Category: Presentation of self in everyday life

“Segmented languages”

In a recent paper (Gorman & Sproat 2023), we complain about conflation of writing systems with the languages they are used to write, highlighting the nonsense underlying common expressions like “right-to-left language”, “syllabic language” or “ideographic” language found in the literature. Thus we were surprised to find the following:

Four segmented languages (Mandarin, Japanese, Korean and Thai) report character error rate (CER), instead of WER… (Gemini Team 2024:18)

Since the most salient feature of the writing systems used to write Mandarin, Japanese, Korean, and Thai is the absence of segmentation information (e.g., whitespace used to indicate word boundaries), presumably the authors mean to say that the data they are using has already been pre-segmented (by some unspecified means). But this is not a property of these languages, but rather of the available data.

[h/t: Richard Sproat]

References

Gemini Team. 2023. Gemini: A family of highly capable multimodal models. arXiv preprint 2312.11805. URL: https://arxiv.org/abs/2312.11805.

Gorman, K. and Sproat, R.. 2023. Myths about writing systems in speech & language technology. In Proceedings of the Workshop on Computation and Written Language, pages 1-5.

Yet more on the Pirahã debate

I just read a draft of Geoff Pullum’s paper on the Pirahã controversy, presented at a workshop of the recent LSA meeting.

It’s not a particularly interesting paper to me, since it has nothing to say about the conflicting data claims at the center of the controversy. No one has ever given an explanation of how one might integrate the evidence for clausal embedding in Everett 1986 (etc.) with the writings of Everett from 2005 onward. These two Everetts are in mortal conflict. Everett (1986), for example gives examples of embedded clauses, Everett (2005) denies that the language has clausal embedding, and Everett (2009), faced with the contradiction, has decided to gloss this same example (Nevins et al. 2009, ex. 13, reproduced from Everett 1986, ex. 232) as two sentences, with no argument provided for why earlier Everett was wrong. While one ought not to reason from one’s own limited imagination, it’s hard for me to fathom anything other than incompetence in 1987 or dishonesty 2005-present. Either way, it suggests that additional attention is probably needed on other specific claims about this language, such as the presence of rare phonetic elements (Everett 1988a) and the presence of ternary metrical feet (Everett 1988b); and on these topics there is far less room for creative hermeneutics.

If people have been nasty to Everett—and this seems to be the real complaint from Pullum—it’s because the whole thing stinks to high heaven; it’s a shame Pullum can’t smell the bullshit.

References

Everett, D. L. 1986. Pirahã. In Handbook of Amazonian Languages, vol. 1, D. C. Derbyshire and G. K. Pullum (ed.), pages 200-326. Mouton de Gruyter.
Everett, D. L. 1988a. Phonetic rarities in Pirahã. Journal of the International Phonetic Association 12: 94-96.
Everett, D. L. 1988b. On metrical constituent structure in Pirahã. Natural Language & Linguistic Theory 6: 207-246.
Everett, D. L. 2005. Cultural constraints on grammar and cognition in Pirahã: another look at the design features of human language. Current Anthropology 46: 621-646.
Everett, D. L. 2009. Pirahã culture and grammar: a response to some criticisms. Language 85: 405-442.
Nevins, A., Pesetsky, D., and Rodrigues, C. 2009. Pirahã exceptionality: a reassessment. Language 85: 355-404.

Alt-lingfluencers

It’s really none of my business whether or not a linguist decides to leave the field. Several people I consider friends have, and while I miss seeing them at conferences, none of them were close collaborators. Reasonable people can disagree about just how noble it is to be a professor (I think it is, or can be, but it’s not a major part of my self-worth), and I certainly understand why one might prefer a job in the private sector. At the same time, I think linguists wildly overestimate how easy it is to get rewarding, lucrative work in the private sector, and also overestimate how difficult that work can be on a day-to-day basis.  (Private sector work, like virtually everything else in the West, has gotten substantially worse—more socially alienating, more morally compromising—in the last ten years.)

In this context, I am particularly troubled by the rise of a small class of “alt-ac” ex-linguist influencers. I realize there is a market for advice on how to transition careers, and there are certainly honest people working in this space. (For instance, my department periodically invites graduates from our program to talk about their private sector jobs.) But what the worst of the alt-lingfluencers do in actuality is farm for engagement and prosecute grievances from their time in the field. If they were truly happy with their career transitions, they simply wouldn’t care enough—let alone have the time—to post about their obsessions for hours every day. These alt-lingfluencers were bathed in privilege when they were working linguists, so to see them harangue against the field is a bit like listening to a lottery winner telling you not to play. These are deeply unhappy people, and unless you know them well enough to check in on their well-being from time to time, you should pay them no mind. You’d be doing them a favor, in the end. Narcissism is a disease: get well soon.

Lottery winners

It is commonplace to compare the act of securing a permanent faculty position in linguistics to winning the lottery. I think this is mostly unfair. There are fewer jobs than interested applicants, but the demand is higher— and the supply lower—than students these days suppose. And my junior faculty colleagues mostly got to where they are by years of dedicated, focused work. Because there are a lot of pitfalls on the path to the tenure track, their egos are often a lot smaller than one might suppose.

I wonder if the lottery ticket metaphor might be better applied to graduate trainees in linguistics finding work in the tech sector. I have held both types of positions, and I think I had to work harder to get into tech than to get back into the academy. Some of the “alt-ac influencers” in our field—the ones who ended up in tech, at least—had all the privileges in the world, including some reasonably prestigious teaching positions, before they made the jump. Being able to stay and work in the US—where the vast majority of this kind of work is—requires a sort of luck too, particularly when you reject the idea that “being American” is some kind of default. And finally demand for linguist labor in the tech sector varies enormously from quarter to quarter, meaning that some people are going to get lucky and others won’t.

Citation practices

In a previous post I talked about an exception to the general rule that you should expand acronyms: sometimes what the acronym expands to is a clear joke made up after the fact. This is an instance of a more general principle: you should provide, via citations, information the reader needs to know or stands to benefit from. To that point, nobody has ever really cared about the mere fact that you “used R (R Core Team 2021)”. It’s usually not relevant. R is one of hundreds of Turing-complete programming environments, and most of the things it can do can be done in any other language. Your work almost surely can be replicated in other environments. It might be interesting to mention this if a major point of your paper is that wrote, say, a new open-source software package for R: there the reader needs to know what platform this library targets. But otherwise it’s just cruft.

Robot autopsies

I don’t really understand the exhuberance for studying whether neural networks know syntax. I have a lot to say about this issue—I’ll return to it later—but for today I’d like to briefly discuss this passage from a recent(ish) paper by Baroni (2022). The author expresses great surprise that few formal linguists have cited a particular paper (Linzen et al. 2016) about the ability of neural networks to learn long-distance agreement phenomena. (To be fair, Baroni is not a coauthor of said paper.) He then continues:

While it is possible that deep nets are relying on a completely different approach to language processing than the one encoded in human linguistic competence, theoretical linguists should investigate what are the building blocks making these systems so effective: if not for other reasons, at least in order to explain why a model that is supposedly encoding completely different priors than those programmed into the human brain should be so good at handling tasks, such as translating from a language into another, that should presuppose sophisticated linguistic knowledge. (Baroni 2022: 11).

I think this passage is a useful stepping-off point for what I think. I want to be clear: I am not “picking on” Baroni, who is probably far more senior to and certainly better known than me anyways; this is just a particularly clearly written claim, and I just happen to disagree.

Baroni says it is “possible that deep nets are relying on a completely different approach to language processing…” than humans; I’d say it’s basically certain that they are. We simply have no reason to think they might be using similar mechanisms since humans and neural networks don’t contain any of the same ingredients. Any similarities will naturally be analogies, not homologies.

Without a strong reason to think neural models and humans share some kind of cognitive homologies, there is no reason for theoretical linguists to investigate them; as artifacts of human culture they are no more in the domain of study for theoretical linguists than zebra finches, carburetors, or the perihelion of Mercury. 

It is not even clear how one ought to poke into the neural black box. Complex networks are mostly resistent to the kind of proof-theoretic techniques that mathematical linguists (witness the Delaware school or even just work by, say, Tesar) actually rely on, and most of the results are both negative and of minimal applicability: for instance, we know that there always exists a single-layer network large enough to encode, with arbitrary precision, any function a multi-layer network encodes, but we have no way to figure out how big is big enough for a given function.

Probing and other interpretative approaches exist, but have not yet proved themselves, and it is not clear that theoretical linguists have the relevant skills to push things forward anyways. Quality assurance, and adversarial data generation, is not exactly a high-status job; how can Baroni demand Cinque or Rizzi (to choose two of Baroni’s well-known countrymen) to put down their chalk and start doing free or poorly-paid QA for Microsoft?

Why should theoretical linguists of all people be charged with doing robot autopsies when the creators of the very same robots are alive and well? Either it’s easy and they’re refusing to do the work, or—and I suspect this is the case—it’s actually far beyond our current capabilities and that’s why little progress is being made.

I for one am glad that, for the time being, most linguists still have a little more self-respect. 

References

Baroni, M. 2022. On the proper role of linguistically oriented deep net analysis in linguistic theorising. In S. Lappin (ed). Algebraic Structures in Natural Language, pages 1-16. Taylor & Francis.
Linzen, T., Dupoux, E., and Goldberg, Y. 2016. Assessing the ability of LSTMs to learn syntax-sensitive dependencies. Transactions of the Association for Computational Linguistics 4: 521-535.

Isaacson and Lewis

It’s amusing to me that Walter Isaacson and Michael Lewis—who happened to go to the same elite private high school in New Orleans, just a few years apart—are finally having their oeuvres as favorable stenographers for the rich and powerful reassessed more or less simultaneously. Isaacson clearly met his match with Elon Musk, a deeply incurious abuser who gave Isaacson quite minimal access; Lewis does seem to be one of a handful of people who actually believed in that ethical altruism nonsense Sam Bankman-Fried was  cooking up. Good riddance, I say.

Use the minus sign for feature specifications

LaTeX has a dizzying number of options for different types of horizontal dash. The following are available:

  • A single - is a short dash appropriate for hyphenated compounds (like encoder-decoder).
  • A single dash in math mode,$-$, is a longer minus sign
  • A double -- is a longer “en-dash” appropriate for numerical ranges (like 3-5).
  • A triple --- is a long “em-dash” appropriate for interjections (like this—no, I mean like that).

My plea to linguists is to actually use math mode and the minus sign when they are writing binary features. If you want to turn this into a simple macro, you can please the following in your preamble:

\newcommand{feature}[2]{\ensuremath{#1}\textsc{#2}}

and then write \feature{-}{Back} for nicely formatted feature specifications.

Note that this issue has an exact parallel in Word and other WYSIWYG setups: there the issue is as simple as selecting the Unicode minus sign (U+2212) from the inventory of special characters (or just googling “Unicode minus sign” and copying and pasting what you find).