Human Uniqueness: Language, Mind, and FOXP2

"Apes can learn many isolated symbols (as can dogs, horses, and other nonhuman species), but they show no unequivocal evidence of mastering the conversational, semantic, or syntactic organization of language." — Herbert Terrace and colleagues, Science, 1979¹

"The gap between us and them is simply too great to provide any understanding of evolutionary precursors or the evolutionary processes (e.g., selection) that led to change over time." — Marc Hauser, Noam Chomsky, Richard Lewontin and five others, 2014²

Summary

This article covers the biology:  what apes taught to use symbols could and could not do, what the gene FOXP2 does, and how human brains differ from ape brains beyond their size.  Where language began, and whether all languages come from one, is argued under Did Language Begin Once?.  The philosophy of mind belongs to Mind, Reason, and Consciousness.

1. The gap is large on every measure anyone has taken. The vocabularies their researchers documented run from about 50 to 161 signs or symbols, and the larger claims made later were never put to a test. An average 20-year-old American knows about 42,000 words.³ Nim Chimpsky's signed utterances averaged 1.1 to 1.6 signs and did not grow over 19 months, while his vocabulary rose from 42 to 125 signs.¹ Kanzi the bonobo, the strongest case, understood novel spoken requests about as well as a two-year-old girl tested the same way.⁴
2. Apes learn symbols. None has shown grammar. Project Nim's own videotapes showed most of his combinations were prompted by his teachers.¹ A statistical test in 2013 found that Nim's combinations "lacked the combinatorial range of a grammar", which very young children already show.⁵ Koko's foundation says she learned "over 1000 signs",⁶ but no corpus of her signing was ever released for testing.⁵
3. FOXP2 is a gene needed for normal speech, not "the language gene". One changed letter in it causes a severe speech disorder in a family known as KE, centred on control of the mouth and face.⁷ Humans differ from chimpanzees at two of its 715 amino acids.⁸ Neanderthals carried both human changes,⁹ and the 2002 claim that the human form swept through our species in the last 200,000 years failed when it was retested on more diverse genomes in 2018.¹⁰ In mice the human form changes squeaks and brain circuits for motor learning;¹¹ in songbirds, turning the gene down spoils song copying.¹²
4. The human brain differs in wiring and in development genes, not in having new language areas. Apes and monkeys have counterparts of Broca's and Wernicke's areas.¹³ The arcuate fasciculus, the fibre tract that links them, reaches much further into the human temporal lobe.¹⁴ Several genes made by partial duplication exist only in humans, among them NOTCH2NL, ARHGAP11B and SRGAP2C; they change how many neurons the growing cortex makes or how its connections mature. None is specific to language, and ARHGAP11B's human form was shared by Neanderthals and Denisovans.¹⁵
5. Noam Chomsky and Robert Berwick argue that the core of language appeared suddenly, in one step. They say the operation that builds sentences, which they call Merge, cannot exist in half-form, so there was "no room" for a protolanguage.¹⁶ Other linguists reply that an all-or-nothing operation need not arise from a single mutation.¹⁷ Eight leading researchers, Chomsky among them, wrote in 2014 that "we lack a connect-the-dots account of any gene to language phenotype".²
6. Creationist and design writers read the gap as the mark of design and the image of God. On the size of the gap they are right, and on FOXP2 Answers in Genesis said early and correctly that it is not a "language gene".¹⁸ But some of their claims go past the data: that FOXP2 shows Neanderthals "were Homo sapiens",¹⁹ that language areas are "non-existent in non-human creatures",²⁰ and that only humans have a hyoid bone holding vocal cords.²¹
7. Ape-language and FOXP2 researchers overreached too, and earlier. The jacket of the 1981 book on Koko credited her with "a vocabulary of more than 600 words" and "speculating on death".²² The FOXP2 team suggested in 2002 that its sweep fitted "the appearance of a more-proficient spoken language".⁸ Both claims outran their data.
8. Two amino-acid changes are not a problem for evolution, and they are not evidence for it either. On the FOXP2 team's own rates, two changes on the human branch is about a 1-in-900 to 1-in-1,600 event for this one gene,²³ which is unusual but expected somewhere among 20,000 genes. Mutation supplies the changes easily.²⁴

Verdict.  On the data, the gap between human and ape language is real, large and qualitative:  apes learn symbols and some word order, but no ape has produced or shown it understands hierarchical grammar, and no gene or brain feature found so far explains how humans do.  That fits the biblical claim that humans alone bear the image of God, and it embarrasses the decades of claims that apes were learning language.  It does not show how the gap arose.  Chomsky, an evolutionist, predicts a sudden gap as firmly as a creationist does, and the fact that no one has found the mechanism is a statement about present knowledge.  The FOXP2 and Neanderthal evidence fits a young-earth view of Neanderthals as human.  The dates attached to that evidence, 300,000 to 400,000 years for the shared ancestor, do not, and are argued in Neanderthals.

Part 1: How large the gap is, in numbers

The table puts the best-documented apes beside human figures.  Every number is the researchers' own.  Vocabulary counts are not strictly comparable:  the ape projects counted signs or symbols that met a usage criterion, and the human figure counts dictionary headwords known well enough to recognise.

Three things stand out.  First, the vocabularies are real.  Apes link arbitrary signs to objects, actions and people, and this is no longer disputed.  Charles Yang, whose 2013 test found no grammar in Nim, still wrote that apes' "ability to learn word-like symbolic units is well documented".⁵  Second, the vocabularies are small:  a few hundred against tens of thousands, a gap of a hundredfold or more.  Third, and most important, the length and structure of what apes produce does not grow the way a child's does.  A child adds words to say more.  Nim added words by repeating himself.

Comprehension is where apes come closest.  Kanzi's scores matched or beat Alia's overall, and he acted correctly on 81 % of sentences whose key words were tested in both orders, against Alia's 64 %.⁴  This shows Kanzi used word order, not just the words.  But the one sentence type that needs two nouns joined into a single phrase, "Give object X and object Y to animate A", was his worst at 33 %.⁴  Alia, not yet two, scored 57 % on the same type.

Part 2: The ape language experiments, one at a time

No ape has been taught to speak, and for monkeys at least the reason is in the brain, not the throat.  W. Tecumseh Fitch and colleagues filmed a macaque's vocal tract by x-ray and found it could make a wide range of vowels.  "Despite repeated attempts, no nonhuman primates have ever been trained to produce speech sounds, not even chimpanzees raised from birth in human homes."²⁷  They concluded that a macaque with a brain "capable of vocal learning and combinatoric operations over speech sounds" would have a vocal tract able to produce "clearly intelligible speech".²⁷  So from the 1960s researchers turned to hands and keyboards.

Washoe: real signs, no record of the prompts

Allen and Beatrix Gardner raised the young chimpanzee Washoe in Reno, Nevada, for 51 months²⁵ and taught her American Sign Language.  In their own summary, "Washoe acquired at least 132 signs.  She asked for goods and services, and she also asked questions about the world of objects and events around her."  Once she had about eight signs, "she began to combine them into meaningful phrases.  YOU ME HIDE, and YOU ME GO OUT HURRY were common."²⁵

What Washoe showed is that a chimpanzee can learn signs and use them to name things it is shown.  Sue Savage-Rumbaugh, who spent her career arguing for ape language, judged the Gardners' blind tests sound:  they "clearly implied that the chimpanzees could recognize and respond to photographs by producing the appropriate sign under conditions that precluded cuing."⁴

What Washoe did not show is grammar.  The Gardners reported almost nothing about the order of her signs, so "there is no basis for deciding whether Washoe's multisign combinations obeyed rules of sign order", Terrace's team wrote.¹  Terrace's team then studied the Gardners' own film, which showed 155 of Washoe's utterances.  Every one of her 35 multisign sequences "were preceded by a similar utterance or a prompt from her teacher."¹  Savage-Rumbaugh agreed on the deeper problem:  "Since the Gardners had no record of the signs directed to Washoe by experimenters, they were unable to provide data to refute these claims satisfactorily."  She added, "Unfortunately, the Gardners continued to fail to address this issue straightforwardly."⁴

Nim Chimpsky: the project that tested itself

Herbert Terrace set out at Columbia University to confirm the Gardners' work and to go further, by recording everything.  Nim was taught by more than sixty teachers over four years.²⁶  The team analysed "More than 19,000 multisign utterances".¹  At first the data looked like grammar.  When Nim used more in a two-sign combination, it came first 85 percent of the time,²⁶ and the team found such regularities "could not be attributed to memorization or to position habits".¹

Then the team studied its videotape.  "That conclusion, however, was invalidated by videotape analyses, which showed that most of Nim's utterances were prompted by his teacher's prior utterance, and that Nim interrupted his teachers to a much larger extent than a child interrupts an adult's speech."¹  Nim's average utterance stayed at 1.1 to 1.6 signs while his vocabulary more than doubled.¹  "The most striking aspect of these functions is the lack of growth of Nim's MLU during a 19-month period."¹  Films of other apes, Washoe and Koko among them, showed the same dependence on the teacher's prior signing.¹

Terrace's conclusion was firm:  "There is no evidence, however, that apes can combine such symbols in order to create new meanings," and "evidence that apes create sentences can, in each case, be explained by reference to simpler nonlinguistic processes."¹

Nim's corpus is still the only primate language corpus ever made public.⁵  In 2013 the linguist Charles Yang tested it with a new statistic.  If a speaker combines words by a rule, the number of different pairings should be high, and Yang showed that young children's speech meets that standard.  Nim's did not:  he "falls considerably below the expected diversity of a rule-based system ( P < 0.004, paired one-tailed Mann–Whitney test)."  Yang concluded that "Nim’s signing lacked the combinatorial range of a grammar, the hallmark of human language evident even in very young children’s speech."⁵  The test does not depend on anyone's reading of the video, which answers the charge that Terrace's judgment was subjective.

Koko: large claims, no released data

Francine "Penny" Patterson began teaching sign language to the gorilla Koko while a graduate student in psychology at Stanford.²²  By the Gardners' criteria Koko learned faster than Washoe:  161 qualified signs to Washoe's 132 at 51 months.²²  Patterson herself reported Koko's average utterance at "roughly 2.7 compared to Nim’s 1.5 signs per statement", and called it "still a low figure compared to children’s sentences, and one that makes it difficult to examine word order."²²

The public claims went much further.  The jacket of Patterson's 1981 book said that Koko "is drawing on a vocabulary of more than 600 words, conversing through a computer terminal, signing with her mate-to-be, Michael, and speculating on death."²²  The Gorilla Foundation's site still says "Koko has learned over 1000 signs".⁶  No transcript corpus of Koko's signing has been released for others to test, so these claims cannot be checked the way Nim's were.  The only Koko record Terrace's team could study was a television film.  In it, "all of Koko's, signs were signed by the teacher immediately before" she signed them, though the team granted that a few films are "much more limited in scope" than their own videotapes.¹

Kanzi: the strongest case

Kanzi, a bonobo born in October 1980, "was exposed to human language and to lexigrams from age 6 months, while still with his bonobo mother."⁴  Lexigrams are symbols on a keyboard.  Like the child he was later compared with, he was not "trained" to talk.⁴  His record is the best ape result there is, and it is still small.

By about 17 months of record-keeping, Kanzi used "about fifty symbols"²⁶ and "understood 150 spoken words".²⁶  In a five-month sample when he was five and a half, he produced 13,691 utterances, of which "just a little more than 10 percent" had two or more elements;  about half of those were spontaneous, leaving "a corpus of 723 combinations".²⁶

His comprehension test is the heart of the case.  From May 1988 to February 1989 Savage-Rumbaugh's team gave Kanzi, then eight, hundreds of novel spoken requests, and later gave the same kind of test to Alia, a girl "between 1½ and 2 years of age".⁴  A typical request was "Make the gorilla mask bite Kelly."⁴  In the blind trials the speaker stayed behind a one-way mirror, so no face or gesture could cue the answer.  "Overall, Kanzi was correct on 72% of all trials and 74% of the blind ones.  Alia was correct on 66% of all trials and 65% of the blind ones."⁴  On requests whose key words were reversed between trials, Kanzi got 81 % right.⁴  He used word order.

Two limits matter.  Kanzi's comprehension was matched by a child not yet two, and his weakest result was the type that needs two nouns joined into one phrase, at 33 %.⁴  And Savage-Rumbaugh herself withdrew the word "grammar" for what Kanzi produced:  "In retrospect, it is clear we should not have used the word “grammar,” either in our letter or in our manuscript."  She preferred "protogrammar", a word she said fits young children too.²⁶

What the four projects establish

Taken together, the projects show four things.  Great apes learn a hundred or more arbitrary symbols and use them to ask for things and to name them.  They put symbols together, mostly in twos.  The best of them understands simple spoken sentences, including word order, at the level of a toddler.  None has shown that it builds phrases inside phrases, the hierarchical structure of human grammar.  Terrace's summary from 1979 has not been overturned:  "Apes can learn many isolated symbols (as can dogs, horses, and other nonhuman species), but they show no unequivocal evidence of mastering the conversational, semantic, or syntactic organization of language."¹

Part 3: FOXP2, the gene that was called "the language gene"

The KE family

FOXP2 entered the story through one family.  In a three-generation family known as KE, a severe speech and language disorder is passed down as a single dominant gene.  Cecilia Lai, Simon Fisher and colleagues at Oxford found the gene in 2001.  It was broken by a chromosome break in an unrelated patient with the same disorder, and in the KE family a single letter change, G to A, swapped one amino acid (R553H) in the part of the protein that binds DNA.  "All affected individuals from the KE pedigree were heterozygous for this mutation, whereas all unaffected individuals were homozygous for the wild type".⁷

The disorder is not a loss of grammar alone.  "Affected members have a severe impairment in the selection and sequencing of fine orofacial movements, which are necessary for articulation".⁷  Their trouble also reaches understanding:  "there is substantial impairment of expressive and receptive language abilities."⁷  Their general intelligence varies:  although the affected members' average non-verbal IQ is lower, "there are affected individuals in the family who have non-verbal ability close to the population average, despite having severe speech and language difficulties".⁷  The authors claimed only that "FOXP2 is involved in the developmental process that culminates in speech and language."⁷

FOXP2 makes a transcription factor, a protein that switches other genes on and off.  Hauser and colleagues note that it acts "in many different tissue types (brain, lung, gut lining) at different times during development as well as throughout life."²  A gene that switches many others is not a gene "for" any one trait.  Fisher, who co-discovered it, wrote in 2017 that "discussions that centre on an abstract “gene for language” are unconstructive".²⁸  Other genes, such as ERC1 and BCL11A, have since been found broken in children with similar disorders.²⁸

Two human changes

In 2002 Wolfgang Enard, Svante Pääbo and colleagues compared the protein in humans, apes, a monkey and the mouse.  FOXP2 "is among the 5% most conserved proteins among mammals".⁹  Leaving aside two runs of the amino acid glutamine, "the human FOXP2 protein differs at only three amino-acid positions from its orthologue in the mouse".⁸  Chimpanzee, gorilla and rhesus monkey proteins are identical to each other, and "two of the three amino-acid differences between humans and mice occurred on the human lineage after the separation from the common ancestor with the chimpanzee."⁸  The two changes are at positions 303 and 325 of the 715-amino-acid protein, threonine to asparagine and asparagine to serine.⁸

The team then made a claim that later failed.  They found an excess of rare variants near the gene in living humans, a classic sign of a recent selective sweep, and estimated "that the fixation occurred during the last 200,000 years of human history".  They added:  "This is compatible with a model in which the expansion of modern humans was driven by the appearance of a more-proficient spoken language".⁸

Neanderthals had it

In 2007 Johannes Krause, Pääbo and colleagues recovered FOXP2 from two Neanderthals from El Sidrón Cave in Spain, with checks against contamination by modern DNA.  Both Neanderthals carried both human changes, on the same background as most living humans.  "Whatever function the two amino acid substitutions might have for human language ability, it was present not only in modern humans but also in late Neandertals."⁹  On the mainstream timeline, that pushed the changes back before the common ancestor of the two groups, "which existed about 300,000–400,000 years ago".⁹  That date is the mainstream reading and is not accepted here without question;  the dispute over it belongs to Neanderthals.

The sweep was not there

In 2018 Elizabeth Atkinson, Brenna Henn and colleagues retested the sweep with hundreds of genomes from around the world.  The 2002 study had sampled mostly people whose ancestors left Africa, a group that went through a population bottleneck.  When Atkinson's team built a sample with the same mix, they reproduced the old signal.  When they tested Africans and non-Africans separately against their own backgrounds, the signal vanished.  "We do not find evidence for recent positive or balancing selection at FOXP2.  The original signal appears to have instead been due to sample composition."¹⁰

The authors did not claim FOXP2 is unimportant.  "We do not dispute the extensive functional evidence supporting FOXP2’s important role in the neurological processes related to language production".  What they ruled out is the story of a recent language sweep:  "recent natural selection in the ancestral Homo sapiens population cannot be attributed to the FOXP2 locus and thus Homo sapiens’ development of spoken language."¹⁰

What the gene does in mice and birds

Enard's group put the two human changes into mice in 2009.  The mice were "generally healthy", but they had "qualitatively different ultrasonic vocalizations, decreased exploratory behavior and decreased dopamine concentrations in the brain", and nerve cells in the striatum, part of the basal ganglia, grew longer branches and showed more synaptic plasticity.  Mice carrying one broken copy show the opposite effects.¹¹  The mice did not talk.  The human form tunes circuits for learning motor sequences.

Songbirds learn their songs by copying a tutor, as children learn the sounds of speech.  In zebra finches, FOXP2 rises in the song-learning part of the basal ganglia while young birds learn.  When Sebastian Haesler, Constance Scharff and colleagues turned the gene down there, "Knockdown of FoxP2 resulted in an incomplete and inaccurate imitation of tutor song," and the syllables of adult song "were abnormally variable, similar to word production in children with DVD", the KE family's disorder.¹²

The mouse and bird work tells the same story as the KE family.  FOXP2 is needed for learning and sequencing fine movements of the mouth and throat, which speech requires.  It is not the source of grammar.  Berwick and Chomsky put it plainly:  FOXP2 is "like the printer attached to a computer, rather than the computer’s CPU."¹⁶

Pricing the two changes

Creationists and evolutionists both treat the two human changes as remarkable, and both can be priced.

Are two changes too many for evolution?  On Enard's own figures, FOXP2 changed once in about 130 million years of branch from the mouse to the human-chimpanzee ancestor, then twice on the human branch since the split "about 4.6–6.2 Myr ago".⁸  If the gene had kept its old pace, two or more changes on that branch would be a 1-in-900 to 1-in-1,600 event.²³  That is unusual for this gene.  It is not unusual for a genome:  with about 20,000 genes, a few such outliers are expected by chance alone.

Can mutation supply them?  Easily.  Each change is a single-letter substitution.  A gene of 2,145 coding letters,⁷ at the measured human mutation rate, would collect about six neutral changes on the human branch if nothing constrained it.²⁴  It collected two that changed the protein.  There is no waiting-time problem here, and nothing in FOXP2 disproves evolution.

Do the changes prove evolution?  No.  Common descent predicts a small number of differences between closely related species, and a creation model in which humans were made distinct predicts a small, telling difference as well.  Two amino-acid changes cannot tell these apart.

Part 4: The brain beyond its size

Human brains are about three times larger than chimpanzee brains, organised differently, and mature over a longer time.²⁹  Size is covered in Brain Size and the Order of the Fossils and in Humans and Chimpanzees.  Two points from that work matter here.  First, size alone does not explain language.  Gorillas match or exceed humans in body size, but their brains are "about one-third of the size of the human brain",³⁰ and the human brain, with about 86 billion neurons, "has just as many neurons as would be expected of a generic primate brain of its size".³⁰  Second, Neanderthal brains were "as large as those of modern humans",¹⁵ so size cannot separate the two either.

No new language areas, but different wiring

A common claim is that humans have brain regions for language that apes lack.  The comparative anatomy does not support it.  James Rilling and colleagues, who argue for real human specialisation, write:  "There is, in fact, considerable evidence that homologs of Broca’s and Wernicke’s areas exist in apes and monkeys".¹³  These are the two areas classically linked to producing and understanding language.

What differs is the connection between them.  The arcuate fasciculus is a bundle of fibres arching from the temporal lobe to the frontal lobe.  In 2008 Rilling's team traced it with diffusion imaging in humans, chimpanzees and macaques and "found a prominent temporal lobe projection of the human arcuate fasciculus that is much smaller or absent in nonhuman primates."¹⁴  In a larger study of 26 humans and 26 chimpanzees,¹³ they found that "in humans the arcuate projections into the temporal cortex are concentrated in STS and MTG, ventral to classic Wernicke’s area, whereas in chimpanzees they are concentrated in STG."¹³  In plain words, the human tract reaches further down into the middle of the temporal lobe, below Wernicke's area, than the chimpanzee tract does.  Rilling's team argued that this pathway "is likely to be the pathway most essential for higher-order aspects of human language such as syntax and lexical–semantics."¹³

A 2020 study traced an auditory part of this pathway in macaques and chimpanzees as well.  Brian Thomas of the Institute for Creation Research quoted its authors' finding of "the differentiation of the human right dorsal pathway connectivity away from the auditory prototype" and replied:  "In other words, human’s language pathways are uniquely human, as is language itself.  That exactly fits the creation model".³¹  Both readings take the same data.  The shared part is what common descent expects;  the human-only extension is what both a creation model and an evolutionary change would need.  The wiring difference is real.  It does not, by itself, say how it arose.

Genes found only in humans

Since 2012 several genes have been found that exist only in humans, made by partial copying of older genes.  Three are the best studied.

The ARHGAP11B story shows how little sequence change can matter.  The human gene differs in function from its parent because "a single C→G substitution that creates a novel splice donor site" cuts 55 letters out of its message.  When the team rebuilt the gene without that one letter, it lost its effect on brain stem cells:  "a single nucleotide substitution underlies the specific properties of ARHGAP11B".¹⁵

The NOTCH2NL genes carry a cost.  They mark the edges of a stretch of chromosome 1 that is often duplicated or deleted, "where duplications are associated with macrocephaly and autism, and deletions with microcephaly and schizophrenia."³⁵

None of these genes is a language gene.  They act on how many neurons the cortex makes and how its connections mature, which language needs, as do music, mathematics and planning.  And ARHGAP11B's human form was shared by Neanderthals and Denisovans,¹⁵ as FOXP2's was shared by Neanderthals.  Young-earth writers read such sharing as evidence that Neanderthals were human.¹⁹  That is a fair reading of these two genes.  It is not a test of every gene, and this article did not survey the others.

Part 5: What the gap is taken to mean

Chomsky and Berwick: one step, recently

Noam Chomsky, the founder of modern linguistics, and the computer scientist Robert Berwick set out their view in Why Only Us (2016).  They define the core of language, its "Basic Property", as "a finite computational system yielding an infinity of expressions, each of which has a definite interpretation in semantic-pragmatic and sensorimotor systems (informally, thought and sound)."¹⁶  The engine of that system is one operation, Merge, which takes two items and joins them into a new unit that can itself be joined again.  "In its simplest terms, the Merge operation is just set formation."¹⁶

Because Merge either applies to its own output or does not, they argue it cannot have come in stages.  "Note that there is no room in this picture for any precursors to language—say a language-like system with only short sentences.  There is no rationale for positing such a system:  to go from seven-word sentences to the discrete infinity of human language requires emergence of the same recursive procedure as to go from zero to infinity, and there is of course no direct evidence for such “protolanguages.”"¹⁶  They place the change "some time before 80,000 years ago if we can judge from associated symbolic proxies", when "individuals in a small group of hominids in East Africa underwent a minor biological change that provided the operation Merge".¹⁶  They suggest it "may have arisen from something as straightforward as a slight rewiring of the brain".¹⁶  They doubt that Neanderthals had language and admit, "We don’t understand the genomic or neural basis for the Basic Property."¹⁶

This is an evolutionary account.  It predicts exactly what the ape experiments found:  apes can learn symbols, because concepts and signals are old, but not syntax, because Merge is new.

The gradualist reply

Pedro Tiago Martins and Cedric Boeckx answered in 2019.  They accept that Merge itself may be all-or-nothing.  They deny that the biology must be:  "we examine this argument in detail and show that the jump from the atomicity of Merge to a single-mutation scenario is not valid".¹⁷  The formal simplicity of an operation, they argue, says nothing about how many genetic steps built the brain that runs it.  Berwick and Chomsky replied in the same issue that the critique "collapses" and rests "on a serious misunderstanding of binary set formation as well as formal language theory".  They also agreed that "there were multiple steps that preceded the appearance of Merge", while noting that neither side had proposed what those steps were.³⁶

The dispute is live, and neither side has data on the genes.  In 2014 Chomsky, Berwick, Lewontin, Tattersall, Hauser and three others reviewed the whole field and concluded that "the richness of ideas is accompanied by a poverty of evidence, with essentially no explanation of how and why our linguistic computations and representations evolved."²  On genes in particular:  "we lack a connect-the-dots account of any gene to language phenotype."²  And:  "Until such evidence is brought forward, understanding of language evolution will remain one of the great mysteries of our species."²

The creationist and design reading

Creationist and intelligent-design writers make three related claims.  Each is stated here in its own words.

Language is a designed system.  Fred Field, writing in the Journal of Creation in 2008, argued that "Attempts to link language and biological evolution have encountered impassible problems but the evidence points to a designed system."  On Kanzi he wrote that "The bonobo’s abilities seem closer to that of humans, but the gap is still formidable, with no evidence of continuity."  His conclusion:  "When a structure functions as an information processing ‘machine’ with such astounding speed and efficiency as that intrinsic to language, the terms intelligent design certainly seems appropriate:  life from life, intelligence from intelligence, information from information."³⁷

Language is all or nothing.  Brian Thomas of the Institute for Creation Research wrote in 2010 that talking "depends on so many perfectly placed characteristics--both material and immaterial--that language must be a gift from God."  He argued that vocal anatomy without concepts, or concepts without anatomy, "would surely be selected out of a population", and that "Removal of any of the many core material or immaterial features would sabotage language."²¹  This is the same structure as Berwick and Chomsky's argument, applied to the whole system rather than to Merge alone.

Language marks the image of God.  Elizabeth Mitchell of Answers in Genesis wrote in 2014 that "The uniquely human ability to understand and produce language is just one more reminder that we alone are created in the image of God ( Genesis 1:26–27 )".³⁸  On FOXP2 she wrote, "If anything, we should be impressed that God designed this important switch to fulfill many roles so perfectly in us and in so many animals."³⁸

The Discovery Institute has pressed the same point through the mainstream literature.  Casey Luskin reported Hauser and colleagues' 2014 paper under the headline "Leading Evolutionary Scientists Admit We Have No Evolutionary Explanation of Human Language".³⁹  When Atkinson's 2018 study removed the FOXP2 sweep, David Klinghoffer wrote, "It’s good to be right, as well as several years ahead of the curve."⁴⁰  On the facts reported, both were accurate.

How strong is the design inference?

The gap argument has two parts, and they have different strengths.

The first part is that the gap exists.  This is well supported.  Every ape project, read from its own data, stops short of syntax, and the leading secular researchers agree that animal studies give "virtually no relevant parallels to human linguistic communication, and none to the underlying biological capacity".²  The creationist writers above were right about this before the FOXP2 sweep collapsed, and right where many popular science accounts were wrong.

The second part is that the gap shows design.  This rests on the claim that no natural path could cross it.  Here the argument is weaker than its writers present it.  Berwick and Chomsky, who are not creationists, predict a gap of exactly this kind from a single biological change.  A gap is what both a sudden natural change and a special creation predict, so the gap alone cannot decide between them.  That no one has found the genes is true today, but it is a statement about present knowledge, and the human-specific genes in Part 4 show the search is not empty.  The design argument is strongest when it is stated as what it is:  an inference to the best explanation of a capacity that, on present evidence, has no known precursor, no known genetic basis and no known path.

Part 6: Where the gap argument is abused

Both sides have stretched this evidence.  A reader checking either side should know where.

Claims that apes learned language

1. Koko's vocabulary and conversations. The 1981 claim of "a vocabulary of more than 600 words" and a gorilla "speculating on death"²² and the later "over 1000 signs"⁶ were never backed by a released corpus. Patterson's own average utterance figure of about 2.7 signs,²² and Terrace's finding that in the one film available every Koko sign was first signed by her teacher,¹ are the data. The claims went far beyond them.
2. Washoe compared with children. The Gardners wrote that "The most significant results of Project Washoe were those based on comparisons between Washoe and children, as . . . in the use of order in early sentences", and Patterson that "language is no longer the exclusive domain of man."¹ Yet the Gardners reported almost nothing about sign order,¹ and Savage-Rumbaugh, who shared their hopes for ape language, said they never answered the imitation critique "straightforwardly".⁴
3. "Grammar" for Kanzi. Savage-Rumbaugh herself withdrew it.²⁶

Claims that FOXP2 explained language

1. The 2002 sweep and the spread of modern humans. Enard's team linked the sweep to a model in which modern humans spread because of "a more-proficient spoken language".⁸ Neanderthals had the same changes,⁹ and the sweep was a sampling artefact.¹⁰
2. A recent date. Francis Collins, in The Language of God (2006), repeated the then-current date, that the two changes occurred "apparently as recently as a hundred thousand years ago", and gave the hypothesis that they "may have in some way contributed to the development of language in human beings."⁴¹ The date did not survive Krause 2007. Collins's wording, though, was hedged. The Discovery Institute later described his account as presenting FOXP2 as "something of a miracle mutation that could have caused human language to develop".³⁹ That overstates what Collins wrote.

Claims from the creationist side

1. "Indicator gene." Peer Terborg and Royal Truman wrote in 2008 that FOXP2 "can thus be used as an indicator gene , a genetic tool to distinguish between humans, primates and other species", and that "The FOXP2 genes found in Neandertals therefore show that they were Homo sapiens ."¹⁹ One gene cannot do this. Mice differ from humans at only three of its amino acids,⁸ and a two-letter difference marks a lineage, not a species. Their conclusion, that Neanderthals were human, may well be right. FOXP2 cannot carry it alone.
2. "Non-existent in non-human creatures." Carl Wieland wrote in 2003 that "these language areas are unique to humans, and non-existent in non-human creatures", and that "An ape’s larynx simply does not have that capability."²⁰ Apes and monkeys have counterparts of both language areas,¹³ and the limit on primate speech is in the brain, not the throat.²⁷ His conclusion, "talking apes do not exist", stands.
3. "In humans only, a hyoid bone." Brian Thomas wrote in 2010 that "In humans only, a hyoid bone, specially shaped cartilage parts, muscles, and connective tissues suspend vocal cords in the throat".²¹ Other primates also have a hyoid bone and vocal folds, and Fitch's x-rays show that a monkey's vocal tract could shape intelligible speech if its brain could drive it.²⁷ Read generously, Thomas meant the human arrangement. As written it is false.
4. "Non-evolution." David DeWitt's 2006 Answers in Genesis article was titled "FOXP2 and the Non-Evolution of Human Language". Its science was sound and early: "Although the gene is clearly linked to language production, one cannot designate it as the “language gene.” An analogy would be calling the mutation that causes muscular dystrophy (and an inability to walk) a “walking gene.”"¹⁸ It also said, correctly, that "FOXP2 is not sufficient for language".¹⁸ But nothing about FOXP2 shows that language did not evolve. The pricing in Part 3 shows the two changes are easy for mutation to supply.

Objections and responses

1. "Linguists keep moving the goalposts. Apes learned words, so they learned language." The test was set by the ape researchers themselves: could an ape create a sentence? Terrace's team at first "grew more and more confident that Nim’s utterances displayed syntax",²⁶ and their own videotape showed they did not.¹ Savage-Rumbaugh withdrew "grammar" for Kanzi.²⁶ Yang's 2013 test is statistical, needs no one's judgment, and is passed by two-year-olds.⁵ Savage-Rumbaugh does have a point that children's early stages have been read generously and apes' harshly: she called it "a double standard".²⁶ That is fair for single words. It does not touch the finding that apes' combinations do not grow in length or structure as every child's do.
2. "Kanzi understood sentences as well as a two-year-old. That is a difference of degree, not kind." For simple sentences, this is a real point for continuity, and it should be conceded. Kanzi used word order, and on some sentence types he beat the child.⁴ The difference of kind appears where structure becomes hierarchical: Kanzi's worst result was the request that joins two nouns into one phrase.⁴ He reached a two-year-old's level at eight.
3. "FOXP2 shows that a language gene evolved recently, by selection." That was the 2002 reading. Neanderthals had the same changes,⁹ and the selection signal turned out to come from how the sample was chosen.¹⁰ FOXP2 helps control fine movements needed for speech. It is not a grammar gene.
4. "Neanderthals had human FOXP2, so they talked." Possibly, but not on that evidence. Fisher, who co-discovered the gene, warns that "the status of a single gene is not enough to resolve whether or not an ancient hominin could speak."²⁸ Young-earth writers who read Neanderthals as people have better grounds in their burials and their interbreeding with modern humans, argued in Neanderthals and Archaic Admixture.
5. "The gap proves that God created humans." The gap is consistent with that, and the Bible's claim that humans alone bear God's image predicts it. But Berwick and Chomsky predict the same gap from one natural change.¹⁶ A gap that two rival explanations both predict cannot decide between them. What weighs toward design is the absence, after decades of searching, of any known genetic path to syntax, which the leading secular researchers themselves concede.² That is evidence, but it is evidence from what has not been found.
6. "This is a god-of-the-gaps argument. Science will close the gap." It may. But "science will explain it" is also a claim about the future, not a finding. The honest position on both sides is the one Hauser and colleagues stated in 2014: a field rich in ideas and poor in evidence.² The genes found since then, such as ARHGAP11B and NOTCH2NL, act on the number of neurons, not on syntax.
7. "A single letter changed ARHGAP11B and enlarged the cortex. So the human brain evolved easily." The single C-to-G change is real and its effect in mice is real.¹⁵ It explains more neurons, not language. Neanderthals and Denisovans had it,¹⁵ so it does not mark the line between those who had syntax and those who did not, wherever that line falls.
8. "Creationists claim apes have no language areas, which is false." Some did. Wieland wrote that language areas are "non-existent in non-human creatures".²⁰ That is wrong, as Part 6 says. Others were careful. Field credited Kanzi's researchers for videos that "appear to show the bonobo’s abilities in interesting, candid and powerful ways."³⁷ DeWitt was right about FOXP2 years before the mainstream correction.¹⁸
9. "The mice with human FOXP2 did not talk, so FOXP2 has nothing to do with language." Too strong. A broken copy causes a severe speech disorder in people,⁷ the human form changes mouse brain circuits for motor learning,¹¹ and Atkinson's team, who removed the sweep, did "not dispute" its role in language production.¹⁰ The gene matters for speech. It does not explain grammar.

Issues and unanswered questions

1. The Gardners' 1969 paper was not read. It is behind the Science paywall and no open copy was found. Washoe's figures here come from the Gardners' own 1989 volume and from Terrace's and Patterson's reports of the 1969 results.
2. No corpus of Koko's signing exists in public. The claims for her cannot be tested, only set beside the few measures Patterson published.
3. Some brain papers were read only in part. Rilling 2008 and Charrier 2012 were read from their abstracts. The 2020 study of the arcuate fasciculus in macaques and chimpanzees was read only through Thomas's quotations of it.
4. The NOTCH2NL evidence favours common descent. Fiddes and colleagues report that NOTCH2NL arose by a partial copy of NOTCH2 before the common ancestor of humans, chimpanzees and gorillas, and that chimpanzees carry eight broken copies.³⁵ Shared broken copies are what common descent predicts. No young-earth treatment of these genes was found; the search was not exhaustive. The general argument is in Humans and Chimpanzees.
5. Other genes that differ between modern humans and Neanderthals were not surveyed. This article says only that FOXP2 and ARHGAP11B were shared.
6. The FOXP2 pricing is rough. It assumes Enard's 130-million-year figure, a steady background rate and a Poisson count, and the comparison with 20,000 genes assumes they behave like FOXP2. Its point is the order of magnitude: the jump is unusual for one gene and expected somewhere in a genome.
7. Whether every human language embeds clauses inside clauses is disputed. Daniel Everett has argued that Pirahã, spoken in the Amazon, does not, and other linguists dispute his analysis. This article did not read that literature. Its claims rest on apes' failure to show phrase structure of any depth, not on recursion in every language.
8. No one has a gene-to-syntax account. This is open on every side. If one is found, the design inference in Part 5 weakens. If decades more of genome work find none, it strengthens.

Verdict

The gap between human and ape language is real and large.  Apes learn a hundred or more symbols, ask for things with them and, in Kanzi's case, understand simple spoken sentences, including word order, at the level of a two-year-old.  None has produced, or shown that it understands, the nested phrase structure of human grammar.  Nim's utterances stayed 1.1 to 1.6 signs long while his vocabulary more than doubled,¹ and a statistical test of his corpus found no grammar.⁵  The loudest claims of ape language, for Koko, were never put to such a test.

FOXP2 is a gene needed for normal speech and for learning fine movement sequences in mice and birds.  It is not a language gene.  Its two human changes were shared by Neanderthals, and the claim that they swept through modern humans recently did not survive larger data.  The human brain differs from the ape brain in the reach of its language pathway and in a few genes that change how many neurons the cortex makes.  None of these explains syntax.

On the facts of the gap, creationist and design writers have been mostly right, and earlier than many in the mainstream.  Some of their claims went too far:  FOXP2 does not prove Neanderthals were Homo sapiens, apes do have counterparts of the language areas, and nothing about FOXP2 shows that language did not evolve.  The gap fits the biblical teaching that humans alone bear God's image.  It does not decide between creation and a sudden natural change, since Chomsky's evolutionary account predicts the same gap.  What it leaves standing is a capacity with no known precursor in any animal, no known genetic basis and no known path, and that is where the argument for design rests.

Sources