Ten findings about language that sound wrong

1

Newborn cries lean toward the melody of their mother tongue

Researchers recorded the cries of 30 French and 30 German newborns, two to five days old, and measured where the pitch peaked in each simple cry. Every cry analysed rose and then fell. In the French babies the peak came a little later in the cry, on average 58% of the way through; in the German babies a little earlier, at 44%. The authors call the later peak a “rising” contour and the earlier one a “falling” contour, and link the difference to the melody of the surrounding speech.

Fetuses can hear and remember melody in the last months of pregnancy. The authors think the babies probably learned this by listening before birth, though they cannot rule out learning in the first days of life. Their summary says only that the influence may come “possibly via vocal learning.”

This is one study. The two groups were recorded in different places, apparently by different teams, and, as Liberman reads the methods, the peaks were marked by hand without the coders being blind to the baby’s language. The gap between the group means, 0.14 of a cry’s length, is about one standard deviation: large by usual standards, but the groups still overlap substantially, and both produced all four contour types. Mark Liberman, who thought the finding probably real, listed these reasons a replication might find less.

Later work has not settled it. Studies of other language pairs, such as Mandarin and German, found differences in pitch variation. A 2017 study of Mandarin and American English newborns found that group differences all but disappeared once each baby’s cries were counted together rather than separately, and said this casts doubt on the idea that newborns match their cries to the ambient language.

Source. Mampe, Friederici, Christophe & Wermke, “Newborns’ cry melody is shaped by their native language,” Current Biology 19 (2009), 1994–1997. doi:10.1016/j.cub.2009.09.064

Limit. A single study of 60 babies in two languages. The babies preferentially produced those contours, not exclusively. The authors cannot rule out learning after birth. The groups overlap substantially. See Liberman’s critique on Language Log.

2

People blind from birth gesture when they talk, even to a blind listener

It seems obvious that people wave their hands because they have seen others do it, and because someone is watching. In 1998 Jana Iverson and Susan Goldin-Meadow filmed 12 children and teenagers, aged 9 to 18, who had been blind since birth, talking with a sighted experimenter, and compared them with 12 sighted peers. All of the blind speakers gestured, at a rate not reliably different from the sighted group. Four younger blind children, aged 5 to 8, were told that the experimenter was blind too. They gestured as well.

Gesture, the authors concluded, needs neither a model to copy nor a partner who can see it, and “appears to be integral to the speaking process itself.” They leave open whether gestures reflect, or even help, the thinking behind speech.

Source. Iverson & Goldin-Meadow, “Why people gesture when they speak,” Nature 396 (1998), 228. doi:10.1038/24300

Limit. A one-page report with 16 blind speakers. Only the four youngest spoke to a listener they were told was blind, and the authors note that the small groups may have made a difference hard to detect.

3

On one Canary island, people whistle Spanish, and the brain handles it as language

Silbo Gomero is a whistled form of Spanish from La Gomera. Nature’s report describes it as a rare and endangered way of communicating over long distances on difficult terrain, used by shepherds; its own summary adds that the language had been almost extinct and was by then taught in local schools. A whistle sounds like music or a signal, so it is a fair question whether the brain treats it as language.

Brain imaging of proficient whistlers listening to Silbo found activity in regions normally tied to spoken language. Spanish speakers who did not know Silbo showed no such activity. In the authors’ words, the language regions “can adapt to a surprisingly wide range of signalling forms.”

Source. Carreiras, Lopez, Rivero & Corina, “Linguistic perception: neural processing of a whistled language,” Nature 433 (2005), 31–32. doi:10.1038/433031a

Limit. One whistled language and a very small imaging study: five whistlers were scanned, usable data came from two or three of them for each task, and there were three controls.

4

“Huh?” is almost the same word in language after language

Linguists say words are arbitrary: no sound is tied to a meaning. Yet the little word people use when they did not catch what you said turns up in nearly the same form around the world. Mark Dingemanse and colleagues recorded everyday conversation in ten languages and found one short syllable with an unrounded vowel in the low front-to-central corner of the mouth, mostly on a rising pitch. A map in their paper shows a strongly similar interjection in 31 languages: those ten, ten more from an earlier survey by some of the same authors, and eleven from other published descriptions.

The authors argue that it is a learned word, not a grunt or an inborn cry, and propose that it converges because every language faces the same job: fixing the moment when one person did not hear another.

Source. Dingemanse, Torreira & Enfield, “Is ‘Huh?’ a universal word?”, PLoS ONE 8 (2013), e78273. doi:10.1371/journal.pone.0078273

Limit. Only the ten languages above were examined in detail, from 196 recorded instances. The other 21 rest on the authors’ earlier survey and on written sources, which are rarely explicit about the exact form. The convergence explanation is the authors’ proposal, not a demonstrated cause.

5

Read aloud, languages that sound fast are not saying more

Some languages sound fast. But a syllable in Vietnamese carries about 8.0 bits of information, and one in Basque about 4.8. Speakers of the first need fewer syllables a second to say the same thing. Christophe Coupé and colleagues had 170 speakers of 17 languages read aloud the same set of 15 short texts. The two tendencies largely offset each other: the information rate averaged 39.15 bits a second (standard deviation 5.1), though individual speakers’ rates ranged from about 4.3 to 9.1 syllables a second.

Source. Coupé, Oh, Dediu & Pellegrino, “Different languages, similar encoding efficiency,” Science Advances 5 (2019), eaaw2594. doi:10.1126/sciadv.aaw2594

Limit. 17 languages from 9 families, all in Europe and Asia; prepared texts read aloud, not conversation. The rate is “similar”, not identical: it varies by language and by speaker.

6

In ten languages, a reply to a yes/no question starts about a fifth of a second after the question ends, on average

Anthropologists have long reported cultures that pause politely and cultures that interrupt. Tanya Stivers and colleagues measured the timing of replies to yes/no questions in videotaped conversations in ten languages, from Japanese and Danish to ‡Ākhoe Hai‖om in Namibia, Lao and Tzeltal: 350 consecutive questions per language, 101 conversations in all. The timing comparison uses the yes/no questions among them, about two-thirds of the total. For each such question they measured the response offset, the time from the end of the question to the start of the reply. A gap counts as a positive offset and an overlap as a negative one. A reply could be a spoken yes or no, a longer answer, a non-answer or a gesture.

Across all ten languages the mean offset was +208 milliseconds, the median +100 and the most common value 0: very often the reply starts just as the question ends. Answers came faster than non-answers in every language, and all ten languages showed a general avoidance of both overlap and long silence between turns. Language means did differ, from +7 ms in Japanese to +469 ms in Danish, and the authors note that Danish sits only about one syllable’s time from the ten-language norm. They argue that sensitivity to such small tempo differences makes them feel like large differences of manner.

Source. Stivers et al., “Universals and cultural variation in turn-taking in conversation,” PNAS 106 (2009), 10587–10592. doi:10.1073/pnas.0903616106

Limit. Yes/no questions in informal conversation, in ten languages. The means cover every kind of reply, so they are not the silence before a “yes.”

7

Deaf signers get words stuck on the tip of the fingers

Everyone knows the feeling of a word just out of reach. Some theories predicted that signers would never have it: in sign, they argued, a word’s meaning and its form are bound together. Robin Thompson, Karen Emmorey and Tamar Gollan elicited the state in 33 deaf American Sign Language signers, who reported 79 episodes. It looked like the spoken version: it often resolved by itself, often involved names, and often brought back part of the form.

Which part? Signers more often recalled a sign’s handshape, location and orientation than its movement. For fingerspelled words, they often remembered the first letter. As with speech, meaning and form appear to be stored apart.

Source. Thompson, Emmorey & Gollan, “‘Tip of the fingers’ experiences by deaf signers,” Psychological Science 16 (2005), 856–860. doi:10.1111/j.1467-9280.2005.01626.x

Limit. American Sign Language only: 33 signers and 79 episodes, 55 of them for fingerspelled names and 24 for ordinary signs.

8

Short, common words tend to carry more meanings, and, the authors argue, that is efficient

Ambiguity looks like a defect that a well-designed language would remove. Steven Piantadosi, Harry Tily and Edward Gibson argue the opposite. If context usually tells the listener which meaning is meant, then a language gains by reusing its easiest words rather than coining a new one for each meaning. Their test in English, German and Dutch found the pattern: on average, shorter, more frequent and easier-to-say words have more meanings.

Source. Piantadosi, Tily & Gibson, “The communicative function of ambiguity in language,” Cognition 122 (2012), 280–291. doi:10.1016/j.cognition.2011.10.004

Limit. Corpus counts in three related European languages. This is a statistical tendency that does not hold equally for every measure and language, and the efficiency account is the authors’ argument.

9

Irregular verbs fade at a rate that follows a square-root law

English once had many irregular past tenses. Erez Lieberman and colleagues followed 177 Old English irregular verbs that are still in the language: 145 were still irregular in Middle English and 98 are today. How fast a verb gives up its irregular form depends on how often it is used. The half-life of an irregular verb scales as the square root of its frequency: a verb used 100 times less often regularizes 10 times as fast. The verbs people use most keep their oddities longest.

Source. Lieberman, Michel, Jackson, Tang & Nowak, “Quantifying the evolutionary dynamics of language,” Nature 449 (2007), 713–716. doi:10.1038/nature06137

Limit. English verbs only, over about 1,200 years. The 177 are verbs that survive today, so verbs that dropped out of the language are not counted.

10

Some of English’s silent letters were added later, one on purpose and one by mistake

The b in debt has never been pronounced. The word reached English as Middle English dette, from Old French dete, from Latin debitum. Scholars later put a b into the spelling to show the Latin ancestry; Etymonline dates that restored spelling to after about 1400.

The s in island has a different story. The Old English form was īeġland, with no s, and the word has nothing to do with Latin. The unrelated word isle, from Latin insula, had been respelled with an s, and in the 16th century island was respelled to match it, by mistake.

Sources. Etymologies of debt and island: Etymonline, debt; Etymonline, island; Merriam-Webster, “Nobody Pronounces the B in Debt”; Wiktionary, island.

Limit. Dictionary and word-history sources, not a study. They agree on the mechanism but not on exact dates for the first appearance of each letter, and the Oxford English Dictionary entries, behind a subscription, were not checked.