Brownian thought space

Cognitive science, mostly, but more a sometimes structured random walk about things.

My Photo
Location: Rochester, United States

Chronically curious モ..

Thursday, August 31, 2006

Gentium success

Finally, finally was able to use the lovely Gentium font for my thesis! Oh and also a hieroglyphics font, which is where the previous hieroglyphic comes from :) Update: For some reason, only the Chapter titles are in Gentium :(


Was wondering about writing systems. Take English- the units of text, the words we write, are not exactly the units of speech. Infact, stuff like cliticization and all is totally non-apparent. How did THAT happen? One possibility is that, while speech uses prosodic units, we actually perceive (recover) morphological units. That is, somehow we "see" word-strings, and NOT strings of prosodic constituents. Which is why phonological phenomena like the British /r/-epenthesis doesn't mislead people into believing there really is an /r/. Turns out, the Egyptian hieroglyphics were pretty much like speech: unless you know the words and the grammar and the meaning, its incredibly hard to segment these babies in any meaningful fashion. For the record, that's my version of my name in hieroglyphic (ps: the cartouche around the name indicates royalty)

Friday, August 25, 2006

Why animals take 1000 trials to learn

Sometimes, experiments with animals can show that they are capable of certain cognitive patterns that mark human language. For example, the work by Keith Kluender on stuff like the perceptual magnet effect or coarticulatory compensation. The recent and much talked about paper is in the line begun by Hauser et al (here), wherein non-humans are shown/not shown to be able to do some grammatical tasks. In this paper, Gentner &al show that starlings can do recursion, supposedly the bastion of human language (see this too). So, starlings can do recursion. But, some critic-people point out, that these starlings take thousands of trials to learn the task. Whereas college grads typically are exposed to a few tens of trials. Of course, logically, if the experiments are well done and sufficiently controlled, showing that starlings do recursion says that starlings do recursion, and the trials don't really come into it. To give a silly example, no amount of throwing pigs off terraces will EVER induce them to take flight. But then why do starlings take so long? Here's a possibility. In the film Cube, 7 people are stuck in a cube-based structure, and lots of deadly traps and dying and all happens. Basically, it takes about 80% of the film to figure out what the trick to solving the cubes is. People DIE in the process. So, the possibility is, that the starlings have (almost) everything needed to solve the task, except knowing what the task itself is. College students have a HUGE head-start here. Without instructions, college students might, if not face death actually, be in a spot. Because, as we have all been told, there are limitless possibilities. So presumably, the starlings have the symbolic representations, they're just going through all kinds of combinations to solve the problem. Probably on each day they try a certain combination; and the combinations keep getting better (= more food), so the task really is to find a set of cognitive combination that maximizes food. A prediction: if it's really this combinatorial story, then maybe they thing module X1 is irrelevant, but that happens to be only because X2-X4 are enough. Come the day X2-X4 asre absent, X1 becomes necessary. So, they would fail. Prediction for an observation: if this story is true, then on an indiviudal level, every once in a while the bird should fail for no apparent reason. So the learning curve won't be like that for a skill, which is gradual, but should be more.. unsmooth. With big dips and strange uprises.

Thursday, August 24, 2006

Fish on the wall

Tuesday, August 22, 2006

Thank you, Adium Duck

Finally, there is a software that does seriously good messaging over multiple thingies. Thank you, Adium Duck.

Wednesday, August 16, 2006

Spectral rotation

People use all kinds of controls for speech-like sounds. One of them is spectrally rotated speech, and I've always wondered wtf that is. So I figured, it must mean that the spectrogram is rotated, but how do you do THAT? So, methinks: consider the FFT; let the real part = power, and the imaginary part = phase. Clearly, we don't want to change phase info, but just swap around the power info. So that's what this little MATLAB script does :) And here's proof of the pudding: first, the normal spectrogram: And then the rotated version: frequencies between 0 and 4 kHz have been rotated (nb: the sound file was 16kHz, so the frequencies run from 0-8kHz; and so 4khz is approx halfway up) :)

Friday, August 11, 2006

Language vs psychology

Back again to the question of what does psychology have to do with language. The problem is this: if we want to study the psychological aspects of language, the only thing we can measure is speech (or sign-language). But, speech is NOT made up of syntactic units, but of prosodic units. And the mapping between syntax and prosody is not fantastic. And so, when originally people tried to find the psychological reality of language, they measured speech, and (unbeknowst to them) were actually finding the psychological reality of speech. One response was a denial of the psychological reality of syntax (is this what brings Jacques to his current position on the issue?). The other was the development of a theory of speech: phonology. Syntax vs phonology One way of looking at it is, that the sequence of words itself is generated by syntax, and the actualization is the domain of phonology. But, here's a problem: Take the case of Heads vs Complements. What determines the relative order of H & C? One possibility is that this is in the output system itself. So, an unordered pair (H,C) is sent to the output module, and the module has to serialize them. So, it might be that the H-C order is a consequence of having to linearize the unordered (H,C) pair. But then what about everything else? What about movement, for example? In general, any two words can be represented as unordered pairs. In fact, in a parallel architecture, one could presumably get rid of all seriality, so that an entire sentence is simply one bowl of spaghetti which in its totality represents the content of a thought. So does movement come out of the necessity for linearization as well? Or maybe it comes about for yet another class of reasons having to do with stuff like pragmatics etc. Maybe, like the ¿ in Spanish, which tells you right at the beginning of a sentence that it will be a query, the wh- elements are just a quick way of letting the other person know there's a question happening. It happens to involve movement, but that might not be the interesting observation about wh- movement.

Thursday, August 10, 2006


Or wasps! Although the Italian word sounds better... it creates this impression of striped insects on little flying scooters. Presumably with crash helmets with the buckle untied and flapping. And mean, goggled ommatidia. Anyhow, I had to put an end to this colony: Which rather reminds me of a previous post. The question, therein really: the nest is beautifully organized, and made up little by little as a coherent structure; does the insect have any kind of representation of the rules? Is there a bit of hardware which, essentially, represents any abstract structural property of the final structure? Or are there just little, stimulus-driven subroutines? And won't they at some level require an abstraction of some sort? One extreme view that I once discussed with Ansgar & Gary (on the lovely island of San Servolo). Basically, these two represent for me the two opposite extremes: Gary believes there are real symbols; Ansgar believes there aren't anything of the sort at all. I'm somewhere in the middle. On the one hand, what else have we got apart from cells in the space between the ears? On the other, where does the systematicity and generative power and great abstractive thought come from? For me, as ever, the answer is somewhere in the middle, and finding it through empirical science will keep me busy for pretty long... :)

Wednesday, August 09, 2006

OMG! Saussure got it all wrong!

Not. Although this is how a recent paper by Farmer, Christiansen & Monaghan in PNAS (103(32), 12203-12208) begins its abstract:
Since Saussure, the relationship between the sound and the meaning of words has been regarded as largely arbitrary. Here, however, we show that a probabilistic relationship exists between the sound of a word and its lexical category.
It's the "however" that's most worrisome. Does the fact that there is a relation between the sound of a word and its lexical category somehow contradict the Sussurean claim? This seems to be what the authors are suggesting. How do they know? Well, they looked at a large corpus of English words. What did they find? (a) That certain sound patterns were more associated with verbs or nouns and (b) English speakers were sensitive to these correlations in bahavioural expeirments. So what's missing? The SAME experiment in Chinese. There are 2 ways of looking at the data: 1) Something about the sounds tells you something about the (syntactic) category. 2) 'Natural' distinctions in the mind (like Nouns or Verbs) might, over linguistic time, cause the accumulation of surface similarities, for example: [i] when verbalizing a noun, a speaker might choose for the verbalized noun to sound more like existing verbs [ii] processing similarities: due to sharing the same syntactic position, words might begin to conform to each other on the surface since this would make processing (producing) them easier. Farmer &al take Line (1). Which might even be correct. But to show that this is so, it is necessary to show that the surface features that separate nouns from verbs are language independent. If they are not, then (2) is the correct description of the facts.
Basically, it seems that there is too much ideological stuff happening, and as a result true progress will suffer.

Ignoring "Ignorance of Language"

Here's a book to be avoided if you don't have too much time: Ignorance of Language by Michael Devitt. Yet another of the "...provocative challenges to [Chomskyan] orthodoxy". What's particularly bad is the repetitive style with enough footnotes to warrant a separate book all by themselves. Worse is the use of "she/her", as in
If a speaker’s competence in a language consists in having knowledge-that of its rules then, assuming RTM [the Representational Theory of Mind], she must represent those rules.
Somehow this is... politically strange. That aside, here's the view from the doorstep (couldn't quite make it past Chapter 1). Take an example. In English, the determiner "the" typically comes before a noun(phrase), as in "the boy" or "the stranger in the strange land". What determines whether "the" comes before or after the noun(phrase)? One possibility is that there is a rule in the mind which says, essentially, to put the "the" before the noun(phrase). The second is that there is no such rule, but when the pair (the, boy) is sent out for `externalization', in speech for example, some process in the motor system forces the order. So, the order is an epiphenomenon of different, presumably rule-governed requirements, and is not part of the linguistic system itself. Devitt says that Chomsky denies the second possibility. But I don' think he does, for example in his Science paper and its followup. Bottom-line. Whether or not something is required as a representation (either a structure or a process) is, as far as I'm concerned, an empirical question. I think it is correct to speak of representations in as far as they help in describing any functionality. So, talk of Nouns or of X-bars is perfectly fine. Just as talk of the DNA repair system is fine, and one can (self included) skip the Biochemistry 101 and mover straight onto Molecular Biology 609..

Monday, August 07, 2006


The problem: You have a speech file, and you have marked the phonemes, using the excellent cross-platform freeware PRAAT. Now you would like to resynthesize using the other excellent cross-platform freeware MBROLA. The solution: My little PRAAT Script for making intonated MBROLA .pho files! (NB: Everything made and tested on a Mac) Setting it up:
  1. Download and unzip the files.
  2. Move the mbrolize.praat (the script) to your favourite location.
  3. Start PRAAT, open the script.
  4. In the script editor menu, click on Add to dynamic menu...
  5. In the following menu, ensure that Class 1 is Sound, Class 2 is TextGrid and Command is Mbrolize as below:
  6. You're set! If you now open a Sound and a TextGrid, and select both, you should see the Mbrolize button like so:
In the zip file there's also one Spanish sentence, Se enteraron de la noticia por este diario, for fooling around with.


- For marking the phonemes you're all on your own, but look at this site for tips. - The script works by looking at the tier you specify in the TextGrid, extracting all the phonemes, and computing -- Their durations -- Their pitch contours, as a list of 20 (%, Hz) values. - Empty slots are replaced by an underscore, so leading and trailing silences are maintained. - For anything else, contact me! :)

Wednesday, August 02, 2006

Typing prosody?

Was just looking at moi fingers typing and making the occasional mistake: What determines these mistakes? I remember someone looked pretty much in detail at typing mistakes, but here's a question: Might typing reflect a "prosodic" organization? Do we type out phrases quickly? Do we pause mostly between phrases? Do we strike the keys most forcefully on phrase-initial segments? (c.f. Keating, Cho & Fougeron, 2003). Are the substitution / swapping /order errors related to the prosodic organization? Probably the best place to study this would be in a chat-room like scenario. I'm sure it could be done already and we would discover the prosodic organization of typing, and it will tell us that prosody is inherent in organizing language onto any motor output.


Here's a good idea for future hand-brakes in cars: Position 1: Off Position 2: On Position 3: Forward brake Position 4: Backward brake The idea is, if you're going uphill and have to stop, and then start again, you can hit Position 4, so that you will not roll back, but will only be able to go frontwards. Etc. You get the idea. Would make driving in Trieste a cinch :)