Linguistic Miscellany Thread

[Qwynegold]

That sounds very fascinating. Has this been done before for Finnish?

I would have assumed that this has been done for every well studied language, but now that you said that, I don't actually know.

Here are my thoughts on measuring vowel length. There is phonetic final vowel lengthening in many languages. This can happen at the end of words, phrases or utterances. Since this is very variable, I would try to measure vowels somewhere else.

Oh, I see. I'll do this over then. Luckily not quite every word was disyllabic, so I can use some of my old recording.

Also, what is your reading pace? Are you trying to pronounce everything very clearly? A slow tempo often pumps up the vowel length a lot. Professional studys often use carrier sentences. The target word is then part of that sentence. It usually does not occur sentence-initially or sentence-finally. This means that you get less utterance final lengthening and you can control a bit for reading pace, e.g. you do not get a list intonation. It is important to note that the relative length of stressed and long vowels (in relation to short/unstressed vowels) is language specific. IIRC, some language have 1:2 ratios and others go as low as 5:6.

I try to use normal medium pace. I'm not using sentences, because so far I have 53 words in my list, and trying to come up with 53 sentences and recording them all is just too much. And after I'm done with these vowels, I want to investigate what allophones I have for /h/ and whether there any difference between diphthongs and vowels in hiatus that can be seen in a spectrogram.

Vowel length is easier to measure between plosives in my experience, because (modulo lenition) there are clear cut-off points. The "silence" in the signal can serve as a cue for the segment boundary. On the other hand, some language devoice vowels between voiceless plosives, which makes it harder to measure them, since the signal and the formants are weaker. That's what I can recall about measuring vowel length.

I'm using plosives as much as possible, but I find them a bit difficult, because there is a really long stretch where the previous vowel blends into the silent bit.

And here are my thoughts on measuring other stuff apart from length, in case you are planning to do this. In my experience, measuring formant value works in most environments, but these are easily influenced by place of articulation and laterals/nasals. This is less prominent towards the (temporal) middle portion of the vowel. That's what I did to minimize any influence of neighbouring consonants here. If you want to measure pitch (f0), it's easier to do this between sonorants. The reason is that you want a continous pitch measurement, e.g. in order to track down extreme pitch jumps that Praat measured, but that are not real and that can be corrected in the settings.

Aha, I had been measuring the average value of F₁ and F₂ across the vowel's whole duration. Should I just measure it at the midpoint of the vowel's duration instead? I probably will not measure pitch. Though later on I'm going to do Swedish, and maybe I'll investigate if I am managing the pitch accent at all.

[Creyeditor]

I'm not using sentences, because so far I have 53 words in my list, and trying to come up with 53 sentences and recording them all is just too much. And after I'm done with these vowels, I want to investigate what allophones I have for /h/ and whether there any difference between diphthongs and vowels in hiatus that can be seen in a spectrogram.

You could use the same sentence for all of them. It is often something like: "Now I am saying X twice."

And here are my thoughts on measuring other stuff apart from length, in case you are planning to do this. In my experience, measuring formant value works in most environments, but these are easily influenced by place of articulation and laterals/nasals. This is less prominent towards the (temporal) middle portion of the vowel. That's what I did to minimize any influence of neighbouring consonants here. If you want to measure pitch (f0), it's easier to do this between sonorants. The reason is that you want a continous pitch measurement, e.g. in order to track down extreme pitch jumps that Praat measured, but that are not real and that can be corrected in the settings.

Aha, I had been measuring the average value of F₁ and F₂ across the vowel's whole duration. Should I just measure it at the midpoint of the vowel's duration instead? [/quote]
I think this is better because the midpoint is less affected by the surrounding consonants. If you are planning on measuring diphthongs you could also do a measurement at the first quarter and the third quarter.

[Qwynegold]

I think this is better because the midpoint is less affected by the surrounding consonants. If you are planning on measuring diphthongs you could also do a measurement at the first quarter and the third quarter.

Thanks for the tips!

There's one more thing I want to ask about. In formant settings, you have to specify how many formants you want to be displayed. I have been using a value of 4, but occasionally I get a really weird value for a vowel, so then I have temporally set it to 5. Should I use 5 all the time? It's really confusing what settings one should use.

[Creyeditor]

I usually use the standard setting (which is 5, IIRC). If I see a strange formant that is really out of the range I adjust the number of formants to 3 and the Hz range to something I expect. This might not be scientifically valid without further justifications, but I think as a hobby it is certainly an okay way of doing it.

[Qwynegold]

*sigh* I had to remeasure a lot of stuff, but now I'm done measuring all these vowels. The next problem is how to draw diagrams so that I can see what the data shows. I tried to use the ones you can generate in OpenOffice, but none of those diagram types do quite what I want.

I need two kinds of diagrams. One is for the vowel lengths. It should have columns with different starting and end points, corresponding to the minimum and maximum vowel lengths measured, and a marking in the middle of the column, showing what the average value was.

The other is basically an IPA vowel chart. I have three instances of each type of vowel, so I want to draw triangles where the corners represents the F₁ and F₂ values of each instance. And then a dot in the middle of each triangle, showing the average value for that vowel type.

I'm thinking should I just draw on paper? That way I can get exactly what I want, but it's hard to work with such a fine grained scale by hand, with values between 200 and 3050. Or should I use Adobe Illustrator to create diagrams? These fine values are no problem when drawing vector art, but I only have Illustrator on my busted laptop which is a pain to work with. Or is there some online tool that can create the kind of diagrams I want?

[KathTheDragon]

I (in LibreOffice, which you should probably switch to from OpenOffice) used a simple scatter plot to make a chart of vowel quality, by plotting F1 on the y-axis, and F2 on the x-axis, with both scales reversed. And it works pretty well. As for vowel length, it sounds like you're looking for a box-and-whisker diagram, but LibreOffice doesn't seem to support them.

[bradrn]

*sigh* I had to remeasure a lot of stuff, but now I'm done measuring all these vowels. The next problem is how to draw diagrams so that I can see what the data shows. I tried to use the ones you can generate in OpenOffice, but none of those diagram types do quite what I want.

I need two kinds of diagrams. One is for the vowel lengths. It should have columns with different starting and end points, corresponding to the minimum and maximum vowel lengths measured, and a marking in the middle of the column, showing what the average value was.

That’s basically a Box plot (or a variant thereof); most statistical software will let you do that.

The other is basically an IPA vowel chart. I have three instances of each type of vowel, so I want to draw triangles where the corners represents the F₁ and F₂ values of each instance. And then a dot in the middle of each triangle, showing the average value for that vowel type.

I'm thinking should I just draw on paper? That way I can get exactly what I want, but it's hard to work with such a fine grained scale by hand, with values between 200 and 3050. Or should I use Adobe Illustrator to create diagrams? These fine values are no problem when drawing vector art, but I only have Illustrator on my busted laptop which is a pain to work with. Or is there some online tool that can create the kind of diagrams I want?

I’ve never used it myself, but I’ve heard that gnuplot is very flexible; you could use that. Alternately, you could just do a regular scatterplot, copy it into Photoshop, and then subimpose a vowel diagram under it.

[Creyeditor]

I'm thinking should I just draw on paper? That way I can get exactly what I want, but it's hard to work with such a fine grained scale by hand, with values between 200 and 3050. Or should I use Adobe Illustrator to create diagrams? These fine values are no problem when drawing vector art, but I only have Illustrator on my busted laptop which is a pain to work with. Or is there some online tool that can create the kind of diagrams I want?

I have to admit that I don't know the answer to these questions. In the past, I have either had someone help me draw diagramms with R(studio) or I drew it "by hand" in a graphics programm like GIMP, where you can also measure in pixels.

[bradrn]

I seem to remember reading a while ago about a language (Mesoamerican, I think) with exactly two adpositions. I’ve been trying to find it again, but to no avail. Does anyone have any idea which language that could be?

[Qwynegold]

Thanks for the suggestions. I realized I could just draw the vowel length diagram in Paint, so I did that. I'll make another thread for this.

I (in LibreOffice, which you should probably switch to from OpenOffice) used a simple scatter plot to make a chart of vowel quality, by plotting F1 on the y-axis, and F2 on the x-axis, with both scales reversed. And it works pretty well. As for vowel length, it sounds like you're looking for a box-and-whisker diagram, but LibreOffice doesn't seem to support them.

I can make a scatter plot in Open Office too, but I can't draw lines in it. Hmm, maybe I'll do a scatter plot anyway, just to get the right positions, and then Photoshop that.

[2+3 Clusivity]

I seem to remember reading a while ago about a language (Mesoamerican, I think) with exactly two adpositions. I’ve been trying to find it again, but to no avail. Does anyone have any idea which language that could be?

Not sure about in Mesoarmerica, but I recall Tok Pisin is claimed to only have two simple prepositions (bilong < belong and long < long) and that the rest are built off of them. See Wiki. Not sure how credible that is as it does not give a source.

In general, if a language has a lot of serial verb constructions and/or relational nouns, I could see getting away with a pretty slim adpositional inventory.

EDIT: Bislama seems similar. Again, same pair: blong < belong and long < long.

[zompist]

I know it's not quite the question asked, but no adpositions is quite possible— pre-contact Quechua is an example. (I'd have to review sources to see if it's borrowed any from Spanish.)

[bradrn]

I seem to remember reading a while ago about a language (Mesoamerican, I think) with exactly two adpositions. I’ve been trying to find it again, but to no avail. Does anyone have any idea which language that could be?

Not sure about in Mesoarmerica, but I recall Tok Pisin is claimed to only have two simple prepositions (bilong < belong and long < long) and that the rest are built off of them. See Wiki. Not sure how credible that is as it does not give a source.

…


EDIT: Bislama seems similar. Again, same pair: blong < belong and long < long.

Yep, I know about those — but I’m not particularly interested in creoles, which tend to have tiny grammars anyway.

In general, if a language has a lot of serial verb constructions and/or relational nouns, I could see getting away with a pretty slim adpositional inventory.

I know this as well; I’ve recently been reading a lot about SVCs, hence my interest in seeing how few adpositions are needed.

I know it's not quite the question asked, but no adpositions is quite possible— pre-contact Quechua is an example. (I'd have to review sources to see if it's borrowed any from Spanish.)

I am aware of this (the possibility is mentioned on WALS, which gives a surprisingly large percentage of languages where this happens). But it’s only really possible for languages with a large case inventory, in which case you’ve really just replaced adpositions with cases.

[Richard W]

Not sure about in Mesoarmerica, but I recall Tok Pisin is claimed to only have two simple prepositions (bilong < belong and long < long) and that the rest are built off of them. See Wiki. Not sure how credible that is as it does not give a source.

I saw the statement in a slim book that was described as a guide for Australian district officers or whatever. I don't recall any contradictions from the very little Tok Pisin I've read.

I am aware of this (the possibility is mentioned on WALS, which gives a surprisingly large percentage of languages where this happens). But it’s only really possible for languages with a large case inventory, in which case you’ve really just replaced adpositions with cases.

It may be a matter of hocus pocus versus God's truth, but Thai is alleged to have very few if any adpositions, and my claim for cases with surface manifestation (genitive and construct) is usually treated as a joke. The words which serve to translate adpositions are (also) verbs or nouns (like English 'top').

[Kuchigakatai]

I know it's not quite the question asked, but no adpositions is quite possible— pre-contact Quechua is an example. (I'd have to review sources to see if it's borrowed any from Spanish.)

Does it accomplish that using something other than its many cases? Looking at the example declension table on Wikipedia, I notice locations aren't listed (below X, close to X, next to X), nor time relations (since X, during X, in X [days, months]), nor "against X", nor "according to X". Out of curiosity, how are these handled, if you know?

In general, if a language has a lot of serial verb constructions and/or relational nouns, I could see getting away with a pretty slim adpositional inventory.

It may be a matter of hocus pocus versus God's truth, but Thai is alleged to have very few if any adpositions, and my claim for cases with surface manifestation (genitive and construct) is usually treated as a joke. The words which serve to translate adpositions are (also) verbs or nouns (like English 'top').

At the same time, it bothers me that people, at least in conlanging land, seem to insist on considering these things actual verbs or nouns rather than words that are in more than one word class, one of them being adpositions.

For example, Mandarin 向 xiàng 'towards; to face or turn [to a direction], to support sb' can't be modified by the aspect markers 了 -le (perfective), 過 -guo (experiential) or 在 zài (action continuative) when it means 'towards'. It can be by 著 -zhe (stative continuative), but then any verb can in the sense of 'while [doing the thing]'. The negators 不 bù and 沒 méi have to be placed before it, so that test doesn't work.

Why bother considering it a "verb" then, as opposed to a preposition? The only reason I can see why 向 xiàng and similar are called "coverbs" (note, not "verbs") in Mandarin grammar is because it's interesting they're homophonous with a verb. Under the same logic, we could refer to English prepositions "in" and "on" as "coadverbs" as they're the same as the stressed locational/directional (pseudo-)adverbs of phrasal verbs like "go in", "be in" ("The doctor is in"), and "put your shirt on", but that doesn't make the prepositional uses actual adverbs...

Mutatis mutandis for the likes of 上面 shàngmiàn 'on X; the top part (of a thing)'. Why bother considering that a locational noun, as opposed to a "postposition", or a "locational" (if you insist)? It can't be modified with a determiner.

(I'd like to add Mandarin also has words that everyone agrees, or should agree, are genuine prepositions and postpositions, like the preposition 於 yú which doesn't have a verbal use anymore, or the postposition 上 shàng 'on X; to go upwards [to a place], to go [to an activity]; part one of two' which doesn't have a use as a noun meaning 'top' anymore unless you want to sound very abbreviated/archaizing.)

This is not to say that in languages other than Mandarin such an analysis doesn't make sense. But I suspect my concerns are likely repeatable in many other such East and Southeast Asian languages...

[zompist]

I know it's not quite the question asked, but no adpositions is quite possible— pre-contact Quechua is an example. (I'd have to review sources to see if it's borrowed any from Spanish.)

Does it accomplish that using something other than its many cases? Looking at the example declension table on Wikipedia, I notice locations aren't listed (below X, close to X, next to X), nor time relations (since X, during X, in X [days, months]), nor "against X", nor "according to X". Out of curiosity, how are these handled, if you know?

It's cases, mostly. Certain constructions that would be prepositional in English (e.g. "for me") would be expressed by other non-case suffixes. Time expressions are handled using the locative cases.

[bradrn]

At the same time, it bothers me that people, at least in conlanging land, seem to insist on considering these things actual verbs or nouns rather than words that are in more than one word class, one of them being adpositions.

For example, Mandarin 向 xiàng 'towards; to face or turn [to a direction], to support sb' can't be modified by the aspect markers 了 -le (perfective), 過 -guo (experiential) or 在 zài (action continuative) when it means 'towards'. It can be by 著 -zhe (stative continuative), but then any verb can in the sense of 'while [doing the thing]'. The negators 不 bù and 沒 méi have to be placed before it, so that test doesn't work.

Why bother considering it a "verb" then, as opposed to a preposition? The only reason I can see why 向 xiàng and similar are called "coverbs" (note, not "verbs") in Mandarin grammar is because it's interesting they're homophonous with a verb. Under the same logic, we could refer to English prepositions "in" and "on" as "coadverbs" as they're the same as the stressed locational/directional (pseudo-)adverbs of phrasal verbs like "go in", "be in" ("The doctor is in"), and "put your shirt on", but that doesn't make the prepositional uses actual adverbs...

In Cantonese at least, there’s a reasonable argument that these are actual verbs rather than prepositions: any word in the first position of an asymmetric serial verb clause is forbidden from taking an aspect suffix, no matter whether it’s a ‘verb’ or a ‘preposition’.

[2+3 Clusivity]

At the same time, it bothers me that people, at least in conlanging land, seem to insist on considering these things actual verbs or nouns rather than words that are in more than one word class, one of them being adpositions.

Not quite sure I am getting your quip. I get it that entries can appear in multiple word classes in certain languages, but I don't get why conceiving (or describing) a given use of an entry as falling into one category or another is a problem.

Perhaps I am miss reading (into) your argument, but I see it as also problematic to think of a form that may appear as an adposition as being adpositional in all forms--take for example a form X that might appear in one sense as an adposition but perhaps elsewhere as a preverb adding a directional sense to a basic verb or perhaps as an applicative changing the syntax, etc. of the clause. Perhaps in some languages you can get a sense of one use as being more basic as others ... diachronically or otherwise ... but I imagine there is some variation.

[bradrn]

I just discovered that zompist got cited in a paper! Possibly zompist might be aware of this already, but Ratcliffe (2012) cites his article on chance resemblances in order to show that both Ehret’s (1995) and Orel & Stolbova’s (1995) reconstructions of Proto-Afroasiatic are very likely to be composed mostly of chance matches. I’m pretty impressed by that — I hear it’s hard enough to get cited even when you publish a paper, and (as far as I’m aware) zompist hasn’t published any!

[zompist]

I just discovered that zompist got cited in a paper!

I wasn't aware of that one, neat. That reminds me that I should put dates on my pages!