Marc Okrand, the creator of Klingon language for the Star Trek TV series, being carried by two men in Klingon costumes. Retrieved from: https://www.cnn.com/style/article/star-trek-klingon-marc-okrand/index.html
If you’ve been on twitter lately then you’ll know that I’ve been at the center of a controversial event. A bioRxiv preprint[1] was posted from the Fedorenko research group at MIT, lead author Saima Malik-Moraleda, concerning a study examining brain activations to five ConLangs (constructed languages): Esperanto, Klingon (from Star Trek), Na’vi (from Avatar), and High Valyrian and Dothraki (from Game of Thrones). Another colleague had tweeted about the posting of the article, which I then proceeded to take a look at, and then I quote-tweeted, saying: “This is "cool" work that doesn't make meaningful scientific contribution”, followed up by some explanation and examples to help motivate my opinion. Needless to say, a large amount of people took offense at my tweet, primarily concerning its tone. Many people thought that I was being rude, mean, or arrogant, claiming to speak on behalf of the entire field.
I acknowledge that the language I used could be easily misinterpreted as rude or mean, and was harsher than it needed to be – I would phrase it differently now. I have never claimed to speak on behalf of an entire field. I was speaking on behalf of myself, an individual with significant knowledge of the relevant domains (neurobiology, linguistics, psycholinguistics, fMRI), but not on behalf of a field. I’ve had cordial communications with Saima in which we have discussed this issue. Let me emphasize that my tweet was in no way intended to be mean-spirited. My tweet was intended to address the work of the authors, not the authors themselves. In the light, I have tremendous respect for Saima for responding to the substance of my critique.
Let me provide some background to my phrasing. I am from a school of thought in which a paper is discussed and evaluated in the following way: the primary aims, and the logic of achieving those aims, are ascertained, and then discussed. All of this is before even considering the empirical data and analyses. If there are fatal problems, then the empirical data are irrelevant to the claims. So, when I went through this paper, in my opinion, there were fatal problems with the main claims and the motivating logic, and that it was incapable of achieving its goal as articulated. This is why I said the work “doesn’t make meaningful scientific contribution” – that the paper did not deliver on its scientific promises. I did not intend to mean that the paper is useless, or that the authors’ work has no meaning.
The substantive issues
Here is the abstract of the paper, which will help frame my response [emphasis added]:
“What constitutes a language? Natural languages share some features with other domains: from math, to music, to gesture. However, the brain mechanisms that process linguistic input are highly specialized, showing little or no response to diverse non-linguistic tasks. Here, we examine constructed languages (conlangs) to ask whether they draw on the same neural mechanisms as natural languages, or whether they instead pattern with domains like math and logic. Using individual-subject fMRI analyses, we show that understanding conlangs recruits the same brain areas as natural language comprehension. This result holds for Esperanto (n=19 speakers)— created to resemble natural languages—and fictional conlangs (Klingon (n=10), Na’vi (n=9), High Valyrian (n=3), and Dothraki (n=3)), created to differ from natural languages, and suggests that conlangs and natural languages share critical features and that the notable differences between conlangs and natural language are not consequential for the cognitive and neural mechanisms that they engage.”
Here is the flaw: ConLangs such as Klingon or Dothraki, as learned by actual humans and assessed in any feasible functional neuroimaging experiment, are no different in any relevant way from other languages. While these languages may sound strange or feel alien to speakers of English or other Western languages, I believe that there are no compelling reasons why the brain activations for e.g. an English speaker proficient in Dothraki listening to Dothraki should differ in gross respects from that same English speaker listening to other languages in which they might be proficient. This is because ConLangs are based on the grammars of natural languages. In this light, there is no reason to expect otherwise - it would be like asking whether Indonesian activates language-related regions. Many functional neuroimaging studies have been published over the years on second language processing showing that second language processing roughly activates the same set of areas as native language processing (although perhaps with some differences in extent and magnitude within this network, or the recruitment of executive function regions, a question of much debate)[2-4]. There are always differences among languages – each natural language has its own unique characteristics and origins – the question is whether there is a specific argument to be made for why those differences would produce fundamentally different patterns of activation.
The article points to the idea that the typical processing efficiency considerations that have been argued to be present in natural languages might not be be present in ConLangs (crucially, the study was testing ConLangs that have been learned and used by groups of speakers, and not other ConLangs that have proved difficult to learn). So yes - e.g., Klingon might be harder to learn and use for a native English speaker than French, because e.g. its grammar might be more complex than French, or its words harder to pronounce. In which case, we might expect greater activation in language areas and potentially some additional activation in executive function areas. In fact, Supplementary Figure 5 seems to show that the dice coefficient between ConLangs and spatial working memory is greater than that between Natural Languages and spatial working memory (however, no numbers are given and no statistical tests comparing these dice coefficients are made). However, the paper does not ask this question (nor does it perform any statistical tests to answer it) - it asks “which network” ConLangs activate; hence the main result being described as the idea that listening to conlangs “recruits” and “engage[s]” the “same brain areas” as natural languages do.
I fundamentally believe that science gains quite a bit from open discussion and debate. In fact, I think that the field of neurobiology of language would benefit from a lot more discussion and debate than we currently have. I will make responses point-by-point, and in a different order that will allow a greater focus on the deeper points later. Saima’s text is bold and in green font.
“…the intuition that conlangs are like natural languages is NOT shared by most people. See here: [reference to previous twitter poll]”
I don’t see the relevance. First, meaningful surveys need to be conducted in a rigorous, scientific manner like other forms of data collection, and not ad hoc twitter polls. More importantly, scientific hypotheses are not a matter of crowdsourcing. If *only* the authors of the study, and nobody else, thought a given outcome might be interesting or surprising, it would be compelling given sound argumentation.
“…one could have hypothesized that generativeness (or another factor) might be critical in determining which conlangs pattern up with nat langs, in which case HighVal [High Valyrian] and Doth [Dothraki] shld not have since it’s v hard for their speakers to come up with new sentences”
There are an infinite number of other factors that one could hypothesize given a comparison between two things. The question is whether these are relevant factors; the relevance must be motivated. I don’t see why the fact that it’s difficult to “come up with new sentences” in Dothraki and High Valyrian means that people proficient in these languages wouldn’t activate the language network, and this is not spelled out in the paper.
The broader issue illustrated here is that the paper offers no significant exploration of what people are doing when they learn a ConLang. In fact, there was almost no discussion of second language acquisition, despite this being the major phenomenon under investigation. What do people *do* when they learn a second language? Focusing on the lexical level, at early stages of learning, it is likely (especially for ConLangs) that people mentally translate the word into their native language. At later stages of learning, they may not need to do this, directly relating the perceived form to the meaning of a word, as we do for native languages. A wealth of research has shown that perception of isolated words often activates a broad set of language-related regions relative to a control stimulus[5-9]. In this study, the critical contrasts were clear audio sentences vs. acoustically degraded versions of the sentences. Thus, it is likely that if the participants in this study were able to at least understand the meaning of a decent amount of the words presented, regardless of their knowledge of any grammar beyond the word level at all or their ability to produce novel sentences of the language, that this would produce activation within language-related regions.
“…your assertion that all conlangs are based on natural languages is in fact incorrect. Klingon was created to differ from natural languages as much as possible. Marc Okrand, its creator, did not think it would pattern up with nat langs because of it. In fact, some Klingon sentences exist because the Star Trek producers reversed the audio track of the intended sentence just because ‘it sounded better’ that way. This is not how natural languages develop.”
This is a central issue, and was the crux of my critique. The paper claims that these ConLangs are “created to differ from natural languages”, yet there is not a single mention or explanation of any grammatical properties in any of the languages tested in this study and how they might be similar or different to natural languages. There are no glosses– that is, element-by-element grammatical descriptions – for the stimuli used in the study. Thus, not only are grammatical structures not discussed at all in this paper, it is impossible for a reader not already versed in these languages to ascertain their grammatical properties for themself.
When reading the preprint, I immediately thought of an older fMRI experiment conducted by Musso, Moro and colleagues[10] which does something like what is hinted at in this study. They began with a theory of certain universal grammatical properties of human languages, then they constructed languages which violated those principles (e.g., violations of the null subject parameter), and then looked at brain activations while people progressed in proficiency. They compared this to brain activations while people learned natural languages conforming to these principles that they simply had not yet been exposed to. They claimed differential patterns of activation in language-related brain regions for the two kinds of languages. Critically, they explained what these grammatical principles were and how they differed between the real and “unreal” (constructed) languages. There is none of this sort of thing in the present paper.
The fact that the structures of ConLangs are shared with those found in natural languages is mentioned in a short paragraph at the top of page 11 of the manuscript (“Of course, it is important to remember that conlangs are created by proficient users of natural language and are therefore typically modeled, in some way, on natural languages…”). However, while this point is briefly discussed, it is not seriously grappled with and is fatal to the logic of the paper.
As far as I understand, the reversed audio of some Klingon sentences in the Star Trek show/movies is not a general grammatical property of the language (there are exceptional cases in every language), and seems like a non-sequitur. It does however call attention to the unusual trajectory of development of the language. In general, the point was made that Klingon was created in certain ways to deviate from some crosslinguistic patterns. Note however that this is applies to *only* Klingon - it is not clear to me why three other ConLangs, Dothraki, High Valyrian, and Na’vi, are also described as “created to differ from natural languages”. In what ways? It is insufficient to say that they “sound weird” or are “alien” languages – I could say the same about many non-Western languages, including the very ones that these ConLangs were based on (e.g., Inuktitut, Turkish, Russian, Arabic, etc.). As far as I am aware, there is no basis for claiming that these three other languages were created to be substantively different from natural languages.
However, Okrand, the creator of Klingon, was incredibly thoughtful when creating that particular language. The ways in which Klingon is fundamentally different from the universal principles of human language primarily concern the phonemic inventory (given this explanation by Okrand). For example, “The combination of an aspirated voiceless alveolar plosive /tʰ/ and a voiced retroflex plosive /ɖ/ is particularly unusual”, meaning that the particular phonological pattern of this language violates the kinds of patterns typically seen in human languages. More generally, if a language has Sound A, it typically has Sound B, but not Sound C. Okrand apparently created Klingon to have Sound A and Sound C, but not Sound B, violating the pattern.
While this is an interesting creative process, there is no explanation of why this property would matter at all for this experiment. There is nothing that makes vocabulary unlearnable in such a system; one could imagine that it might make generating new words difficult in some way. However, this wouldn’t prevent their brains activating to known Klingon words as described above. In addition, second language acquisition often involves limited creation of new words by people learning the language as adults; for example, college students learning a second language are typically not going around inventing new vocabulary. Beyond the phonetic/phonemic structure, there are some unusual grammatical choices - for example, Klingon has Object-Verb-Subject word order, which certainly is rare across the world’s languages, but it would not seem “alien” to speakers of Hixkaryana, an indigenous language of Brazil, which does use this word order. Thus, while Klingon may be unusual from a cross-linguistic perspective, it does not fundamentally contain anything not found in other languages, at least at the grammatical level.
One stated motivation for why brain activation to these ConLangs might not have activated language-related brain regions is the fact that these languages are artificial and not natural, and thereby might be “harder” in some way. However, we know that second languages are often harder than native languages (ask an undergraduate class learning any language). Yet we know that they activate regions implicated in native language processing, and maybe some additional executive-function-related areas. Why would ConLangs be different? Again, what is missing from this paper is any exploration of the psychological experience of the learner of these languages rather than vague speculations. As mentioned before, no tests of this hypothesis were actually made, and there were no apparent attempts to select sentences that might induce processing difficulty and compare them to analogous structures in natural languages.
In fact, this is another broad point worth raising. If one is interested in whether Klingon is “harder” than learning, say, French, this kind of fMRI experiment is a very bad way to go about it. One way would be to measure and report on the structural properties of the grammar. Another way might be to collect more measurements of processing load/difficulty. All of these things would be great! But it's highly indirect and fraught to infer this sort of thing from an fMRI experiment, because reverse inference is an extremely tricky endeavor[11].
Throughout this whole engagement, I enjoyed learning about the process of how Klingon was created, and the linguistic sophistication of its creator. Okrand put very serious and impressive thought into this particular ConLang. In this way, I get that it is very entertaining for those that are into this sort of thing to see pictures of brain activations to these ConLangs. My critique here is not about why some people might find that to be interesting and exciting, but rather about the scientific goals and claims of the paper.
References
1. Constructed languages are processed by the same brain mechanisms as natural languages
Saima Malik-Moraleda, Maya Taliaferro, Steve Shannon, Niharika Jhingan, Sara Swords, David J. Peterson, Paul Frommer, Marc Okrand, Jessie Sams, Ramsey Cardwell, Cassie Freeman, Evelina Fedorenko bioRxiv 2023.07.28.550667; doi: https://doi.org/10.1101/2023.07.28.550667
2. Jasinska, K. K., & Petitto, L. A. (2013). How age of bilingual exposure can change the neural systems for language in the developing brain: A functional near infrared spectroscopy investigation of syntactic processing in monolingual and bilingual children. Developmental cognitive neuroscience, 6, 87-101.
3. Perani, D., & Abutalebi, J. (2005). The neural basis of first and second language processing. Current opinion in neurobiology, 15(2), 202-206.
4. Sebastian, R., Laird, A. R., & Kiran, S. (2011). Meta-analysis of the neural representation of first language and second language. Applied psycholinguistics, 32(4), 799-819.
5. Binder, J. R., Rao, S. M., Hammeke, T. A., Yetkin, F. Z., Jesmanowicz, A., Bandettini, P. A., ... & Hyde, J. S. (1994). Functional magnetic resonance imaging of human auditory cortex. Annals of neurology, 35(6), 662-672.
6. Binder, J. R., Rao, S. M., Hammeke, T. A., Frost, J. A., Bandettini, P. A., Jesmanowicz, A., & Hyde, J. S. (1995). Lateralized human brain language systems demonstrated by task subtraction functional magnetic resonance imaging. Archives of Neurology, 52(6), 593-601.
7. Binder, J. R., Frost, J. A., Hammeke, T. A., Cox, R. W., Rao, S. M., & Prieto, T. (1997). Human brain language areas identified by functional magnetic resonance imaging. Journal of Neuroscience, 17(1), 353-362.
8. Fedorenko, E., Hsieh, P. J., Nieto-Castañón, A., Whitfield-Gabrieli, S., & Kanwisher, N. (2010). New method for fMRI investigations of language: defining ROIs functionally in individual subjects. Journal of neurophysiology, 104(2), 1177-1194.
9. Matchin, W., Hammerly, C., & Lau, E. (2017). The role of the IFG and pSTS in syntactic prediction: Evidence from a parametric study of hierarchical structure in fMRI. Cortex, 88, 106-123.
10. Musso, M., Moro, A., Glauche, V., Rijntjes, M., Reichenbach, J., Büchel, C., & Weiller, C. (2003). Broca's area and the language instinct. Nature neuroscience, 6(7), 774-781.
11. Poldrack, R. A. (2006). Can cognitive processes be inferred from neuroimaging data?. Trends in cognitive sciences, 10(2), 59-63.