A map of the vowels in the mouth, with Romanian ă and î in the middle

Your map: seven vowels, twenty seconds#

Say seven vowels one after another, about a second each, as if singing them softly. The first three, i, a and u, are the corners of your mouth: from them the page sets your scale, so that a child’s voice, a woman’s and a man’s can sit on the same map.

Why are ă and î rare?#

On the map, ă and î sit in the middle: the tongue is neither forward, as for i and e, nor pulled back, as for u and o. î is high, with the mouth nearly closed; ă is lower, with the mouth half open. (Romanian writes î and â for the same vowel.)

▶ Listen: a, ă, î, i

PHOIBLE, a collection of 3,020 descriptions of the sounds of 2,176 languages (counted by Glottolog code), shows how unusual that spot is. About one language in six has a vowel written like î (371 languages) and about one in four a vowel like ă (504). Both are found in 129 languages, one in fourteen to seventeen depending on which description of each language you count (161 if any description will do). Among them: Amharic, Marathi, Kashmiri, Khmer, Komi and Ossetian.

Some of Europe’s big languages have only half the pair: in at least one PHOIBLE description, Russian and Polish have a vowel like î, and French and German one like ă. And among the Romance languages, writes Margaret Renwick, whose measurements are on the map, none but Romanian has î.

Yet ă and î tell very few words apart. Renwick gives the three pairs usually cited: rău (bad) and râu (river), țări (lands) and țâri (sea mackerels), văr (cousin) and vâr (I thrust). The first two come from Latin reus and rivus: in both, the r at the start pulled the vowel to the middle of the mouth.

Linguists do not even write them the same way. Renwick writes ă as ʌ rather than the usual ə, because ə suggests a weakened vowel, and her speakers gave ă its own place on the map and the length of a full vowel.

Which English vowels does Romanian lack?#

English vowels can sit on the same map, from recordings of 92 Americans (Hillenbrand and colleagues, 1995). Each American speaker is scaled by their own i, a and u, as you are; the Romanian averages are scaled by group (women, men).

• One Romanian e, two English ones. Romanian e lies near both the vowel of head and that of had, closest to head and a little higher in the mouth. English keeps them apart (bed, bad); Romanian has a single e in that area.
• î, close to heard. In the first two resonances, î sits next to the American vowel of heard. The difference is an r: Americans say it inside the vowel, and the r lowers a third resonance the map does not show, to about 1,711 Hz in men and 1,931 Hz in women, compared with at least 2,355 Hz and 2,734 Hz for the other vowels.
• ă, nearest to hud. The vowel of hud (or cup) is ă’s closest neighbour, but not the same: it is more open.
• Three vowels with no neighbour. The vowels of hid, hawed and hayed are the English vowels farthest from any Romanian vowel.

Aim for a Romanian vowel

Pick a target, then say the vowel and steer the dot into the circle. It works once you have made your map (the page needs your i, a and u). For î, try the vowel of heard with the r taken out, and let the dot tell you how close you are.

How does the page read your voice?#

When you say a vowel, your vocal folds buzz, and your mouth and throat, a tube of a certain shape, strengthen a few frequencies of the buzz. The first two, called formants, are enough to tell vowels apart: the first rises the wider you open your mouth, the second rises the further forward you push your tongue. The map has the first on the vertical axis (down means a more open mouth) and the second on the horizontal axis (left means the tongue further forward).

Dozens of times a second, the page takes the last 40 or so milliseconds of sound and finds the two resonances with a standard method from phonetics (linear prediction). For high voices, whose harmonics are sparse, it fits the model to the harmonics alone (El-Jaroudi and Makhoul, 1991).

The method was run on 1,658 recorded vowels (Hillenbrand, 1995) and compared with the resonances the authors measured and corrected by hand. On these studio recordings, at each vowel’s steadiest point, both resonances came within 10% of their value for 94% of the men’s vowels, 89% of the women’s and 86% of the children’s. In children and very high voices the first resonance is the hardest to catch, above all in i, u and î, where it sits close to the voice’s own pitch: there the dot shakes more.

The Romanian average comes from Margaret Renwick’s thesis (Cornell, 2012): 14 women and 3 men speaking standard Romanian, reading words in the frame “Spune ... de trei ori” (“Say ... three times”), measured at the middle of the stressed vowel. Each vowel’s patch is the average plus and minus one standard deviation (Renwick gives mean standard deviations: how much a vowel varies across words and repetitions). The map’s average is the midpoint of the women’s and the men’s averages, so the three men weigh as much as the 14 women. The American speakers were scaled here the same way as you; Renwick gives only averages for women and for men, so those were scaled by group.

The mouth sketch in the corner of the map shows what the two directions mean in textbooks: the tongue rises where the dot is. It is a diagram, not an X-ray: rounding the lips, for one, also moves the second resonance.

Questions#

Sources#

• Renwick, M. E. L. (2012). Vowels of Romanian: Historical, Phonological and Phonetic Studies. Teză de doctorat / PhD thesis, Cornell University. Tabelele / Tables 4.5–4.6.
• Hillenbrand, J., Getty, L. A., Clark, M. J., Wheeler, K. (1995). Acoustic characteristics of American English vowels. JASA 97(5), 3099–3111.
• Datele și înregistrările Hillenbrand 1995 / Hillenbrand 1995 data and recordings (copie, cu permisiune / mirror, with permission)
• Moran, S., McCloy, D. (eds.) (2019). PHOIBLE 2.0. CC BY 4.0.
• El-Jaroudi, A., Makhoul, J. (1991). Discrete all-pole modeling. IEEE Transactions on Signal Processing 39(2), 411–423.
• Klatt, D. H. (1980). Software for a cascade/parallel formant synthesizer. JASA 67(3), 971–995.
• Traunmüller, H. (1990). Analytical expressions for the tonotopic sensory scale. JASA 88(1), 97–100.

The Romanian measurements are Renwick’s tables for stressed vowels; the English ones are the steady-state values in vowdata.dat for the 45 men and 47 women who said all of i, a and u. The PHOIBLE numbers count the symbols ɨ and ə; a language whose middle vowel is written another way (Bulgarian, for one, with ɤ) is not counted.