The solution, of course, is to use TIPA, which is an excellent system for typesetting IPA symbols (and in fact, the only reliable way I’ve found to ensure that diacritics are placed in the right places). Unfortunately, actually typing in TIPA is a terrible endeavor, as many of the symbols are represented with several characters, or with longer escape sequences (“Columbia” is [k@" l\textturnv mbi@] in TIPA), and although it makes sense at some level, it’s far from intuitive.

So, fed up with manually looking up TIPA symbols, I stumbled upon the magnificently wonderful E-linguistics Toolkit, and from that, have a solution which makes using TIPA tolerable, a simple terminal command which, when run, converts text from unicode IPA into TIPA markup, allowing me to type quickly and still use my IPA in LaTeX.

Here’s how to set it up on your Mac (or linux box), assuming you’ve got a bit of command-line knowledge:

1) Download and install the eltk from their website. You’ll download the zip file, then follow the directions in the readme.

2) Download this file, my tipafy script and unzip it. I’ll assume it’s on your desktop

3) Run these two commands in your terminal:

sudo cp ~/Desktop/tipafy /usr/bin/tipafy
sudo chmod +x /usr/bin/tipafy

Then, you’re done! (Sorry, Windows people, I don’t really know how python and creating/running executables works for you :()

Now, let’s say you’ve got a chunk of unicode IPA, like “fownɛˈtɪʃn”, and you need it in TIPA form. Just open a terminal window and type the below:

tipafy fownɛˈtɪʃn

… and it’ll output:

\textipa{fownE" tISn}

Which is ready to be dropped into a LaTeX document.

If you have multiple words, just put the whole thing in quotes:

tipafy "ðə 'sowldʒə˞ də'sajdɪd"

… and it’ll output everything as a TIPA command.

It’s not perfect, but it gets you a heck of a lot closer than just typing blindly, and in my experience, it’s been faster to type unicode and convert than to actually try and remember the TIPA commands for everything. And, of course, the real credit goes to Scott Farrar and the eltk people, who made it possible for this whole thing to work.

\textipa{EndZoj}


I sent this video to my undergraduate phonetics class today, and one student emailed me back, asking “How accurate would that representation of the waveform be? What would it sound like if you played it back?”. Well, I have a conference call to prepare for, and I don’t want to be doing that right now, so I decided to go full nerd and do a bit of analysis instead and answer that question.

Let’s assume that the records were cut _extremely_ carefully, and that every record represents the amplitude at that moment perfectly. A stretch, but they seem to have done their homework. We’re also assuming (as seems to be the case) that the records are cut to a flat edge (rather than representing additional detail. Even with that assumption, does this represent the sound well? Let’s do the math.

The first record goes on at 27 seconds, the last at 1:52. That’s 85 seconds of audio. Those 85 seconds are represented by 960 vinyl records. This is a new record representing a slice of the waveform roughly every 0.088 seconds (the period). Using our frequency formula (f=1/0.066), that’s a record addition rate (sampling rate) of roughly 11 Hz. Bam.

So, we have a new amplitude sample 11 times per second. Now, we’re basically treating this as a digital audio file, which samples amplitude repeatedly at a given rate. The Nyquist theorem (which is awesome) states that the highest frequency captured accurately by a given sampling rate (the “Nyquist Frequency”) is 1/2 of the sampling rate.

For your average CD, your sampling rate is 44,100 Hz, which has a nyquist frequency of 22,050 Hz. This means that the highest frequency captured accurately in the recording is 22,050 Hz. Higher than humans can hear, but your pet dolphin isn’t impressed by your “Hi-Fi” stereo.

In this video, our sampling rate is only 11 Hz (11 records per second of audio). This means that the Nyquist frequency would be 5.5 Hz. Which means that if you were to “play back” the waveform here by pulsing the amplitude shown by the next record every 0.088 seconds, it’d be completely inaudible to humans (who can only hear 20Hz and above) as periodic sounds, and would just sound like a series of bursts of noise.

In effect, they’re representing a zoomed-out version of the waveform, which is incredibly awesome looking, but not a very good way to store audio information for playback. But that doesn’t take away from the fact that this is one of the more creative things done in a music video in a long time. So, I’d still chalk this one up as a victory for the producers of this video.

Alright, alright. I’ll go prepare for my conference call.

They all sit down around a big table and start doing field research with the speaker, asking grammar questions, trying to pick apart the phonology (sound patterns) of the language. After a while, I get sucked in, and the faculty member baits me into joining, hinting towards what I was already thinking was a phonemic initial glottal stop contrast. (For the less linguisty among you, that means that in this dream language, the sound in “Hawai’i” or “Uh-oh” can occur at the start of the word or not, and whether it’s there or not changes the meaning of the word).

So, at this point being unable to resist, I jump in. I quickly start trying to elicit the speaker to highlight the contrast by having him repeat words, partly for my own joy and partly to show the undergrads what’s going on. Then, as is always wise in a field methods class, I start trying to produce those contrasting words myself, something I’m quite comfortable doing having spent as much time in phonetics as I have.

Then, in my dream, I realize that I couldn’t. No matter how hard I try, I just couldn’t make that initial glottal stop, I just kept producing the words without it. I knew it was there, I knew how to make it, and I knew it SHOULD be working, but I couldn’t do it. And worse still, the speaker was getting frustrated, the faculty member was judging me, and the undergrads were all starting to mock me. Finally, scared, confused, and completely glottal-stop-less, I woke up.

I think I have a problem.

Having dreams about linguistics is nothing new to me. Heck, I’ve even analyzed dream languages for hours while sleeping. But this one, to my mind, crosses a line. I’ve heard that police officers sometimes can’t fire their guns to save themselves in their dreams, and maybe firefighters sometimes run out of water in their dreams.

Now, I know for sure I’m in the right field because apparently for me, in a nightmare, it’s not that I’ll be naked in class, that my gun won’t fire, or that my car won’t start. Instead, I’m up at night worried about laryngeal misfires. If that doesn’t make me a phonetician and a linguist, I don’t know what would.

For example, English has a phonological process called “Dental Assimilation”. In this process, an alveolar sound (like /t/, /n/ or /l/) becomes dentalized (made with the tongue behind the teeth) before dental sounds (/ð/ or /θ/). In order to demonstrate this (or better still, to test students), you need to come up with sets of words in which one word ends with an alveolar sound and the next starts with a dental sound.

Of course, there are a few common cases (“that thing”), but inevitably, in front of 66 people (or when asked to give an example on an exam), you can’t think of something reasonable like “can’t think”. So, you come up with something on the spot, and end up with something like “stupid thyroid” or “bell thief”.

To get these wonderful word pairs the recognition they deserve, I propose a neologism (a new word for an existing concept). I recommend that henceforth and forever more, a word combination which would be completely absurd in any context other than demonstrating phonology should be called a pine thug, in honor of the best/worst pair I’ve ever come up with in front of a classroom.

I know, I know, it’s tough to get a good neologism going. Most are gone within a few months (cf. “linsanity”), and barring political necessity (as has propelled “santorum“, a neologism created explicitly to mock Rick Santorum’s anti-gay stances), making a neologism stick is very difficult. However, I’ve known enough phonetics instructors (and students!) who acknowledge the agonies and ecstasies of pine thugs that maybe, just maybe, this one will take root.

To install, download the tgz file (the below code assumes you’ve downloaded it to the desktop), double click it to expand, then open a terminal and type:

cd ~/Desktop/sndpeek-1.3-exe

sudo cp bin/sndpeek /bin/

Once you hit enter after typing the above command, OS X will then ask for your OS X administrator password (to copy the file deep into the filesystem), and once you’ve done that, at any point in the future, you’ll be able to just type “sndpeek” into a terminal and it’ll pop up a window displaying whatever sound source is selected in your Sound input preference pane. I usually give a more complex command to produce a prettier output, “sndpeek –logfactor:0.5 –lissajous:OFF –features:OFF –depth:150′”

I encourage you to play with the software as there are few better tools to help understand what a spectral slice is, how it works, and how spectrograms can be made. Enjoy!

With a name like that, it couldn’t be anything but judgmental pedantry, but even in an otherwise eyeroll-worthy article, I found that several of these words are actually completely reasonable pronunciations, and several of them demonstrate interesting phonological processes. So, I’m going to discuss them a little bit.

Athlete (pronounced with a schwa in the middle, “Ath-uh-leet” /æθəlit/)

This is a very reasonable and common pronunciation, which I noticed extensively in the speech of even experts on the subject (Michael Lewis, the author of Moneyball: The Art of Winning an Unfair Game is a notable /æθəlit/ speaker. Here, the change likely comes from our dislike of having an interdental sound (/θ/) right next to a lateral (/l/). If you attempt to make the “correct” pronunciation, you’ll notice that your tongue is, in a sense, trapped between your front teeth, and to make a smooth gesture, you end up having to attempt to curve the sides of the middle and back of your tongue down. Which is unpleasant. So, it’s not shocking at all that speakers who use the word often may add the schwa.

(It’s also worth noting that there is no ‘H’ in Athlete, despite the author’s smug assertions that “there is no vowel between the ‘H’ and the ‘L’ in any of these words”. The English “TH” in this word is actually a single sound, a voiceless interdental fricative, which is nothing resembling an /h/. Once again, pedantry is seldom done well enough to be immune to further pedantry.)

Utmost (pronounced as “upmost”, /ʌpmowst/)

This is an awesome example of assimilation, two sounds becoming more like one another to make the speaker’s life easier, a phenomenon I’ve discussed before. Here, in the “correct” pronunciation, /ʌtmowst/, we have a /t/ sound, created at the alveolar ridge (just behind the teeth, try it) followed immediately by /m/, a bilabial sound created by pressing the two lips together.

When speakers are “mispronouncing” the word as /ʌpmowst/, they’re actually being more efficient, substituting in a /p/, also a bilabial sound, which allows them to simply close their lips (creating the /p/), then lower the velum (allowing nasal airflow) and start voicing to begin making the /m/. Going from /p/ to /m/ requires no additional tongue or lip movement, whereas going from /t/ to /m/ requires reconfiguration of the tongue and lips. Efficiency. Not quite the idiot pronunciation he’s claiming.

Sherbet (pronounced as “sher-bert”, /ʃɜɹbəɹt/)

Why does Primer Magazine hate assimilation? The first syllable has an “err” (/ɜɹ/) sound, why not the second syllable too? If we can keep the whole word vaguely “r-sounding” (“rhotic”, in phonetic terms), all the better. Speakers love regularity. Primer Magazine doesn’t.

“For all intensive Purposes”

This is really a horsed zebra. For further discussion of this, see a post I made last week.

Often (pronounced as “offen”, /ɑfɪn/)

How many Americans say “often” with the /t/, ever? This is textbook deletion of an unpleasant sound to simplify a cluster, and it’s one carried out by many, many people. Why bother with a /ft/ cluster when there’s no need to keep it around? It’s not like there’s another word, “Offen”, which this form of “often” could be confused with, and frankly, for speed, fluidity, and social reasons (in the US), the “offen” pronunciation is really a better choice.

Edit: OK, I misread this one completely in my anti-pedant rage. The author of the quoted article is actually _in favor_ of “offen” as the “proper” form, and I responded assuming that he, like so many others have, was arguing that “often” (with a /t/) is the only proper form. So, I’ve culled some of the anger from the post, and kept the phonology. Thanks, commenter!

Awry (pronounced as “aw-ree”, /’ɑɹi/ instead of “uh-rye” /ə’ɹaj/)

This word is a textbook example of why our writing system needs to be taken out behind the barn and dispatched as humanely as possible. Although “wry” is used for the proper /ɹaj/ pronunciation in the word “wry” (and only there), usually the “aw” digraph represents /ɑ/ (as in “claw”, “maw”, “awful”, “awkward”) and the “ry” represents /ɹi/ (as in “fury”, “worry”, “scurry”). I can understand the author feeling the need to state the proper pronunciation of the word, but his indignation at the thought that anybody could EVER think “awry” is pronounced “aw-ree” is just silly.

So, there’s a bit of phonological goodness wrung out of an otherwise dry and pedantic bit of prescriptivism. Which I am going to pronounce as “per-scriptivism” for the remainder of the day. Just to anger Justin Brown.

Purple haze all in my brain
Lately things just don’t seem the same
Actin’ funny, but I don’t know why
‘Scuse me while I kiss the sky

But many people hear the last line as “‘Scuse me while I kiss this guy”, and that misperception actually reveals something very interesting about how English consonants work.

What makes /k/ different from /g/?

Both /k/ and /g/ are what linguists refer to as “stops”, they’re consonants where the airstream out of the mouth is completely obstructed, and actually, both /k/ and /g/ are “velar” stops, made with the tongue up against the soft palate, or velum. Try it, making a /k/ as in “cap” and a /g/ as in “gap”, one after the other, and you’ll notice that your tongue isn’t changing position when you switch from /k/ to /g/ at all.

The simplistic explanation is that /k/ is a voiceless sound (meaning that our vocal folds/cords aren’t vibrating while we make the closure), and /g/ is a voiced sound, involving glottal vibration during the closure. Unfortunately, like most things in phonetics, it’s not quite that simple or easy.

Voice Onset Time

In reality, stop consonants are classified by their voice onset time, the amount of time that elapses between when the stop is released (when the tongue stops blocking airflow) and when the voicing starts (when the vocal folds start vibrating) for the following vowel. By looking at voice onset time (VOT), we can actually classify consonants in three different ways. (I’ve actually discussed voice onset time before, but now that I’ve already made nicer looking graphics for teaching, it seems worth doing again.)

First, [kʰ]. In English, any voiceless stop that’s at the start of a syllable (so the /k/ in “cap”, but not “pack”) is “aspirated”, meaning that there’s a considerable time gap with a burst of air between the opening of the stop and the start of voicing (it has a positive voice onset time). In the word “cap” /kæp/, we bring our tongue back to the velum to make a closure, we release that closure, and then, around 100 ms (milliseconds) later, we start voicing for the vowel /æ/. Viewed in terms of the acoustical waveform of speech, here’s what aspiration and VOT looks like in [kʰa]:

[g], on the other hand, is a voiced stop, where voicing actually starts during the closure. So, the tongue moves up to the velum, the vocal folds begin vibrating, and then, when the stop is released, the vowel begins immediately. The voice onset time is negative, as the voicing started before the closure. See yet another waveform diagram below, this time showing /ga/:

There’s a third option. Imagine that you started voicing at the exact moment that you released the stop, as shown below:

Then what you have is [k], what linguists refer to as a “voiceless unaspirated stop”, with a voice onset time of 0 (or close to it).

So, we have three stop choices: Voiced stops, voiceless unaspirated stops, and voiceless aspirated stops, which are all used differently in the different languages of the world. But how does this affect Jimi Hendrix?

English makes stops oddly

Our problems with Jimi Hendrix kissing guys (not that there’s anything wrong with that) come from three fundamental oddities in the way that English produces stops.

First, English only distinguishes between Aspirated and Voiced stops. “cap” starts with a /k/, which is produced with aspiration, and “gap” starts with /g/. We don’t have a three way contrast between voiced [g], voiceless unaspirated [k], and voiceless aspirated [kʰ]. Korean, as I’ve mentioned before, has that three way contrast.

Second, English word-initial (at the start of a word) voiced stops are actually produced as voiceless-unaspirated stops, with a VOT of ~0. This is because we, as English speakers, have really strong aspiration in our voiceless stops, so even if we produce something without much voicing during the closure, listeners will still be able to understand that it’s not aspirated, so clearly, the speaker must be intending to express voicing. Here’s a waveform of the word “guy”, to prove the point. Note that there’s a very little VOT here.

Finally, when following an /s/, English voiceless stops are not aspirated. So, in the word “sky”, we have an unaspirated stop, rather than the normal, aspirated [kʰ] which our writing system would lead us to expect. Here’s a waveform showing the very small VOT in “sky”:

So, in effect, the /g/ in “guy” and the /k/ in “sky” are the same sound! Still don’t believe me? Well, first listen to sky, then listen to guy, then listen to “sky” where I’ve digitally removed the /s/. Your writing system has been lying to you!

So what does Jimi Hendrix kissing men have to do with Stop Acoustics?

When we look at the acoustics of “guy” and “sky”, it’s very easy to see that the difference the two different perceptions of the lyric (“kiss the sky” and “kiss this guy”) are incredibly similar. When we realize that in English, [k] and [g] are functionally the same thing, the difference between our two choices:

… is seen to be only a question of where you put the /s/, and thus, really, no difference at all.

So, we see that not only are sounds in English not what our writing systems makes them out to be, but that this “error” of perception is not only understandable, but linguistically fascinating as well.

So, next time you find yourself listening to Purple Haze, Thank Jimi Hendrix for providing one of the best examples of the perceptual troubles which can come from our lack of a voiced/voiceless-unaspirated contrast in the English language. Or, curse me for linguistically corrupting an otherwise good song. Either or, really.

“Without further a due, you can get the latest nightly builds [at this website]“

This is a form of phonological re-analysis. When we re-analyze a word or phrase, we’re usually replacing an uncommon or non-transparent word with something that’s phonologically similar (that sounds alike), but is much more common or makes more sense. These are also referred to as “eggcorns”, a term coined by Geoff Pullum.

So, the speaker stumbles with “Ado” is a Middle English word, according the New Oxford American Dictionary, “from northern Middle English at do ‘to do,’ from Old Norse at (used to mark an infinitive) and do”). Rather than using “further ado”, the speaker (typer?) replaces it with a phonologically identical pair of words (“ado” /ədu/ “a dye” /ə du/) which are much more common in the English language. In short, the speaker replaces the word “ado”, a certified Zebra, with a common set of English words, “a due”, and thus, thinks horses.

A whole herd of Zebras, all horsed

We really like, as speakers of language, to turn zerbras into This happens relatively frequently, with varying degrees of phonological similarity. I’ve seen “do process” for “due process” (homophones like above), “play it by year” instead of “play it by ear” (/plej ɪt baj iɹ/ vs. /plej ɪt baj jiɹ/), where word segmentation makes the difference. Google gives 216 hits for “Torn ass under”, a (creative!) re-analysis of “torn asunder” (/tɔɹn əsʌndəɹ/ vs the original /tɔɹn æs ʔʌndəɹ/) to get around the ambiguity of “asunder”, meaning “into various pieces”. Entertainingly, this same “sunder” root causes yet another Zebra reanalysis. Not infrequently, you’ll hear people talking about “various insundry goods” in case of “Various and Sundry Goods” (/vɛɹiəs ɪnsʌndɹi ɡʊds/ vs. /vɛɹiəs ən sʌndɹi ɡʊds/). “Sundry” is definitely a zebra if you’re not familiar with “sundries”, items of various kinds, although interestingly, here, it’s replaced with another zebra, “insundry”.

With a bit more phonological difference, we get the reanalysis that many love to hate: “all intensive purposes” can be swapped for “all intents and purposes” (/ɑl ɪntɛnsɪv pəɹpəsɪz/ vs. /ɑl ɪntɛns ən pəɹpəsɪz/). And if we do this at a whole-phrase level while listening to music, we can get Mondegreens, a term for misheard song lyrics (hearing Jimi Hendrix’ “‘Scuse me while I kiss the sky” as “‘Scuse me while I kiss this guy”).

So, this is a relatively common phenomenon, and gives us great information about how speakers are coping with the amount of homophony in our language. In closing, thanks for reading Lingua Stick Miss Tick, and more importantly, thanks for not spelling it that way.

The first Hammered Dulcimer I ever owned (a Masterworks Russell Cook Edition) was very nice, and it had one characteristic odd note: E320 had an almost low-back-vowel-ey sound to it. Unfortunately, it also had a rattle inside the instrument which couldn’t be repaired, so it went back to their shop for diagnostics, and Russell (very generously) built me a new one, almost identical, save a few small changes.

That second dulcimer, my current dulcimer-love, is a beautiful instrument, but playing around this evening, I realized that that note doesn’t have that same quality (which I do now miss, ever-so-slightly), but I wasn’t sure what that quality was. Many of the rest of the notes sound nicer, so I don’t fault it, but it’s definitely different.

The Phonetics of Musical Instruments

This left me wondering, how do I determine that difference? Then, I remembered I’m a phonetics nerd, I do acoustical analysis of sounds for a living, so I decided to apply the same methodologies to my dulcimer(s).

First, I pulled up two recordings of the same song, one per dulcimer, new-and-old, in Praat. Then, I found the same timepoint of the same hit of that same note from both songs. I made a Fourier spectrum at that point for each of the two dulcimers for comparison (using the procedure described in Section 6.9 of Using Praat for Linguistic Research). This shows the amplitude (power) of all of the frequencies which make up the signal.

Below is the result:

(Also, for those interested, here’s the parts of the songs extracted with the odd note. The note in question is the third prominent note hit. Don’t mind the tempo change, my playing had improved in the two years between allowing me to play the song at faster speeds. Listen to the First Dulcimer and the Second Dulcimer.)

One can very clearly see how different these two notes are in terms of Timbre. The first dulcimer shows an almost vocalic set of strong resonances and weak ones. These resemble the resonances in vowel formants (see this spectrum of the vowel /ɔ/ for comparison), which explains why that note sounds so vowel-like to me. We can see that the second dulcimer shows a much more consistently tall set of harmonic peaks, with a much more consistent spectral tilt.

So, a little bit of phonetic analysis lets us see that although two instruments may be made by the same company, they may be the same instrument model, but they can have two fundamentally different voices.

If you’re interested, check out:

My Linguistics Bookmarks

My Phonetics/Phonology Bookmarks

My Natural Language Processing Bookmarks

Thanks, and I hope you all are enjoying your linguistic lives!