a fit of madness, some have even proposed they represent uvular stops.

However, if you spend more than a few seconds considering this, it doesn’t seem likely. Even Beekes took time out of his life’s work purging the Greek inherited lexicon from the face of the earth to point this out:

I think that there is a general objection to velar fricatives (…). The general objection is that they are not the most probable candidates for a group of sounds, the 'laryngeals', of which one of the most remarkable things is that they were often vocalized. This phonetic property of the laryngeals makes it rather improbable that they were velar fricatives. (For uvular fricatives the same objection holds.)

We do know that at least one of the laryngeals ended up as a velar (or uvular) fricative sound in Anatolian. However, this isn’t good enough grounds to assume the same in PIE.

But in spite of this, Beekes is very keen (as are many) on the idea of laryngeals as pharyngeal fricatives. This assumption appears to have been made originally by Keiler via analogy to Semitic, which contrasts voiced and unvoiced pharyngeal fricatives with velar-uvular and “glottal” equivalents.

It’s augmented with reference to the interior Salish language Shuswap. The only major descriptive study of the language was published in 1974 by Aert Kuipers, via whom it eventually reached the attention of Beekes.

We’ll talk about all this in detail some other time. Suffice it to say that the standard assumption is that *h₁ represents a glottal fricative h, *h₂ a pharyngeal fricative ħ/ʕ, and *h₃ a labialised version of the same, ħʷ/ʕʷ.

Hʕ̞lˀt ywr hʕ̞ʷˀrsɰs

However, there’s no good reason to assume this. In fact, the only argument for this being the case is that *h₁ doesn’t colour an adjacent *e… and that we already call these sounds “laryngeals” as a term of art.

This is why you need to be careful with your terminology, because if you aren’t, it influences your thinking in ways that make your thinking less effective. We can call this weak version of the Sapir-Whorf hypothesis semantically induced stupidity.

In reality, there’s many sounds that could be vocalised in marginal positions that wouldn’t change the character of an adjacent *e. But nobody’s bothering to give them a decent look. I’m quite shocked that nobody’s at least tried to make a serious case for PIE *h₁ as ɰ on the basis of Shuswap.

However, the subject of today’s study of terminology-induced brainrot is a different kind of supposed para-glottal sound.

Infectious pharyngitis

Glottalic theory is one of the stranger ideas in philology that has unusually outsized staying power. It’s come and gone in numerous guises and under numerous proposals over the years.

Let’s briefly review it as described by its most sane and cogent advocate, Frederik Kortland.

In brief: instead of the usual tenuis, voiced, and “voiced aspirate” stop series, PIE actually had a system like this:

            Tenuis               Voiced            Preglottalised

Labial        (p)                  b                     ˀp

Coronal        t                   d                     ˀt

Palatal        c                   j                     ˀc

Velar         k, kʷ              g, gʷ                 ˀk, ˀkʷ

(Note: If the idea of palatal stops so offends you as to require it, feel free to interpret the palatal and velar series as velar and uvular and shuffle the labialised ones around according to your preference. It really doesn’t make much of a difference for the purposes of this article. We’ll do the dorsals later.)

In this system, the tenuis stops take the place of the series usually described as “voiced aspirates”. Preglottalised stops take the place of “voiced” stops. The “voiced” stops are the traditional reconstruction’s tenuis series.

For this arrangement, Kortland claims direct evidence from Indo-Iranian, Germanic, Baltic, and Armenian (which should be a red flag in and of itself) and indirect evidence from Slavic, Greek, Latin, and Indo-Iranian again.

In the rough, his argument precedes thus:

1. The only direct evidence of “voiced aspirates” comes from Indic, with indirect evidence from the aspirates in Greek and fricatives in Italic. All other branches have plain voiced stops, including the closest relatives of the three aberrant groups (Iranic, Phrygian, and Celtic respectively). This is enough to cast suspicion on the idea that they were an original feature.

2. It could in fact be the case that Sanskrit, Greek, and Latin simply happen to have innovated an underlying voiced stop series in ways that suggest they were originally “aspirated”, and we’ve held onto this idea because Sanskrit, Greek, and Latin are also among the earliest attested and most widely studied languages when it comes to the reconstruction of PIE.

3. The main obstacle to reinterpreting “voiced aspirates” as simple voiced stops is that we already have an assumed series of voiced stops in PIE.

4. But giving up this assumption allows us to create much more satisfying explanations for specific features seen in the families cited above.

5. This possibility wasn’t accounted for in the past because phenomena like glottalisation weren’t especially well known until long after the “conventional” reconstruction of PIE obstruents was elaborated.

The material given under point (4) is as follows:

1. Lubotsky’s Law in Indo-Iranian. Lubotsky identifies 14 Sanskrit lemmas (with 5 correspondences in Old Iranian) in which a laryngeal is lost before a voiced (or “glottalic”) consonant when it is followed by another obstruent with no compensatory lengthening of the preceding vowel. Using G as a cover symbol for “glottalic”, we can frame this law as VHGC > VGC, e.g. *peh₂ǵ- > pājas- but also *pajra-. This is explicable as a result of the laryngeal merging with the glottalic element of the “voiced” consonant. This is the key argument for pre-glottalised consonants as opposed to any other form of glottalisation.

2. Kortland argues that in modern Sindhi and its neighbours Kutchi and Saraiki – which phonemically distinguish implosive voiced stops from plain voiced stops – implosives are a direct continuation of PIE voiced stops while plain voiced stops reflect PIE “aspirates”.

3. He also argues that the current phonology of Punjabi can only be explained by the recent existence of pre-glottalised stops in its history: “voiced aspirates” in Punjabi have become tenius and yielded a low tone on the following vowel. Since voiceless aspirates remain aspirated, Kortland argues that this presupposes that “voiced aspirates” were not actually aspirated at the time when this change occurred. Since “plain voiced” stops don’t produce a lowered tone, he argues this is evidence for the preservation of their “glottalic” character.

4. There are no voiced aspirates in Kashmiri or Nuristani.

5. In Classical Armenian we see a 3-way contrast between voiceless, voiced, and aspirated stops similar to Greek. However, because Armenian is Armenian, their origins are different: PIE’s *ptk become CA PIE pʰtʰkʰ PIE *bdg > CA ptk, and PIE *bʰdʰgʰ > CA bdg. In some Eastern Armenian dialects, sounds corresponding to CA p,t,k are realised as ejectives. Kortland presumes these to be continuations of original PIE preglottalised consonants and concludes they demonstrate glottalisation is “ancient” in the reflexes of PIE voiced unaspirated stops.

6. In Germanic, glottalic theory is recruited to explain preaspiration in Icelandic, Faroese, and some dialects of Norwegian and Swedish, and the stød in Jutlandic Danish. It’s also recruited to explain a variety of other aspects of Germanic phonology – including but not limited to: gemination, the High German consonant shift, sonorant devoicing, and the assimilation of nasals to stops in North Germanic. And, for some reason, the tendency of medial stops in English to become gloˀal stops.

7. Lachmann’s law in Latin and Winter’s Law in Balto-Slavic specify the lengthening of a short vowel before a voiced stop in various environments. (In Lachmann’s, the lengthening occurs when a voiced stop is immediately followed by an unvoiced stop, in Winter’s, it is more general, but can be blocked when the voiced stop is followed or preceded by a variety of consonants). Kortland takes this as evidence of glottalisation rather than lengthening in both cases. In BS, he states that his preglottal stops dissolved into a glottal stop and a voiced stop, the first part of which “merged with the glottalic reflex of the laryngeals”.

I can’t reproduce Kortland’s entire arguments here in encyclopaedic detail because this piece is going to be long and boring enough already. I hope I’m not misrepresenting the general thrust of any of them here, because I want to evaluate them on their actual merits. If you think I have failed to do this, please let me know.

With that out of the way, let’s go.

Lubotsky’s law

Lubotsky’s law appears to hold with some reliability in 12 of his provided examples (in 2, the existence of a laryngeal is difficult to confirm). There are also a small handful of words only attested in Iranic such as *teh₂g- “command, arrange order”, which surfaces in Greek as τᾱγός, (presumably from a full grade, with τάσσω from zero-grade *th₂g-), but in Old Persian as ham-ataxsata (“they have put in order”).

These plus a few direct Iranian matches point to preforms where a laryngeal has been deleted without a trace, which suggests that Lubotsky’s law is an Indo-Iranian specific example of the kind of laryngeal deletion seen also in PIE.

However, its existence doesn’t demand a preglottal element in voiced obstruents for it to work.

In Proto-Indo-Iranian, the laryngeals merge as a unified sound *H. This coincides with a collapse of the distinction between *e, *o, and *a.

This is significant because if we liberate ourselves from the need to see *H as a glottal stop (as per conventional wisdom), we can explain Lubotsky’s law as a re-emergence of the tendency to delete a laryngeal in a sequence with more than 2 other consonants.

To simplify this radically without having to give a full rundown of optimality theory as applied to PIE (and PII): In PIE, the maximum allowable cluster in any syllable is 2 consonants (except word initially, where one consonant is *s, in which case 3 are permitted).

If more than 3 consonants collide word medially, they must be resolved as belonging to either the coda of the first syllable or the onset of the second, which means that any proposed syllabification has to create a valid onset cluster, and a valid coda cluster.

If it’s impossible to do this with the sounds available, one or more consonants may be considered “extrasyllabic”, in which case they are usually deleted. Sometimes, they may be metathesised or a “valid” cluster created via epenthesis.

We can explain Lubotsky’s law as the resolution of such a phonotactic conflict. In the example of *peh₂ǵ-, PII *paHjas- syllabifies as *paH.jas and the laryngeal is retained, but *paHjra- presents us with two untenable options: *Hj is not a valid coda cluster in PII, nor is *jr a valid onset cluster. *H is therefore deleted, because this preserves the semantic and pragmatic nature of the underlying morphemes with as little possible phonetic change to the output at the stem level.

As to why this only happens with voiced stops: PIE and PII both have a tendency to disallow two similarly-articulated sounds to appear together in sequence. If we presume *H was considered to have similar properties to voiced stops (i.e. it was voiced), its loss here makes perfect sense.

My suggestion for the identity of this sound is an alveolar, post-alveolar, or retroflex voiced approximant [ɹ]~[ɹ̠]~[ɻ]. This would also explain why this sound vocalises to *i in certain positions. PII i would have been the nearest equivalent vowel to [ɹ]~[ɹ̠]~[ɻ].

It would also explain its apparent survival (per Beekes) between vowels in hiatus in Avestan. Beekes interprets this as a glottal stop, but a (by then) weakly realised non-phonemic glide of some kind would also make sense. As we’re about to see, “weakly realised non-phonemic glides” are a recurring phenomenon in Indo-Iranian (and potentially also in Balto-Slavic).

Sindhi

Traditional accounts of the origin of implosives in these languages explains them as a unconditioned development of initial voiced stops and voiced geminates arising from the Middle Indic simplification of medial clusters (b- d- ɖ- j- g- and -bb- -dd- -ɖɖ- -jj- -gg- all > ɓ ɗ/ᶑ ʄ ɠ). Voiced geminates have been noted to have similar aerodynamic constraints and articulation to implosives, indicating that the change happened word medially in the first instance before spreading to initials.

We can interpret this development as a contrastive “strengthening” of voiced stops and their geminates against the general Middle Indic tendency for intervocalic stops to weaken, first by voicing or degemination where relevant, then by spirantalisation, and then by reduction to a weakly articulated semivowel, before vanishing entirely.

Kortland objects to this analysis on the grounds that there are some anomalies in the alternation of the expected distribution of j and ʄ . He posits the following explanation:

1. Because Sindhi keeps Sanskrit y and j separate, y > j in Sindhi while j becomes ʄ (which he renders ‘j). He cites for this: jō/jā 'who' (< yah, yā), jō 'because' (yatah), jańyo 'sacred cord' (yajñopavīta-), jāhu 'coitus' (yābha-).

2. Medial y sees similar treatment: Skt.-īya- (Pali -īya- or -iyya-) is reflected in Sindhi as -ija-, Sky. diyate “to be given” becomes S. ᶑijan(u) (< Prakrit dijjai, Pali diyyati).

3. This indicates that OI y-, -yy- became j in Sindhi while j-, -jj- became ʄ.

4. But by contrast Skt. cʰidyate “to be broken” becomes S. cʰiʄan(u). (This isn’t an isolated example, but these special characters are doing my nut so you’ll have to take my word for it).

5. Similarly we seem to get ʄ from jy (bhaʄan(u) 'to be broken' < bhajyate etc).

6. We also see implosive ʄ arise from -ry- (kārya- > kāʄ (u) “ceremony, work”) and -rv- (carvati > caɓan(u) “chew”), but not -vy- (kārtavya- > katabu “business”) .

His answer to this is that “r was not assimilated to a following j or v, but developed into glottal closure. I think that -ry- and -rv- became -dy- and -dv- at an early stage and subsequently shared the regulaar development of these clusters in Sindhi.”

He also suggests on the basis of the stem pēʄ- which is unexpected by his rule (< *péh₃-ye- via Prakrit peyya) must reflect a retention of the laryngeal in an early Indo-Iranian form *paHy- which has caused the creation of a marginal phoneme *ˀy which also became ʄ. He also suggests that the root jiva- (< *gʷih₃w-) shows a dissimilation conditioned by the retained preglottalic residue of the laryngeal before w.

There are numerous flaws with this argument. The most obvious is that the late retention of y in Sindhi doesn’t preclude ʄ from arising from j, -jj-. Furthermore, Kortland appears to have confused some of the precise developments in the Middle Indic clusters he cites. Let’s break them down:

dy, jy: A sequence of any dental + y from Sanskrit reliably yields a palatal in Middle Indic (ty > c, th> cʰ, dy > j, dʰy > jʰ). Medially, these are always geminates, (-cc-, -ccʰ- -jj-, -jjʰ-). Likewise, palatals + y regularly resolve as simple or geminate palatals initially and medially respectively. We’d thus expect them to fall together with medial voiced geminates from other sources in early Sindhi and be caught up in the expected shift -jj- > -ʄ-.

ry: Also regularly becomes -jj- in Middle Indic, and would thus be caught up in the implosive shift.

rv, vy: In medial clusters, v reliably assimilates a preceding or following r, giving -vv- which is then strengthened to -bb-, which is treated as -bb- from any other source in terms of the implosive shift, including -dv-, which is also a regular MI development.

y, yy: Insofar as medial -y- survives to Sindhi as -j- it must do so via gemination. This change is seen in many Indo-Aryan languages, where -y- becomes -yy- and then, in most cases, -jj-. It’s particularly common in the case of passives formed from -ya-, as in diyate above and its Prakrit and Pali reflexes.

In Sindhi, however, this geminate -yy- must have remained separate from -jj- and resolved as -j- instead, forming a chain shift when j- -jj- became implosive ʄ-, -ʄ-.

The only thing we really have to explain here then is vy, which becomes -b- medially but w-/v- initially. We have to assume that this shift happened after the implosive chain shift, or that it occurred as part of a wider resolutions of geminates: Sindhi is conservative in preserving MI initial w-/v-, but intervocalically its -w- is a lenition of MI -m-, and its -m- is a weakening of MI geminate -mm-.

It’s possible that the same process that reduced geminate voiced stops to implosives also created the environment that caused medial weakening of resonants, triggering the reflex of MI -vv- and -yy- to strengthen to -b- and -j- while their initial counterparts shifted by analogy to the medials.

The preservation of initial w- from vy- would represent the only “incomplete” step in the chain shift that gives *y > j, *yy > jj and *b- *d- *ɖ- *j- *g- and their medial geminate equivalents > ɓ ɗ/ᶑ ʄ ɠ.

Beyond this: contention that Sindhi’s implosive stops are a direct preservation of sounds found in Proto-Indo-European stretches credulity. Especially if laid in parallel with Kortland’s own arguments against Sanskrit “voiced aspirates” as a continuation of a genuine PIE phonation.

Just as Sanskrit is an aberration within Indo-European, Sindhi, Kutchi, and Saraiki are aberrations within modern Indo-Aryan, being the only languages to display implosive consonants. They are also clearly innovative in comparison with other Indo-Aryan languages, with a much larger and more complex consonant inventory than the rest of the family.

It’s of note that the many much more phonologically conservative branches of Indo-Aryan show no evidence of implosives, and ample evidence of the “voiced aspirate” group. Some even resolve historical “voiced aspirates” as ejectives, which seems to be an intermediate step in the process of tonogenesis in related dialects.

Punjabi

Per Kortland: “In this language, the voiced aspirates have become voiceless and unaspirated, yielding a low tone on the following vowel, e.g. kòṛā ‘horse’, Hindi ghoṛā. Since the voiceless aspirates have been preserved as a separate category, it appears that the dh series was not phonemically aspirated at the time of the devoicing while the glottalic stops were preserved at that stage.”

This is a particularly surprising comment as it completely misunderstands the actual nature of the phonemes under scrutiny (another consequence of inherited imprecisions in terminology).

“Voiced aspirates” as they are usually called in Indo-European studies are not in fact “aspirates” at all, they represent a completely different type of phonation which modern linguistics refers to as “murmuring” or “breathy voice”.

True “aspirated” voiced stops are extremely rare, and may not exist at all. They are said to occur in exactly one natural language, and in that case might actually represent cluster of a voiced consonant and its equivalent aspirate.

We therefore shouldn’t necessarily expect “true” (voiceless) aspirates to behave in exactly the same way as their “voiced” counterparts: their “voicing” and “aspiration” are not separable qualities: they’re a “conjoined” feature of murmured phonation.

This explains why Pubjabi murmured consonants appear to have “become voiceless and unaspirated”. They’ve instead become tenuis by losing their murmured phonation. The statement about their voicing also doesn’t hold entirely true, as previously murmured stops in Punjabi surface either as voiced or voiceless in different environments.

The loss or influence of murmured phonation is seen to give rise to a “low tone” in a handful of other Indic languages. In Gujarati, proximity to murmured stops has created contrastive murmured vowels as well through a process of lowering or reduction.

The “default tendency” in Proto-Indo-European stops

The explanation of voiced aspirates as “murmured” consonants also allows us to explain their tendency to disappear in other branches: if “murmuring” on stop consonants starts to lose its contrastive force absent any other changes, the next nearest equivalent sound will be “plain” voiced stops in most cases.

We would actually expect this as PIE seems to have operated on a tendency towards making all voiced consonants “murmured” by analogy: two voiced stops can’t occur together in a root, but two murmured stops can, and the number of roots with two murmured stops far outnumbers those with one murmured and one voiced.

It’s quite possible that in most branches this process simply “edged out” the distinction between voiced and murmured stops over time, with learners re-analysing the contrast as a simple voiced/voiceless distinction.

This provides a neat explanation for the disappearance of the three-way contrast in Celtic, Albanian, Anatolian, Balto-Slavic, and potentially Tocharian (which also loses its voicing distinction, either as a parallel process, or as a later development). We should see this as the “default resolution” of the PIE three-way contrast.

This leaves us to explain Italic, Armenian, Germanic, Greek, and Indo-Iranian.

Greek: murmurs develop into true aspirates

Greek’s treatment of breathy voice is the simplest on paper, and parallels a similar development in some varieties of Chinese. This allows the synchronically-unstable PIE distribution of stops to be maintained while “translating” the opposition into something more stable. This likely represents another data point in favour of Greek’s early exit from the PIE dialectical system, and its extreme phonological conservatism.

Indo-Iranian: persistence of “breathiness” as a contrastive feature

The usual reconstruction of the development of PII assumes that it retains PIE’s murmured stops and added a set of “true aspirates” conditioned when a stop was followed immediately by a PIE laryngeal.

However, since the emergence of aspirates in Indo-Aryan appears to follow the second palatalisation (i.e. *kʰ doesn’t become *cʰ before i, y, etc), it’s actually possible that the creation of aspirates is an entirely Indic phenomenon.

In Proto-Iranian, the conventional assumption is that murmured stops merged with voiced stops, and that aspirates became fricatives. But since all voiceless stops became fricatives prior to another consonant in Proto-Iranian as well, it’s entirely possible that aspirates never existed as a phonemic series here, instead representing the resolution of a sequence *CH in Proto-Iranian itself from which *H was later lost.

If this is the case, we can neatly explain Kortland’s point (4): aspiration is absent in Iranian and Nuristani because Iranian and Nuristani are also examples of the “default” paradigm for murmur resolution, which might even have occurred at the dialectical level in Proto-Indo-Iranian (as neither Iranic nor Nuristani show any traces).

It’s only the Indic branch of Indo-Iranian that preserves murmur (and innovates aspiration) as phonemic features. Indeed, both murmur and aspiration, as we’ve mentioned above, appear to have retained a strong contrastive quality in all Indo-Aryan languages, with many modern members of the family expressing contrastive aspiration and murmur on nasals, liquids, fricatives, flaps, and even in some cases semivowels and vowels.

Why exactly this is the case is anyone’s guess, but Indo-Aryan seems to be very prone to “phonological overcrowding” with interesting marginal consequences. It’s possible that a unique Indic resolution of a stop followed by a non-syllabic “laryngeal” as an aspirate served to maintain a distinction between murmured and voiced stops which would otherwise have been lost.

Germanic and Armenian: chain shifts

In Armenian and Germanic the loss of murmuring triggers a compensatory chain shift with two very different outcomes.

In Armenian, it’s likely that the loss of murmuring “crowded out” the space previously occupied by voiced stops, leaving them as voiceless/tenuis, while PIE’s voiceless series became aspirated via compensatory fortition.

Here we can take a moment to object to point (5) of Kortland’s argument for the glottalic theory: there’s no evidence that dialectical ejective/glottalised stops in Eastern Armenian are anything other than a modern aerial realisation of historical voiceless stops.

In early Greek loanwords into Armenian, Greek’s voiceless, voiced, and aspirated stops correspond more or less exactly to their phonological equivalents in Classical Armenian. Furthermore, some EA dialects of the modern age are recorded as pronouncing the voiced stops of CA as “murmured” or “breathy voiced”. I somehow doubt that Kortland would conclude this is evidence of the original realisation of PIE voiceless stops preserved only in Armenian.

Moreover, as we’ll cover below, “glottalised” and “preglottalised” for which read “implosive” stops, although they do pattern together when they both occur in the same language, are not directly phonologically equivalent, and we would not expect one to develop into the other.

In Germanic, it’s possible that the Grimm’s law chain shift was conditioned by the early spirantilisation of voiceless stops before voiced and murmured stops had a chance to merge, which allowed both to occupy new territory while remaining phonemically distinct.

A brief diversion into Icelandic phonotactics

This is where we can address Kortland’s point (6): supposedly “glottal” retentions and reflections in Germanic.

The only word I can use to describe these is “incoherent”.

Not only do they attempt to cast phonological features of modern Germanic languages so far back into the past that they would represent retentions from Pre-Proto-Germanic (as with many examples above), they do so in a way that blatantly disregards existing diachronic explanations for these features.

I’ll sketch this out most fully with reference to North Germanic “pre-aspiration” as most famously reported in Icelandic.

Kortland wants pre-aspiration in Icelandic, Faroese, and dialectical Swedish and Norwegian to reflect a retention of PIE “preglottal” consonants as a direct continuation of what we conventionally call Proto-Germanic “voiceless” and therefore PIE “voiced” stops.

While it’s trivially obvious that PG’s voiceless stops often become aspirates (including allophonically in English), the big problem with Kortland’s analysis is that, in all cases, “pre-aspirates” are not distinct phonemes.

Instead, they appear to be allophones of aspirates created by the metathesis of the “aspiration” of a consonant. This is treated for as a separate sound h (i.e., “aspirates” in these languages, regardless of their historic origins, are considered bisegmental for phonotactic and phonetic purposes).

In Icelandic, “pre-aspirates” occur only in the codas of syllables with short vowels. “Post-aspirates” only occur in syllable onsets, or following long vowels and voiced sonorants. Further, Icelandic allows no aspirated geminate consonants, and resolves a combination of an aspirate + an analogous non-aspirate as a pre-aspirate (as seen in the neuter sigular (marked by unaspirated -t) of words ending with -tʰ).

This lets us explain “pre-aspiration” as an allophonic resolution of a word form that violates Icelandic’s phonotactic constraints. h can only occur before a vowel in syllable onsets, after a non-geminate consonant word-finally, and following a short vowel in a stressed syllable word-internally.

When h occurs as the second element of a consonant cluster (which for our purposes includes “aspirated” stops and their geminates (p)ph, (t)th, (k)kh, or an aspirated stop + n, l) it shifts to the coda of the preceding syllable in order to maintain these constraints.

In the case of “geminate aspirates”, gemination is also lost in the process, which also shortens a long vowel in a stressed syllable. Icelandic and other North Germanic languages demand all stressed syllables be bimoraic, which means their rime either consists of a short vowel and a cononant, or a long vowel. “Superheavy” syllables with more than two morae are not tolerated.

Word initially and following long vowels, an aspirate sequence ph, th, kh is analysed as either an acceptable onset, as straddling the syllable boundary in an acceptable way (e.g. as p.h, t.h etc), or with the h standing as an “extrasyllabic” .

Pre-aspiration is also “blocked” before voiced sonorants, with differing dialectical outcomes. Some northern dialects retain the original aspiration in such a sequence (e.g. m+ph = mph, ð+kh = ðkh), while in the south, h is lost but the preceding sonorant devoices (m̥p, θk). This reflects differing phonotactic preferences between dialects. Northern dialects syllabify m+ph as m.ph, Southern dialects, in which pre-aspiration is much more generalised, attempt to resolve the cluster by metathesising ph to hp, at which point it merges with the sonorant.

Other North Germanic languages and dialects exhibiting pre-aspiration do so in similar ways, although sometimes with simpler underlying rules. (In Faroese, for instance, pre-aspirates occur only intervocalically, and cannot occur before a close vowel).

All of this serves to demonstrate that Kortland’s recruitment of NG pre-aspirates as direct reflexes of supposed PIE “pre-glottals” or implosives effectively amounts to linguistic fanfiction, as does his attempt to connect it to the otherwise unrelated Danish vestjysk stød.

In brief: recent analyses of the stød suggest it is a definite innovation, as its distribution has been influenced by sound changes occuring since the 12th century AD. At least one article has appeared directly correcting Kortland’s attempted reconstruction here, and it does so convincingly, so I’m not going to pay it any more attention.

Beyond this, Kortland recruits his glottals to account for multiple discrete processes of stop gemination in multiple Germanic subfamilies. First, he wants *mp, *nt, *nk > pp, tt, kk in Scandinavia to reflect the devoicing of the nasals before assimilation to the stop via a preglottal element.

This is clearly inspired by the example of the Southern Icelandic dialects mentioned above, but as we’ve seen, this change isn’t even uniform in Icelandic and is explainable by other means. All we need for nasal+stop combinations to resolve as geminate stops is for nasals to drop before them and compensatory lengthening to take its course, or for the nasals to be strengthened by the following stops.

We should also consider (a) Kluge’s law, which states that phonemic contrasts in PG stops were levelled in the environment of a nasal anyway before the application of Grimm’s law and (b) that Kortland’s treatment of Winter’s law (below) also demands that preglottalisation be lost in the environment of nasals rather than altering the nasal).

He also wants the same thing to explain the gemination of *k and *t before *j, and the gemination of all West Germanic consonants less *r before *j as well. These can be dismissed out of hand. Both of changes already have isolatable environments that absolutely do not require phonological exoticisms to explain them. Finally, his suggestion that laryngeals created a phonemic glottal stop in Old English is perhaps the worst of the bunch here: all of the etymologies he proposes have better alternatives.

That’s quite enough of that, let’s get back to the freaks.

Italic: fricativisation

In Italic, murmured consonants were first reduced to voiced fricatives *β, *ð, *ɣ, *ɣʷ, which were then devoiced initially to *ɸ, *θ, *x, *xʷ . It’s often said that they were allophonically voiced in medial positions, but the opposite is much more likely because voiced fricatives are the most likely outcome for the spirantalisation of a murmured stop.

Proto-Italic quickly merged *ɸ, *θ and *xʷ as *ɸ~*f, while *x became *h except before a liquid, where it strengthened to *g. All the voiced allophones remained separate, except *ɣ intervocalically > ɦ, (written “h”). These eventually strengthened to voiced stops in Latin, but fell together as *β~v (written “f”) in Osco-Umbrian by a similar merger as occurred in the voiceless allophones.

(Note: the Italic resolution of the murmured stops doesn’t necessarily preclude glottalic theory, but all other “odd” developments necessarily do).

It’s now time to talk about item 7 on Kortland’s list: Lachmann’s law and Winter’s Law.

Lachmann’s law: not quite as simple as advertised

Lachmann’s law (short vowels are lengthened when followed by a cluster of a voiced stop and another stop, which is always voiceless in practice) is fairly easy to address: Glottalic theory posits that Lachmann’s law operates before voiced stops in PIE and not voiced aspirates because “glottals, dear boy”.

This once again fails to take into account the fact that “voiced aspirates” are not actually “voiced”.

In any case: voiced stops appear to have allophonically devoiced when immediately followed by a voiceless stop in PIE:

*h₂eǵtos > *h₂eḱtos (whence Latin āctus)

There’s no reason to assume that a murmured stop would similarly “demurmur”. The evidence bears out the fact that they don’t. If anything, murmured stops are more likely to assimilate everything in close proximity to themselves, perhaps accounting for why they don’t occur in roots alongside voiceless stops. The only apparent exception to this is when they are followed by *n, as seen in Winter’s Law below.

But how to we explain this apparently unique resolution in Italic? Jasanoff’s treatment provides a nice defence of the idea that voiced consonants were restored by analogy in transparent formations, after which the vowel was lengthened and a subsequent change devoiced the stops again. Basically:

*h₂eǵtos > *h₂eḱtos > *aktus > *agtus > *āgtus > *āctus

He also accounts for some other Latin irregularities like pēs and māximus as arising by a similar process.

However, I think we can simplify this. With two exceptions, tāctus < tangō and tēctus < tegō, all examples of words displaying Lachmann’s Law have a coronal voiced stop: they’re either d or ǵ.

It’s quite possible that the resulting Pre- or Proto-Italic *-dt- *-ǵt- > *-tt- *-ḱt- were dealt with by the deployment of normal PIE methods for resolving such clusters:

*h₂eǵtos > *h₂eḱtos > *assos > *āsos (paralleling the usual Latin treatment of thorn clusters)

*kh₂dtos > *kh₂ttos > *kh₂tstos > *kastos > *kassos > *cassus > *cāsus

In these cases (and in all other cases where Lachmann’s Law operates), the resulting participles would show a long vowel, but also be conspicuous in their missing root-final consonant. It would be very easy to restore it with reference to the perfect and present forms of the verb.

Our two exceptions argue against this, but we’re quite fortunate in that they both have two near-identical counterparts which do show Lachmann’s law: pāctus < pangō and frāctus < frangō for tāctus, and lēctus < legō and rēctus < regō for tēctus.

It doesn’t seem out of the realms of possibility that the long vowels in these two participles were introduced by analogy to their neighbours.

My proposed amendment also helps us explain the apparent anomalies axis and tussis, which are recruited by Kortland as an example of the “early loss of glottalisation before *s” (which, again, is a bit of a reach). Axis by rights should be āxis, but we instead see an alternative older form assis, matching my hypothetical preform *assos for eventual āctus.

It’s possible that the double -ss- here survived in early Italic before *i while it was otherwise deleted (this effectively mirrors Jasanoff’s proposal for the non-operation of Lachmann’s Law in stems with an i vowel: short i resists lengthening, and also blocks the shortening of the adjacent geminate).

In this case, a -k- was restored to assis rather than āsis; the vowel was never lengthened, so it can’t be lengthened in the analogically restored form. The same applies for tussis, where any attempt to restore -d- would be reduced back to -s- next to -s-, which is then retained as a geminate. The sigmatic aorist forms of adigō (with a stem adāx-) can also be explained as a normal operation of this law applied to an inflected form: -g- or -k- is restored to a reduced stem adās- < *adass- to the other parts of the verb.

Pushing the antecedent for Lachmann’s Law back to the dialectical PIE level also raises the possibility that the reduction of *-dt- to *-ss- sometimes reconstructed for Proto-Celtic, is part of the same shift. This shift is known to be early (with some apparent exceptions in certain dialects of Welsh), being already found in 1st Century BC glosses of Celtic terms given by Caaesar. That the PC examples don’t show lengthening of the vowel would be down to the fact that Celtic was able to retain -ss- as a geminate in all positions, whereas Italic, much less tolerant of geminates, reduced this sound to a single *s early on aside from the examples mentioned above.

This means we have two layers of analogy operating on a marginal case rather than one. But it seems to me a more satisfying answer that also allows us to incorporate Jasonoff’s argument for māximus < *mh₂ǵ(i)sm̥mos via *maḱsomos > *massomos > *māsmus > *māgsimus (with analogical restoration of g > k via reference to magnus and *magiōs (> maior) and pēs via a similar route. The i is apparently restored here as well by analogy to other Latin superlatives, despite this form having only arisen in the first place because of its reduction. Funnny old world, innit?

In any case, whether you like my modification to Jasonoff’s modification to Lachmann’s law or not, there’s absolutely no need for glottal elements on PIE voiced stops to get us there.

Winter’s Law

Winter’s Law a significantly more complicated case. The summary I’m going to provide here doesn’t really do the extent of the debate over the thing justice. I apologise for any Balto-Slavic fans in my audience.

In its original formulation, WL describes a similar concept to Lachmann’s Law—a short vowel before a voiced stop in Proto-Balto-Slavic is lengthened.

This again must date back to the dialectical PIE stage, as it discriminates between “voiced” and “aspirated” stops which have otherwise merged in Balto-Slavic, and before the general merger of short *o with *a, because it sometimes creates long *ō where we’d otherwise expect short *a.

This change has always been controversial: although some reflexes of some short vowels before voiced stops do seem to show lengthening, many don’t, and on the surface there appear to be a number of exceptions.

Various reformulations of WL have been offered. The most successful in terms of its explanatory power is Dybo’s, which has also been adopted by Leiden. In this formula, short vowels before voiced stops were lengthened as a result of their receiving the Proto-Balto-Slavic acute.

The acute is a unique prosodic feature of PBS, the existence of which is the only way to explain vowel length, accent placement, and tone outcomes in attested languages.

It is independent of the position of the “original” PBS accent (which is a minefield in and of itself), and can appear only on:

• Any syllable with a long vowel, regardless of coda

• Any syllable with a vocalic diphthong (limited to *ei, *au, *ai in PBS)

• Any syllable with a short vowel-sonorant *l, *r, *m, *n nucleus (these behave in the same way as diphthongs for most purposes)

This means that all PBS syllables could be either short, long, or long+acute. Since the acute can’t occur in a short syllable, later shortening of vowels for any reason cause the loss of the acute. This is most visible in consonant stem nouns and verbs, where inflectional endings that start with a vowel have caused the shortening of vowels (and deletion of acutes) in the stem.

The origin and nature of the acute is murky. Traditionally, it was believed that all long vowels in PBS became acuted at some stage, with non-acuted long vowels appearing later on. But more recent interpretations distinguish between inherited long vowels/diphthongs in PIE and long vowels/diphthongs arising from the loss of a laryngeal.

Per Kim, acutes rise from a PIE combination *VH(C) > *V̄ˀ(C), and from *VRH(C) > VRˀ(C), where R can stand for any of *r, l, m, n, y, or w as PBS phonology permits them to form diphthongs with a short vowel.

The acute above is reflected with a glottalisation marker because, you guessed it, the Leiden model casts the acute as a glottal residue of the PIE laryngeals.

Again, there is no real reason to assume that this is the case, other than that you’ve already decided that laryngeals are “laryngeal” because Jerzy Kuryłowicz called them that for want of a better term a hundred years ago, and you are not inclined to think for yourself.

Here’s what Kortland has to say about the acute and Winter’s Law (citations etc omitted):

Winter’s law yielded glottalization of a preceding syllabic nucleus, not lengthening of a preceding vowel. The glottalization merged with the glottalic reflex of the Indo-European laryngeals and remained distinct from vocalic length in Balto-Slavic. At a later stage, glottalization could yield short or long vowels in the separate languages, e.g. short o in Polish krowa ‘cow’ but long *ō in the Upper Sorbian cognate kruwa < krówa, similarly Polish słodki ‘sweet’ but Upper Sorbian słódki with an acute from Winter’s law. Glottalization was preserved in Russian at the time of the earliest Latvian borrowings. It has been preserved up to the present day in conservative varieties of Latvian, e.g. pȩ̂ds ‘footstep’, nuôgs ‘naked’, as in British English foot and naked.

The thing is though

The hiatus that can represent a glottal closure in some dialects of English on words like foot and naked is a recent allophonic development due to vowel reduction (in some cases these sounds are realised as aspirated, in dialects where the long vowel or diphthong remains, they are unmodified).

That aside, we’ve got some real problems here:

1. We don’t need the acute to be glottal for it to manifest in the way it does. At the level of elementary phonetics, vowels are often pronounced slightly longer before voiced stops than they are before voiceless ones (the vibration of the vocal cords associated with both voicing and the vowel “bleed over” into one another as the result of motor planning). We could just as easily assume that a consonant of sufficient sonority can prompt the lengthening of the vowel which manifests the acute in the same that vowels lengthened by “process” rather than inheritance do.

2. It doesn’t appear that the acute could ever manifest in a short syllable. This is proven by its erasure in cases where a vowel bearing it was shortened. Regardless of whether or not it was due to “glottalisation”, it would have to have resulted in a long syllable in PBS. We therefore couldn’t explain short or long reflexes in the descendants with reference to the acute (or WL) itself, we’d have to look for processes that might have blocked or reversed vowel lengthening and “suppressed” the acute in the process. (We’ve already mentioned one above).

3. As we’ve shown amply above, just because Russian suggests glottalisation at the time of the earliest Latvian borrowings doesn’t imply that glottalisation is a retention in the slightest: it could be a secondary development due to vowel reduction as in the proffered English examples. The same argument can be applied to the Samogitian and Latvian “broken tone”, which manifests as a creaking of the vocal cords in the middle of a long vowel or diphthong, and is often explained as a reflex of the PBS acute proper.

Kortland (after Dybo) further adduces some “blocking environments” for Winter’s Law. These include:

When *s precedes a voiced stop

This one is fairly straightforward, I don’t have much of an argument with it, and it seems to hold true in all cases. The blocking of whatever “acuting” process is going on in proximity to a voiced consonant by an intervening consonant seems trivial as a process.

When *s follows a voiced stop

This is presumed to account for PS *loza “vine, stick”, next to Lithuanian lazdà (also “stick” but also “hazel”) and Old Prussian laxde alongside PS *lēska (both “hazel”), all from PIE *h₁loǵ- presumably with an original suffix -teh₂ (?) which must have manifested in PBS as *lagzdā < *lagstā at some point (the g is apparently retained in dialectical Latvian lagzda).

The manifestation of WL in *lēskà can be blamed on this originating from an e-grade *h₁leǵ-keh₂, which presumably resolved to *lēzkà with the palatal devoicing thereafter.

In the other cases, it becomes difficult to explain the short vowel without reference to the s or z. Dybo in his examination of WL has curated multiple examples.

It’s unclear why a following sibilant would block the shift. Kortland of course leaps onto the scene to explain this as a consequence of the early loss of glottalisation before *s, and recruits axis and tussis as examples of this same principle from Lachmann’s Law. However, as I showed above, this is much more likely down to the marginal retention of geminate *-ss- resulting from a voiced coronal + any voiceless coronal in proximity to *i.

What’s more likely happening here is that in examples with following *-s- like *lagzdā, a similar assimilation occurred between the expected palatal outcome of *ǵ and the affricative element in *st. In some cases, this leaves us with a voiced palatal and no evidence of the acute, in others, it leaves us with a voiced palatal+stop combo and no evidence of the acute.

This is interesting because it suggests Winter’s Law is actually the complete opposite of Lachmann’s Law. As it turns out, Lachmann’s long vowels have nothing to do with voiced stops as such, but are the result of allophonic devoicing, assimilation, and reduction of voiced stops next to *s and *t.

In Winter’s Law, on the other hand, the lengthening (or “acuting”) of the vowel is actually triggered by the voicing of the following stop, and doesn’t occur when a following *s neutralises that voicing.

In cases with an unstressed nasal diphthong

Diphthongs of a short vowel + *n seem to regularly escape Winter’s Law. Kortland advances this as a loss of glottalisation before *-n-. But other nasal diphthongs do in fact appear to undergo Winter’s Law, as do resonant diphthongs with *r and *l (i.e. they receive the acute when followed by a voiced stop).

It appears instead that what’s happening here is the result of a paradigmatic levelling associated with “strong suffixes”. Basically, the nasal is lost early on and takes the acute with it, because removing the second element of the diphthong shortens the syllable and the acute can’t exist in short syllables.

Syllabic resonants on the whole also appear to undergo Winter’s Law, including *n̥ in many examples, suggesting that two different processes of accentological analogy operating at slightly different times. Interestingly, *un appears to lengthen to *ūˀn (with acute) when followed by any stop, while *in never does under the same conditions, giving us another acute that can’t be blamed on either glottalisation or laryngeals).

When followed by a resonant

This is the most difficult exception to reconcile because it comes with a few exceptions itself.

However, the bulk of them are cases where a murmured stop has been allophonically “voiced” by a following *n—which to the perceptive should constitute a gigantic black mark against the idea of preglottalisation of voiced consonants. (How can a following n provoke glottalisation before the preceding consonant of all things?)

It appears which might be going on here—some of the time—is that a resonant following a voiced stop devoices it, hence no WL, and an n specifically following a murmured stop voices it (perhaps by forcing a more aggressive, “open” vibration of the vocal cords). Again, this is fairly natural to assume and we don’t need preglottalisation to explain it.

Overwintering

The distribution of Winter’s Law and its exceptions along with basic considerations of PBS phonology leads us to conclude:

1. It was caused specifically by the proximity of a vowel sound to a voiced stop consonant, not by any other secondary process or articulatory factor.

2. Other phonological processes that separated the vowel from the triggering voiced stop (whether by devoicing the stop or standing between them) block the change.

3. The change itself was the lengthening of the underlying vowel sound, at which time it also became “acuted”. This is demonstrated by four things:

   1. The acute can only occur in syllables with long vowels or diphthongs.

   2. Any shortening of acuted syllables deletes the acute.

   3. Diphthongs in positions that trigger Winter’s Law also receive an acute.

   4. Acute and “non acute” (or “circumflex”) long vowels and diphthongs have some quality that sets them apart from each other, as they behave differently in subsequent sound changes.

Based on both the Balto-Slavic evidence itself and the evidence or lack thereof for “glottalic” influences here, there’s no reason to specifically connect the acute produced by Winter’s Law with the acute produced by laryngeals, except for the fact that they are both associated with the prosodic lengthening of previously short vowels.

As usual, I tend to go for the simplest plausible explanation, which is this: the acute isn’t a distinct phonemic segment that descends from the PIE laryngeals, or indeed any other sound.

Instead, it’s a property specific to PBS that’s introduced whenever a syllable acquires an extra mora, whether by the means of the loss of another consonant (the principle source of long vowels in “late” PIE), conditioned changes triggered by proximity, or as the result of any other process.

In other words, the acute is an example of chroneme: a suprasegmental element that signifies a specific (in this case conditioned) contrastive mark of length.

This is the only explanation that accounts for its contrast with “retained” PIE long vowels, seen especially in:

1. Absolute word final position (which were never acuted).

2. The shortening of word-final acuted long vowels in contrast to non-acuted final long vowels in Lithuanian (Leskien’s law).

3. The “sporadic” acuting of vowels that could never have been influenced by laryngeal deletion (including in Winter’s Law).

4. The apparent tendency for vowels lengthened as a result of the inherited PIE system of ablaut to have the acute applied to them as well.

It seems that Proto-Balto-Slavic speakers had some understanding of a distinction between a vowel which was “inherently long” and one which was “lengthened”, which is manifested in the acute.

We don’t actually need to know what the acute is in order to arrive at this conclusion. Although we’d still have to reconcile this with its presence in diphthongs, and its absence in cases where a laryngeal has been lost between vowels.

One possible way to explain this (as suggested by *ūˀn < *un before a stop) is by assuming that simple diphthongs were considered to be monomoraic in PBS—they were viewed by speakers as qualitative rather than quantitative differences in the articulation of a short vowel sound. They could therefore undergo moraic lengthening in the same manner as any other short vowel.

The lack of acuting on long vowels created by the deletion of an intervocalic laryngeal also support this view. The sequence VHV resolves as either an unacuted long vowel (where both original vowels were the same) or an unacuted diphthong (where they weren’t), except when the first vowel is *u or *i, in which case the laryngeal becomes either *w or *y.

This suggests that, as in Avestan, an intervocalic laryngeal was preserved not as a glottal stop, but as a weakly-articulated non-phonemic glide.

Since PBS and PIE don’t appear to have permitted “rising” diphthongs, this glide strengthens into a semivowel after a high vowel, but otherwise disappears leaving two short vowels in hiatus. These then resolve as what we can understand as an inherently bimoraic segment (which is thus not acuted, because it isn’t an “extension” of an otherwise monomoraic segment),

This also raises the possibility that “inherently long” vowels in PBS actuall became trimoraic (as in Germanic overlong final vowels descended from PIE final long vowels) in order to distinguish themselves from bimoraic segments created by “lengthening” processes.

In any case

All of this counts as evidence against the trigger of acuting in Winter’s Law as related to a “glottalic reflex” of either the layrngeals or the voiced stops.

We should instead see Winter’s Law as an example of conditioned lengthening triggered by voicing itself, as confirmed by its absence in “devoicing” environments.

Further, there’s a purely phonological argument against many of Kortland’s suppositions that can be made without reference to comparative evidence.

The reason I opened with mention of the laryngeals is that many of the arguments for glottalic theory are contingent on the conventional reconstruction of laryngeals as some kind of backed fricative. As seen in the Leiden conception of Winter’s law, this is presumed to have become a glottal stop or something adjacent to it in many of the descendants, often merging with the “glottal” element of “preglottalised” consonants.

This is highly implausible. The only real candidate for “preglottalised” stop sounds of this kind is implosives, as seen in Sindhi etc. Although implosives are described as being “glottalic ingressives”, their articulation is actually opposite to ejectives, which are truly “glottalised” sounds, and appear to be the model for Kortland’s conception of “preglottalised” consonants.

In ejectives, the glottis is lowered and fully closed (as in a glottal stop) before being raised sharply to create a “burst” of outward pressure. In implosives, the glottis is raised and then sharply lowered to create a “suction” effect. (This means the glottis never fully closes, and implosives are almost always realised as voiced, while ejectives are voiceless by definition).

Unless hypothetical PIE implosives were reanalysed as a bisegmental cluster of a glottal stop and a voiced stop—possible but unlikely—in many branches, a hypothetical “glottal” reflex of any laryngeals would not pattern as a “relative” of the implosives. As we’ve seen above, we’d also have to explain the realisation of laryngeals as some kind of weakly realised glide in PBS and PII and their purely vocalic realisation in Greek by other means. (The most likely explanation, as I briefly mentioned in the Lubotsky’s Law section, is that they were originally some kind of approximant allophone to specific vowels, but we’ll cover this another time).

Moreover, in terms of their phonetic realisation, implosives are fairly similar to voiced geminates, not to a glottal stop + voiced stop. If they did bisegmentalise, we would expect to see at least occasional medial reflections in the form of voiced geminates, which we don’t. In theory, I suppose, you could explain Winter’s Law and Lubotsky’s Law this way, but you’d have to do a lot of heavy lifting to explain why these residual geminates weren’t resolved by synchronic tendencies towards the deletion of geminates dating back to PIE times.

This means that all purported evidence of preglottalisation advanced for PIE is more correctly explained as the result of independent changes in the daughter languages. We can therefore regard the glottalic theory as disproven. (See, I can do it too).

Where does this leave us?

The main objection to the conventional reconstruction of PIE stop consonants is that it’s “synchronically unstable”, as mentioned above. But just because it’s synchronically unstable doesn’t mean it’s impossible. It merely means that it’s likely the result of some overarching transformation of the language that immediately preceded its breakup into different branches.

When we consider this possibility, we should consider relative distributions and easily observed phonotactic constraints.

Murmured consonants appear far more often than voiced consonants in PIE reconstructions. Voiced consonants are also unusually distributed: *b is very rare and usually secondary.

Furthermore, PIE is very finicky about which sounds it allows to appear together in roots and affixes. Two voiced stop can’t appear together in the same root, and murmured and unvoiced stop similarly can’t occur together either. But roots with two murmured stops are very common.

Alongside the specific distribution of “laryngeals”, “semivowels” and “resonants”, we can—and eventually will—make a case for the phenomenon of murmured phonation in PIE being a direct result of a general tendency towards vowel reduction in a variety of phonological and morphological processes. This tendency can be traced back to early PIE’s application of stress and ultimately provides an explanation for the origin for many of PIE’s more unusual morphophonemic qualities.

But this piece is already over 9000 words long, so we’ll leave that for another time.