Recently in the sister blog

In the article, “Testing the role of convergence in language acquisition, with implications for creole genesis,” Marlyse Baptista et al. write:

The main objective of this paper is to test experimentally the role of convergence in language acquisition (second language acquisition specifically), with implications for creole genesis. . . . Our experiment is unique on two fronts as it is the first to use an artificial language to test the convergence hypothesis by making it observable, and it is also the first experimental study to clarify the notion of similarity by varying the levels and types of similarity that are expressed. We report an experiment with 94 English-speaking adults . . . A miniature artificial language was created that included morphological elements to express negation and pluralization. Participants were randomly assigned to one of three conditions: congruent (form and function of novel grammatical morphemes were highly similar to those in English), reversed (negative grammatical morpheme was highly similar to that of English plural, and plural grammatical morpheme was highly similar to that of English negation), and novel (form and function were highly dissimilar to those of English).

A miniature artificial language! Cool.

And here’s what the researchers found:

Participants in the congruent condition performed best, indicating that features that converge across form and function are learned most fully. More surprisingly, results showed that participants in the reversed condition acquired the language more readily than those in the novel condition, contrary to expectation.

It’s funny they find this result surprising. Speaking as an outsider to this field of research, based only on introspection and experience, I’d’ve thought that the novel condition would be more difficult than the reverse condition. To me, reversal’s pretty trivial; but an entirely new system, that sounds hard. For example, in our apartment we have a backwards clock (the gear is flipped so the hands go counter-clockwise), a clock with the minute and hour hands switched, and a 24-hour clock. I’ve never had any problem reading the backward clock—once you realize it’s reverse, reading it is automatic—but the other two clocks still give me difficulty and I have to consciously work it out each time I read them.

Sure, language is verbal and clock-reading is visual. Still, as Phil would say, I’m surprised that Baptista et al. are surprised that learning something reversed is easier than learning something new.

If the result really is a surprise, I’d like to see a replication study. Also some graphs of the data.

12 thoughts on “Recently in the sister blog

  1. It seems like the reversed condition in the experiment is closer to the clock with the minute and hour hands being switched rather than the clock running backwards. And, as you said, this intuitively seems harder (and this hardness would be predicted by many theories of learning as well).

      • At a café a week or two ago I overheard two people, probably college students, talking about a 24-hour analog clock. One, who had seen such a clock, thought it was great. The other was very confused about how the minute hand could work. The first person gave a confused explanation. The second finally seemed to conclude that there must be a separate dial for the minute hand. I think he couldn’t get around the fact that 60 is not divisible by 24. I managed not to bang my head against the table, and then I tried to focus more on coffee and work.

        • I have a 24-hour watch that I use when I travel overseas; if you search for [dolphin 24 hour watch] or similar you can see one. It has an outer bezel marked with even hours, and an inner face marked with odd hours. The inner face is fixed but the outer bezel can be rotated. I leave the inner one set to the time back home, and rotate the outer bezel to show the correct time in whatever time zone I’m in at the time. So if I want to know if it’s a good time to call my wife I can just read the time like normal; if I want to know the local time I have to look at the number the hour hand is pointing to.

          The thing about “60 not divisible by 24” is funny, and reminds me of a similar story. Way back when we were in high school, Andrew and I went to a pizza place with our friend Craig. We asked ‘how big is a large’, and were told “16 inches”. “OK, we’d like one large pizza please. Could you cut it into 12 pieces to make it easier to split three ways?” The waitress said “No, we can’t do that.” Why not? “Because 12 doesn’t go into 16.”

          She was serious.

  2. Coming from within the language acquisition field, it is indeed considered surprising that the reversed language would be more difficult than the novel. The reason is that it’s considered more difficult to put aside your native language bias; thus, structure that directly conflicts with the way your native language works is more difficult to acquire in a second language because you are not only learning something new but are actively suppressing what you expect based on your native language. Bayesian metaphors are used in language acquisition, where native language expectations are your prior, so I guess it’s kind of like finding data that directly conflicts with a strong prior. The posterior will not move too much from the prior, following exposure to the conflicting data; this is the reverse case. But if you don’t have a strong prior (the novel case), then you can base your posterior basically just on the data you’ve just been exposed to. I think that would be the explanation from Baptista et al.

    • I’ve heard this before but can’t understand: no matter what language is presented to me I have a language bias. That prior directly affects whatever language I learn. The idea that you can base your posterior ‘on the date you’ve just been exposed to’ reads to me like nonsense because we’re not tabula rasa. Example: an older woman put the hard cider 6 pack into a paper bag with minimal motion, opening the bag just with a tap, touching the top side with only a finger as she glided the carton in and down what was otherwise a tight fit. I asked her how many times she did that before it became natural: it’s a skill, it’s a language acquisition. She said a lot. She had obviously probably as a human put things inside other things, likely inside bags or boxes at some point over her life. That’s her prior, right? So she gets this new skill of putting cartons of beer and the like in bags efficiently. How do we escape this problem of her having a prior? That invokes Gödel, but I won’t go there. Instead I’ll just note that you would need to construct a ‘novel language’ to be acquired that either isn’t done by humans or which occurs in some distant remote form from that person’s experience, perhaps like putting an Amazonian native from the jungle in front of an accounting spreadsheet. By description, the ‘novel’ language involved stuff like plurals and thus I assume some recognition of and facility with stuff like plurals. That’s not like the spreadsheet. That’s pretty much how humans think and use language and sort screws and bolts and count how many cookies they have for dessert. To what extent is the prior actually isolated by such a ‘novel’ acquisition? Maybe as Andrew speculates the partially isolated prior actually makes reversal easier because you learn the task of putting the carton in the bag because you’ve put stuff inside stuff before. My guess is she could also remove stuff from a bag and might even be able to take a carton out of one bag and put it in another. I don’t mean to be snarky. I’ve had a rough day.

  3. The U.S. military uses the 24-hour system. I once asked my father, a 30-year veteran, how long it took him for that to seem natural. Two weeks, he said, but he knew plenty of old hands who never got used to it.

  4. As I read the “surprising” results, I wasn’t very surprised. I doubt I can explain this well – perhaps someone else can give the general rule I’m trying to get across – but my thoughts can be understood, I think, from an example I thought of while learning Chinese and Japanese.

    In Japanese, the verb and object are reversed in regard to their positioning in English. However, this is an easily understood change, expressed in a mere sentence, albeit unintuitive to native English speakers. Beginning learners don’t make the mistake of creating sentences in the form [subject] [object] [verb] – at least I’ve never witnessed such a mistake. This example is “reversed,” is it not?

    In Chinese, the copula changes based on the form of word following it. For example, [subject] [copula] [noun] uses “shi,” while [subject] [copula] [adjective] uses “hen” (not technically a correct explanation, but this is essentially how the language deals with our “be” copula in translating to English). Students, even into intermediate levels, make mistakes here, using “shi” when they should be using “hen.” I believe this example is “novel.”

    Mathematically, Andrew’s Bayesian explanation seems to explain this phenomenon. The reverse situation would be like flipping a prior – transposing the distribution with a single function. The novel situation would require something like a piecewise function or using an entirely different distribution. As the latter seems harder, should we not have hypothesized that the novel language situation would be harder for students to learn?

    • I’m not sure if the Chinese copula, when pitched against Japanese SOV order, is a very good example of a ‘novel’. The difference between ‘novel’ and ‘reverse’ is that in the reverse case, negative transfer from the L1, in this case English, is likely. English speakers do know what a copula is; under your analysis of Chinese syntax, there’s simply a distinction between two types of copulas, which seems to be a variation on something English already has. So if English speakers are accustomed to the one-copula system, they will still have to actively suppress the negative L1 transfer to deal with a two-copula (actually multiple-copula, if you count other degree words) system. Also, frequency is a huge factor here. One would expect that adjectival and nominal predicates are less common than verbal ones, and hence Japanese learners would be exposed to SOV sentences much more than Chinese learners would be exposed to two different copula types. (English speakers may have also had exposure to SOV through English poetry, so it’s not an entirely unfamiliar either.)

      Flipping a distribution is definitely a conceivable notion, but we don’t know a priori whether the human minds are capable of doing that w.r.t. language acquisition. My suspicion, in addition to the explanations proposed in the paper, is that there’s a difference between explicit and tacit learning; adult participants may be capable of explicitly formulating rules like ‘negative is new plural, plural is new negative’, but if they did not consciously search for patterns but simply implicitly learnt the language, they may have fared worse in the reverse condition. It would be interesting to see the study done on kids instead.

Leave a Reply to David Cancel reply

Your email address will not be published. Required fields are marked *