Mora and accent

Since I'm on about Latin phonology as of late, I might as well talk about accent and mora. I found a paper talking about it. Lehman says this:

P12.  Latin word accentuation
     1. The weight of the last syllable is stipulated to be one mora.
     2. Word accent falls on the third-last mora.
     3. If the word is shorter, word accent falls on the first mora.
While this does not answer the question of whether the mora is a primitive or a derived unit of Latin phonology…

But I don't like the notion of counting the last syllable as one mora for accentuation purposes. It screws up poetic scanning and junks up the whole rest of the system of mora counting. It also fails to account for words that have exceptional accent locations, like illūc.

Making a slight modification to Lehman's rules clears the whole mess up. I propose this:

1. From the last syllable onset position (whether filled or not), count back two moras.
2. Stress the syllable with the mora penultimate to the last onset.
3. If no penultimate mora, stress the mora before the last onset. 

Here's why I like this: No exceptions. Here it is in action. To make things clear, I've turned the ultimate syllable onset red as well as the moras that are counted.

	Standard orthography	IPA with syllables	IPA with moras indicated	Onsets and mora count	Stress placed
Antepenultimate stressed syllable	paenitet	paɪ.nɪ.tɛt	paɪμμ.nɪμ.tɛμt	paɪμμ.nɪμ.tɛμt	'paɪ.nɪ.tɛt
Penultimate stressed syllable	amāre	a.maː.rɛ	aμ.maːμμ.rɛμ	aμ.maːμμ.rɛμ	a.'maː.rɛ
Ultimate stressed syllable	illūc	ɪ.lːuːk	ɪμlμ.luːμμkμ	ɪμlμ.luːμμkμ	ɪ.'lːuːk

Presumably the first two rows look like standard action. The last row needs some explaining. First, the /lː/ is part of the second syllable, but it is also a participant in the previous syllable for the purpose of mora. I split it in the IPA with moras indicated column to make the bisyllabic participation explicit. So where do I get off on calling the final /k/ the onset of the ultimate syllable?

Well, for those of you paying attention to your Bennett's, check this:

6.3. When the enclitics… -ce… are appended to words…

And that is exactly what the situation with illūc is. The -c at the end is a remnant of the -ce enclitic. So the actual situation is that the word was originally illucce. Both the ll and the cc, as /lː/ and /kː/, would be ambisyllabic. Ok, so there's some sleight of hand going on by invoking the /k/ at the end of illūc as part of the onset of a non-pronounced syllable, but it kills the irregularity in Lehman's mora-based rule and syllable-based accentuation rules generally. (That said, now we've got a violation of *μμμ. Dammit, nothing works.)

Or am I missing something? I can't help but feel like I'm missing something when I'm putting forward a new idea.

*VVV? No, *μμμ

Well, I found what I was looking for. I was down the right trail with my suggestion that *VVV was a constraint in Latin. The problem is that I wasn't going far enough: I needed to take it all the way to the mora.

While searching, I found a research paper that focuses on six constraints. Brennan, the author, clearly has sunk more time in than I have. He went far enough to get to *μμμ (at least in syllables that aren't word initial). Makes sense to me. What's cool is that it fits in nicely with what Allen suggests for syllable weights.

canem ➝ [kanẽ:]

And that is ok. The [e] is one mora and the [:] is the other. But then

canēs ➝ [kane:s]

Wait. Isn't that last syllable e_μμs_μ? No. According to Brennan, /s/ and /n/ aren't moraic. So we've really got is e_μμs. And that's not violating *μμμ. So far so good. It gets better. Sort of.

*μμμ solves the [ju:li:] problem quite nicely. In the nominative we lose nothing, and the constraints explain it all.

/juːlɪʊs/	* μμμ	MaxIO
☞juːlɪʊs
juːlʊs		*

After all [ju_μμlɪ_μʊ_μs] never has more than three moras in a row—even if the /ɪ/ and /ʊ/ are in separate syllables. But in the genitive we violate * μμμ, though admittedly across the syllable line.

/juːlɪiː/	* μμμ	MaxIO
☞juːliː		*
juːlɪiː	*!

So /ju_μμlɪ_μi_μμ/ packs three moras into the last two syllables, which, while it is allowed, seems not to be favored. But when we move to the dative we get this.

/juːlɪoː/	* μμμ	MaxIO
☞juːlɪoː
juːloː		*

Three moras. No deletion. My suggestion is that /ɪ/ and /i:/ are somehow considered to be the same. And their orthography and presentation in textbooks would suggest that. The existence of stuff like nihil/nīl, pronounced [nɪhɪl̴] and [ni:l̴], also suggests that /ɪ/ and /i:/ are related quite closely.

The constraint of *μμμ  answers some questions, but brings me back to my initial question of just what is the nature of the relationship between long and short vowels in Latin? Something is afoot. 

Deleting the perfect v

In the last post I looked at syncopation in perfect verb forms. And then I remembered poor īre. It syncopates too, but it doesn't seem to have a theme vowel. So we get this.

īvī → iī
īvistī → īstī
īvit → iit

This reveals something about the v-deletion. First, *VVV is seemingly broken by iī. Second, if v deletes and takes ī to i, then something really odd, though by no means without parallel elsewhere, is going on with īstī. But then it doesn't happen with iit. Strange things are afoot at the Circle K.

But I got another reminder yesterday in class. v-deletion is even more widespread. Remember nōrit? That's right.

nōverit → nōrit

So now we've got -vi- and -ve- deleting. Now we've got some patterns to go hunting for. The question is this: can I find a -vī- or -vē- that deletes? If so, this is really simple. It's about front vowels. Will I find -vo- or -vu- deletion? Then it is short vowel deletion, but for now all I can say is -vi- and -ve- delete.

/wV+front+short/  → Ø / V_C

If only -vi- and -ve- deleting, then there are, as I said before, strange things afoot at the Circle K. Then it is confined to front short vowels. (And I really don't care for the terminology long/short since it also is a vowel difference.) What is is about v and these vowels that are prone to deletion? While something like nōrit isn't ambiguous, it isn't wholly clear either. 

You say Iuli, I say Iulii…

There is something to this *VVV constraint in Latin. If you remember, as I did this afternoon, there is an alternate genitive for 2nd declension -ius nouns, the constraint gets some more evidence for its existence. Exhibit 1: the most famous gens Romana: Iulius. But in the genitive it can be either Iulii or Iuli. Curious that. (And thank the dead language phonologist's friend, Latin grammarians, that surface forms got written down.)

What is interesting—to me, and since I write the blog, we look at what I find interesting—is that even though the /ɪiː/ at the end is two syllables, it is close enough to violating *VVV that the underlying /ɪ/ gets deleted.

/juːlɪiː/	*VVV	MaxIO
☞juːliː		*
juːlɪiː	*!

While both forms pick up one violation the *VVV violation is worse, so [juːliː] is favored. Now tell me that doesn't make sense and give some strong evidence for the existence of *VVV in Latin.

Constraints in Latin phonology

In phonology we're moving away from rules, which seem to be language-specific, and towards constraints. I've got my objections to constraint-based phonology, but it also seems elegant. Or at least clean.

So here's a rule in Latin. It's not a hard rule, but it seems to be frequent enough that it works. And good enough to get to modern Spanish. Maybe some more tinkering would make it solid, but then the number of rules multiplies. Anyway, the rule. Or something really close to it. There's probably a morpheme boundary involved, but let's ignore that for now, shall we?

/w+ɪ/ ➝ Ø / Vː__

It is at least productive enough to yield parāstī from parāvistī. So that's the constraint table I'll put up. I'm going to focus on just a few forms to keep things under control.

Clearly there is a *VVV constraint in Latin. There are long vowels and diphthongs, which are effectively VV. There are short vowels, which are V. But no VVV—extra long vowels or long vowel diphthongs. Ever. So that's an easy constraint. It even seems fatal in all situations—unless you can somehow kludge in hiatus. That would violate DepIO (the constraint that says not to add things in), so I feel pretty safe in ignoring that possibility here.

There also seems to be a constraint against eliminating the conjugation's theme vowel going on. I can't put my finger on any scientific reason you'd want to do this, but it seems like a way to avoid creating ambiguity. In any case, syncopated forms in Latin preserve the conjugation's theme vowel in the perfect. If there's no theme vowel in the perfect stem, there's no syncopation.

On to the tableau.

/pɑrɑːwɪstiː/	*VVV	*theme vowel deletion	MaxIO
☞pɑrɑːwɪstiː
pɑrɑːstiː			*
pɑrɑːɪstiː	*!		*
pɑrɪstiː		*!	*

An ugly problem turns up. The tableau predicts the wrong winner. We want to get [pɑrɑːstiː] as the clear winner so that it is the favored colloquial form to get passed to Spanish. The problem is that we need to violate MaxIO by deleting [wɪ] and still get the win. Meanwhile the full form [pɑrɑːwɪstiː] violates no constraints in the above tableau, but needs to lose. 

Or maybe no such problem existed in 1st century BCE Latin. You could use either form. The MaxIO violation must have been so weak that something like [pɑrɑːstiː] could start competing with the winner [pɑrɑːwɪstiː]. After all, [pɑrɑːstiː] doesn't take information out of the verb: in fact, it eliminates redundancy. The -istī ending is unique to the perfect tense, so dropping the perfective -v- couldn't create ambiguity. From there all you've got to do is push the first i in -istī over to the stem and get an underlying second person perfect morpheme of -stī. Then there's no good reason to keep the first i. (Sounds simple when I put it that way.) But that's beyond phonology, no?

EDIT: I think there's something to all of this. Some goodies about SR in Latin orthography that may provide a springboard for further investigation. Some evidence for *VVV.

Double EDIT: Nemo Oudeis tells me that the v-deletion can also occur in other places (divitior > ditior; divitissimus > ditissimus), but not remembering these for sure I understated the matter.

Note about the table: Support for tables in Blogger poor. Nay, non-existent.

Spectrograms

I just got done with a frustrating analysis of vowel duration in English. From what I can tell, the rules only count in theory. In measurable spectrograms, all bets are off.

This makes me want to make some spectrograms of Latin aphorisms to see if I'm putting the theory into practice. Praat much?

Say ah!

Little Boy and I were playing yesterday. He's become quite the chatter box, though he's only starting to grasp words. Because he loves to make lots of noise with his voice, he opens his mouth really wide. So wide that I saw his tongue move down and back before he started yelling /ɑ/ at me. Every time he went to yell, down and back went the tongue.

I mention this because I had been having a hard time picturing the front/back and high/low distinction when producing vowels. I can hear the distinctions, well, insofar as any native English speaker can. This is to say I can hear the distinctions English makes quite clearly. Others may be a bit more difficult, but I digress. It was good to see the actual mechanics of speech production put in front of me so accidentally.

It was nice to see that. 

Phonology of Mandan vowels

For those of you interested in the phonology of Mandan, here's a summary of a presentation I recently sat in on.

Jonina Torres presented information about the way Mandan, a Siouan language spoken in North Dakota, differentiates its vowels. She had analyzed a recording of a Mandan speaker to see how the vowels varied in duration. Mandan appears to differentiate vowels in both length and nasalization. The inventory Torres investigaged is /a, a:, ã, ã:, e, e:, i, i:, ĩ, ĩ:, u, u:, ũ, ũ:, o, o:/.

Using a piece of software called Praat, she looked at sonograms and blocked off each vowel for duration. For the most part, short vowels were held for about 0.1s and long vowels were held for 0.2s. This doesn't surprise me given the describers long and short. Torres found two pairs that did not fit the pattern. The first is the distinction between /ã/ and /ã:/. The short /ã/ was held for about 0.25s, and the long /ã:/ was held for about 0.15s. This is odd, since the “short” vowel was in fact longer than the “long” vowel. The other surprise in her data was the variance between /ũ/ and /ũ:/. Short /ũ/ was held for 0.1s, which is in line with the other short vowels. On the other hand long /ũ:/ was held for 0.5s, which was more than twice as long as all the other long vowels, save /ã:/. Concluding, Torres suspected that these variations might be evened out if she sampled a larger data set.

Hearing things in The Hunger Games

I went to go see The Hunger Games with my wife over the weekend. It was ok. Well, more accurately, it was subversive, but that's another post for another blog. There were a few interesting points.

Some of the characters had distinctly Roman names: Cinna, Cato and Caesar leap to mind. All of their names are pronounced English-style. /s/ for Cinna and Caesar. /k/ for Cato. I knew about the Roman names in advance, so I really was hoping for /k/ throughout. Oh well.

But then the really interesting part. After watching the movie, my wife referred to one of the characters, and I thought, "Hm. That's odd." In case you didn't know, my given name is Peter. She said /pitə/. Wait. Nice midwestern girls like my wife do not say /pitə/. Most midwesterners, my wife included, say /pitəɹ/, so I was surprised to hear /pitə/.

Since I've lived in New England, I've been called /pitə/. A lot. My coworkers were particularly fond of displaying their non-rhotic accents in this manner. Suffice to say when I heard the character was named /pitə/ in the movie, I heard Peter. After the movie my wife, without meaning to, set me straight.

Turns out the young man's name is Peeta. Our ears are liars.

John Wells

This guy has a blog about nothing but phonetics. It's wonderful. If you can tolerate to add another blog to your reading list, add this. To help convince you, here is his latest post about phonetics, regional and social variance. Wonderful stuff.

English places

Comments on yesterday’s blog addressed the fraught question of proper names in pronunciation dictionaries. I thought it might be useful if I tried to say what my policy was in LPD, at least as concerns place names in England.

I must confess that I did not set up a set of principles before starting work. Rather, what follows is a post-hoc attempt to express the principles I think I generally followed.

Let’s start from the difficult fact that in England everything is complicated by social class factors. A hundred years ago, certainly fifty years ago, and still to a large extent today, most English people spoke and speak with a local accent. Broadly speaking, the lower your social class, the more your pronunciation diverges from RP; the higher your social class, the closer to RP. Whereas RP speakers can be found in all parts of the country (or could when I were a lad, when not only the local landed gentry but also the vicar and the doctor probably spoke RP or something very close to it), “local” implies non-RP. The local accent typically includes various features that are regarded as non-standard and have traditionally been considered unworthy of mention in normative reference works such as dictionaries. (Note to nonNSs: when I were a lad is a stock phrase with non-standard werefor was, used for comic effect.)

So let’s agree, for the purposes of argument, that most people who live in Hull call it ʊl. But in RP it’s unquestionably hʌl. We can leave it to the sociolinguists to determine the precise details of who uses which of these pronunciations and under what circumstances, and to what extent there are also intermediate forms such as hʊl, həl and perhaps also ʌl.

Continue reading at John Wells's blog…

Allen's Vox Latina

Dude, IPA: 1888. Allen's Vox Latina: 1965. Why didn't he use IPA for everything?

Seriously, he's killing me. For the consonants Allen mostly used IPA. But then just to keep everyone on their toes, he makes a new system for vowels. Srsly? I can forgive the horrible triangle of vowels thing, mostly. The lack of IPA in the vowels is just inexcusable—it makes an otherwise clear book much more difficult to understand. I sometimes go nuts at classicists. Too many won't get on board with the rest of the world, be it Allen with IPA or shamefully too many teachers with second-language acquisition. <rant/>.

As for the book itself, it's an interesting read that is squarely aimed at folks that want the information without all of the nasty technical details. But I've got a bone to pick: how is he seriously saying that "gn" is pronounced, as near as I can tell /ŋn/. I don't know about you, but that's a tricky combo for me to make in flowing speech. Allen says that we should pronounce gn like "hang nail", but that is a low frequency word in English. "Magnus" is a bit more, shall we say, high frequency (the "gn" digraph occurs 93 times in Caesar's de Bello Gallico, Book 1). Maybe if I get motivated, I'll look into the pronunciation of "gn" thing.

Anyway. The whole thing seems like an invitation over to /ɲ/ to me. Or at least this is how I got to /ɲ/ in my ordinary pronunciation of Latin, despite aiming for /ŋn/.

IPA for Latin

We got introduced to the IPA in class this week. It was done in the context of English, which I understand. Everyone in the room speaks English and thus has a common point of reference—well, and really, would you want to learn IPA in the context of Quenya or something dangerously unfamiliar? So English sweeps in to save the day. But I find the exclusive study of English dull. So let's spice things up with Latin. Hopefully there aren't too many mistakes.

We have consonants.

	Labial	Labiodental	Alveolar	Palatal	Velar	Glottal
Stop	p   b		t   d		k   g
Aspirated stop	ph		th		kh
Fricative		f				h
Nasal	m		n	ɲ	ŋ
Trill			r
Flap			ɾ
Approximant	w			j	w
Lateral Approximant			l

Most of these are what you'd think they are. Latin writes c where IPA uses /k/. Latin writes n, but does both /n/ and /ŋ/ depending on where it is. So regina gets /n/ and angulus gets /ŋ/. R is kind of tricky too. I've heard some people make the r/rr distinction be /r/ and /rː/, but I've also heard—and use personally—/ɾ/ and /r/ for that distinction.

Then we have ASCII art hiding out as vowels.

  i+y--------------+---+u

    \              |   |

     \   ɪ         |   |

      \            |   |

      e+-----------+---+o

        \          |   |

         \         |   |

          \        |   |

           +-------+---+

            \      |   |

             \     |   |

              \    |   |

               +---+--a+

As you can see, not only is this really ugly, but Latin doesn't have much in the way of vowels. The sad thing is that /y/ is really only found in Greek loan words. There is also the matter of /ɪ/. I grayed it out, as I don't think (not having read Allen's Vox Latina, which you can buy for me if you love me) the Romans used it. I can tell you that /ɪ/ is taught for short i in the United States. I can't vouch for elsewhere. 

Each of the vowels is either long or short, as Latin vowels come in both flavors. Hic and hīc form a minimal pair that shows off vowel length.

hic (this) = /hic/ or /hɪc/, the former is probably better
hīc (here) = /hiːc/

Yes, I picked i. It shows off the /i/ versus /ɪ/ thing that we teach. Lastly there are nasalized vowels. I really wish these were more commonly taught. When vowel + m is the last part in a word, the vowel nasalizes.

am, as in tam = /tã/
em, as in hem = /hẽ/
im, as in turrim = /turĩ/
om as in the archaic quom collapsed into um
um as in virum = /wiɾũ/

There are other places that nasalization occurs, but this should be enough to get the idea. I haven't listed the diphthongs yet, so here you go:

ae = /ai/au = /au/ei = /ei/eu = /eu/oe = /ɔi/ or /oi/, but the latter is probably better.ui = /wi/

So now let's have some IPA fun at Catullus's expense. Here's Catullus 85:

odi et amo. quare id faciam, fortasse requiris
nescio, sed fieri sentio et excrucior.

And in IPA:

/oːd et amoː. kwaːɾ id faciã foɾtasːe rekwiːɾis/
/nescio sed fieɾiː senti et ekskɾukioɾ/

The elisions were kind of tricky to show, so I just dropped 'em. I probably should've done something ever so slightly different, but my transcription somewhat preserves the word divisions as is. Shall we do this for Ancient Greek sometime?

For fun, here it is in Latin.