Figure 1 summarizes some of the facts about these two groups. Both are mostly men between the ages of twenty-five and forty. It is not necessarily true that only English speakers become polyglots, though most of them grew up with only one language. It’s also not necessarily true that either talent or achievement is linked to IQ (the high skew probably stems from the fact that IQ was self-reported).
Figure 1
N = 390
Category 1: “I know >6 languages” N = 172
Gender – 69.2% male
Age – 44.8% 25 to 40
Mother Tongue – 43 reported >1 mother tongue; 84 had English as (one) mother tongue
IQ – 45.7% reported IQs over 140; 42% reported IQs 120–140
Origins – 1 from South Africa, China, Australia; 2 from India; 167 from Europe, the US, Canada, or South America
Category 2: “I learn languages more easily” N = 289
Gender – 65.6% male
Age – 41.7% 25 to 40
Mother Tongue – 53 reported >1 mother tongue; 162 had English as (one) mother tongue
IQ – 35.3% reported IQs over 140; 50.4% reported IQs 120–140
Origins – 1 from Vietnam, Pakistan, Singapore, China, India; 2 from Philippines; 3 from South Africa; 279 from Europe, US, Canada, or South America
I was also able to gather information about how many languages people know. Figure 2 shows the distribution of language repertoire size among people who participated in the survey. I asked them, “How many languages do you say that you know (spoken and written)?”
Figure 2 shows that people who know more than six languages are rare, but not as rare as those who claim to know eleven or more, who represent the true modern extremes of human language learning. Only two individuals had more than twenty (using their own definitions of “knowing” a language). This finding puts reports of Mezzofanti-size repertoires among modern hyperpolyglots into perspective. Over the entire curve, self-reported talent apparently makes a bigger contribution to language accumulation than does growing up bilingual. However, for the seventeen people with eleven or more languages, six of them grew up with one mother tongue. Thus, we can say that early bilingualism made a larger relative contribution to their overall accumulation than for the whole group.
Here are a few more facts about extreme modern language accumulators. (There’s more available at www.babelnomore.com.) Of the seventeen, 82 percent of them were male. Five were in the United Kingdom, four were in the United States, two each in Canada (one from Quebec) and Germany, and one each in India, Denmark, the Netherlands, and Latvia. They’re not particularly mobile; 62.5 percent said they live in the same place they grew up in. Twelve had English as one mother tongue and eleven said that English was now their dominant language. They claimed languages from an average of nine language families; as a group, their language families ranged from five to seventeen. Assuming that they’re developing literacy skills, each one worked in an average of five writing systems, ranging from two to nine.
Given the size of their repertoires, it would have been surprising if they were able to learn only major world languages. Among these extreme accumulators, the bulk of these linguistic collections consisted of European languages (representing the Romance, Germanic, Slavic, Finno-Ugric, Celtic, and Hellenic families), most of which are state languages (except for Macedonian, Latgalian, Welsh, Bavarian, Catalan, and Occitan, and historical versions of languages such as Old High German). Nevertheless, non-European languages were also represented: Arabic (in several regional forms), Hausa, Igbo, Afrikaans, Farsi, Hindi/Urdu, Kazakh, Kinyarwanda, Hawaiian, Japanese, Mandarin, Cantonese, Mongolian, Vietnamese, Malay/Indonesian, Korean, Southern Min, Wu, Thai, Hindi, Marathi, Gujarati, Sanskrit, Innu-aimun, and Cree. Apart from Farsi, Japanese, Mandarin, Cantonese, Korean, Thai, and Hindi, all the languages were reported by only one person apiece. Overall, there were only two Native American languages, Innu-aimun and Cree.
Figure 2. Polyglot language repertoires.
I asked them in what order they learned these languages. A common assumption is that language accumulators learn within language families in order to “rack up numbers,” but my analysis shows this isn’t strictly true. From language to language, they were more likely to move between different language families (e.g., English → Arabic) than within the same family (e.g., English → German → Dutch). About a quarter of the time, they studied languages from families they’d encountered before. Another interesting pattern was that Asian languages, nonstate languages, and languages with smaller populations of speakers were more apt to be learned later than earlier.
All of them said that learning languages was easier for them than for others. The survey asked them why. Sixty-two percent said it’s due to an innate talent; 69 percent said it’s because they’re more motivated; and 88 percent said it’s because they like languages. In other studies with similar findings, motivation is treated as a personality trait, which means you can’t predict who will and who won’t have it. To the contrary, the neural tribe theory, borrowing from Ellen Winner’s descriptions of gifted children, suggests that you’re born with the capacity to be motivated to learn foreign languages. How that capacity is developed is a matter of biographical detail and historical circumstance.
In the stairwell of Gregg Cox’s house in Bremen, Germany, I’d seen the plaque from The Guinness Book of World Records: “Gregg M. Cox of Oregon, USA, the Greatest Living Linguist, is able to read, write and speak 64 languages, 11 different dialects, speaking 14 fluently.” It had once hung in the hallway, but his wife, Sabine, moved it to the staircase because her clients (she’s a cosmetologist) always wanted to talk about her husband. “You don’t want to tell every customer your entire history and who you’re married to,” Sabine said. “I mean, I’m proud, but . . .” Her sentence trailed off.
The three of us were standing in Cox’s office; she was talking while Cox, a short, bald man, rummaged through papers in his desk.
“I got tired of it, so I hung it in a different place,” she said. “I go shopping sometimes,” she said wearily. “They say, ‘Cox, oh, are you his wife?’ I say yes.”
“Do they expect you to speak a lot of languages?” I asked.
“I get that a lot,” she replied. “I say, I only speak two languages. But then I say, it doesn’t matter how many languages I speak, because he doesn’t understand me anyway.”
When Cox and I sat down to talk again, I asked him if there were any myths about hyperpolyglots. He immediately replied, “That they can jump back and forth between all their languages. That’s the biggest myth. I’ve met several other polyglots, and we’ve been able to bounce back and forth in seven or eight languages, but not further than that,” he said. “The most languages that I’ve ever had back and forth with somebody was seven.” (I can attest that I heard Cox speak English, German, and Spanish in our several days together.)
As Cox well knew, there’s a persistent myth that hyperpolyglots are able to maintain very high skills in each and every one of their languages. When this can be proven false, then they’re considered to be discredited. You can graph this by putting on the y axis a global measure of language skills and proficiencies, and on the x axis the number of languages, which looks like this:
Figure 3. Imagined distribution of a hyperpolyglot’s language abilities.
Yet hyperpolyglots are not unique in this—the same myth applies to bilinguals. The reality, as bilingual researcher François Grosjean notes, is that “some bilinguals are dominant in one language, others do not know how to read and write one of their languages, others have only passive knowledge of a language and, finally, a very small minority have equal and perfect fluency in their languages.”
As I had discovered, the actual curve of skills and proficiencies over number of languages looks (roughly) more like this:
Figure 4. Actual distribution of hyperpolyglots’ language abilities.
Where you can find data for specific people, you can plug th
em into the curve. Let’s use Charles Russell’s careful accounting to generate one for Mezzofanti. We know that he didn’t speak or read all of his languages to equal degree—one group he had to a very high degree (though the figure of thirty might be exaggerated), and his “surge” languages to some middling degree, along with the “bits of language” on the tail end.
Figure 5. Distribution of Mezzofanti’s language abilities.
You can produce a similar curve by plugging in modern criteria and data from contemporary hyperpolyglots. For instance, Air Force officer Edgar Donovan scored 3 (out of 5) in listening and reading in Spanish, Italian, French, and Brazilian Portuguese, and 2 (out of 5) in European Portuguese. In his other languages, he has negligible proficiency, and he received a number of zeros (meaning that he knows practically nothing of the language) in Arabic, Hebrew, Turkish, and Serbo-Croatian.
A more extensive picture came from Graham Cansdale, the European Commission translator I’d described to Loraine Obler. He possessed a cluster of Geschwind-Galaburda traits (gay, spatially limited, verbally gifted). After meeting Graham in his office, I stayed in touch with him. Later he agreed to rate his abilities in all of his languages using surveys of speaking, listening, and reading that were adapted by the American Council on the Teaching of Foreign Languages. The surveys are based on a scale first developed by the US State Department’s Foreign Service Institute in the 1950s (the test became mandatory for all foreign service officers in 1958) and modified over the following decades by the Interagency Language Roundtable (ILR), an informal group in the federal government that shares information on language teaching and testing.
As is the case with the Defense Language Proficiency Test, the ACTFL scale makes the educated native speaker the pinnacle of achievement. Still, the scale suited my purposes well, especially since everyone knows its shortcomings. Using a full version of a proficiency test would have been expensive and time-consuming. We also know there’s good correlation between the skills that someone reports and their actual skill level.*
I interviewed Graham about his language learning history, which provides the basis for the x axis of Figure 6; the y axis is the ratings of the ILR scale. Graham then filled out the self-assessments. The standards grade on a scale from 0 to 5. In speaking, for instance, a 0 means no proficiency, 0+ refers to an ability to use rehearsed utterances, while 5 is a “functionally native proficiency.”
Figure 6 is Graham’s total language system, which is a patchwork of proficiency. Here’s a native English speaker, raised monolingually, who now lives in predominantly Francophone Brussels with a native Slovakian speaker (which explains his strong oral skills on all three) and who uses a range of languages for his work. There are higher scores in reading than in speaking because he’s a translator; even so, he said he’s reached a “critical mass” in some of the languages, so that his abilities degrade more slowly. Not only does practice matter, but having consolidated certain languages, such as Spanish, by living in a place also helps. (One benefit of his job is substantial opportunities for travel and professional development.)
Figure 7 graphs his total speaking proficiency.
Of particular interest is that he claims no perfect score in any language but English, and in one-third to one-half of the languages he has studied, he has no skills to report in speaking, reading, or listening.
Figure 6. Distribution of Graham Cansdale’s self-reported proficiencies in speaking, listening, and reading, arranged by language biography (2010).
This curve appears to map onto Graham’s biography, such that his later learned languages appear lower on the curve. Alternatively, it might reflect how much practice he gets in which languages. Or it might map both. The fact that there appear to be three clusters of abilities—very high ones, a set of middling ones, and a long tail of very low ones—reflects the three levels of retained knowledge that linguists Kees de Bot and Saskia Stoessel laid out in an exploration of how people lose and relearn languages. At the highest level is knowledge that remains active. At the next level are items that can be recognized passively. At the third is knowledge that was thought lost but is still there—things that had been “saved.” Saved languages are important to hyperpolyglots—over and over, people mentioned their “surge” languages proudly. But such knowledge isn’t quantified in the modern language testing world. Aptitude can’t account for it, either.
Earlier I suggested that reading and translating activities require less mental effort than speaking. Indeed, in Figure 8 is how Graham assessed his reading proficiency; note that the line doesn’t drop off as sharply as the speaking curve. He’s able to read more languages than he can speak.
Figure 7. Distribution of Graham Cansdale’s self-reported proficiencies in speaking, arranged by score (2010).
Figure 8. Distribution of Graham Cansdale’s self-reported proficiencies in reading, arranged by score (2010).
I also asked experts how many languages a person could control—how many they could switch back and forth between with no confusion. They said that there was no theoretical limit to the number of languages one could learn. Time, not cognition, seemed to be the limiting factor. “There’s really no limit to the human capacity for language except for things like having enough time to get enough exposure to the language,” said Suzanne Flynn, a psycholinguist at MIT who studies bilingualism and trilingualism. “It gets easier the more languages you know.” Harvard University psycholinguist Steven Pinker agreed. Asked if there is any theoretical reason someone couldn’t learn dozens of languages, he replied: “No theoretical reason I can think of, except eventually interference—similar kinds of knowledge can interfere with one another.”
But there are real limits—ask hyperpolyglots themselves. Out of respect for Erik Gunnemark, I’ll count only the contemporary superlearners. Gunnemark, in a letter to Alexander, wrote that “if you read or hear that a certain person ‘can speak’ (or ‘speaks’) a large number of languages (for instance twenty or more) you should always be a little skeptical.” Gunnemark insisted that only thirteen languages be attributed to him, those he spoke “fluently” or “fairly well.” He never counted the fifteen languages he spoke at a “mini” level. (Amorey Gethin, who collaborated with Gunnemark on The Art and Science of Learning Languages, told me in a phone call that “I don’t think [Erik] could speak very many languages very well—but he could read them. He could read quite a few languages.”)
A clearer limit came from my survey. Out of 167 respondents, only twenty-eight said they knew ten or more languages; only seven knew fifteen or more; and the highest number claimed was twenty-six. One thing to note is that my standard was very broad—people didn’t have to speak the language (they could also write it), and the definition of “knowing” was up to them. I don’t know if they were counting all the languages they had ever encountered or all the ones they claimed to have readily accessible. Either way, one consequence is that if you thought Ziad Fazah and Gregg Cox had exaggerated, you’d really think so now. The map of what’s possible, even at its extremes, doesn’t seem to extend to fifty-nine (for Fazah) or sixty-four (for Cox) languages simultaneously. Among other things, this would seem to invalidate Cox’s Guinness record. (Yet I’m not inclined to diss Guinness record keeping altogether, as you’ll see.)
I’d started out looking for what the human limits might be in practice, not in theory. The hypothetical immortal with a million lifetimes to learn everything adds about as much understanding to the phenomenon as the native speaker of one language does—it tells us that exceptional language learning takes place within the very real constraints of our very real world. Since that’s the case, we should look to real hyperpolyglots and the results of their natural experiment.
Because working memory capacity is finite, one could predict that there should be a limit to how many languages someone can keep active. Lomb Kató described active languages as the ones that “lived” inside of her; Claire Kramsch described them as languages one “resonate[d]
” with. Indeed, if you want to understand the upper limits of language control rather than language learning or memory, you should look to the hyperpolyglots, too. Lomb said she only had five languages “living” inside her; Cox told me he could switch back and forth only in seven. Gunnemark reported being fluent in six.
Presumably, they could have controlled more languages if they needed them, given that they claimed to have “surge” languages as well. These seem to range between five and nine, though hyperpolyglots are able to manage more for short periods of time. When Helen Abadzi worked as an interpreter at the Athens Olympics, she said she worked in ten languages simultaneously, but she carried a PDA loaded with dictionaries. There was also the polyglot contest she’d participated in—to win that, you’d have to keep many languages powered up for at least a day or two (but not longer).
Such a limit has been proposed before. Psychologist and Russian hyperpolyglot hunter Dimitri Spivak deemed it “the rule of 7.” For his book Kak stat’ poliglotom (or “How One Becomes a Polyglot,” which is only available in Russian), Spivak interviewed polyglots across Russia and asked how many languages they felt they really knew. Though his rule is disputed, no one has offered counterevidence. Spivak adds that it doesn’t matter if the languages are spoken or written: “The brain tends to treat each set of homogeneous units as a simple set,” he said in an email. “There’s no substantial difference between storing or recalling from long-term memory seven languages, or seven systems of writing.”
Babel No More Page 24