The Happiness of Pursuit: What Neuroscience Can Teach Us About the Good Life (18 page)
Read The Happiness of Pursuit: What Neuroscience Can Teach Us About the Good Life Online
Authors: Shimon Edelman
BOOK: The Happiness of Pursuit: What Neuroscience Can Teach Us About the Good Life
10.85Mb size Format: txt, pdf, ePub
Collocation with zero, one, or more slots in it is in fact the most general kind of linguistic pattern. As a formulaic construction, it allows the speaker to express intent in a manner that is ritualized, hence likely to be interpretable by the listener. The particulars of the message are conveyed by selecting a filler for each slot out of a set of possibilities that may be large or small but is always limited and is specific to the construction in question.
Stringing such constructions together sequentially and nesting them (that is, filling a slot with another construction that itself has slots) supports the expression of extended, hierarchically structured intents and messages, which can thus grow to be as complex as any conceivable structure in the mind. The resulting system nevertheless remains manageable because it relies on a single organizational principle: the codification of conditional probabilities of some elements, given others. This same principle extends to the very fundamentals of the use of language for communication, insofar as it captures the dependence of the choice of constructions on the speaker’s intent and on the situational context in which communication is taking place.
I find it extremely pleasing that a communication system that has evolved to fit inside a brain made of neurons boils down to the one computational operation that neurons implement most naturally—making certain represented quantities depend on others. Given that the system of constructions (unlike, say, the physiology of a liver, but very much like the actual connection strengths among neurons in a brain) needs to be learned from a corpus of experience, it also makes great sense that the slotted collocation construction can be learned by tallying conditional probabilities over the discrete elements of language.
To do so, the learner that monitors a stream of utterances needs merely to align and compare them. The parts that match across multiple utterances signify a recurring sequence or collocation, whose statistical significance can then be estimated by assessing the probability of the recurrence arising by chance. At the same time, a localized mismatch signals a slot in the collocation, the differing elements being the options among which a choice must be made.
If, like me, you are not fluent in the language of Homer, you can appreciate the nature of the computational opportunities for (if not the actual experience of) discovering structure in a profoundly foreign stream of data by examining the following passage. These are the six lines from
The Odyssey
whose translation appears earlier in this chapter (“My lady goddess . . . ”). Here I offer them in the original ancient Greek, complete with polytonic accents, but with spaces omitted, to simulate the feel of regular speech, in which there are no pauses between words:
The Odyssey
whose translation appears earlier in this chapter (“My lady goddess . . . ”). Here I offer them in the original ancient Greek, complete with polytonic accents, but with spaces omitted, to simulate the feel of regular speech, in which there are no pauses between words:
πότναϑεάμήμοιτόδεχώεοοἶδααὶαὐτὸς
πάνταμάλοὕνε
εἶδοςἀ
ἡμὲνγὰρβροτόςἐστισὺδ᾽ἀϑάνατος
ἀλλὰ
οἴ
By approximating the acoustic speech stream, which is analog (in the sense of Chapter 2), with text, which is digital, this example focuses on how a learner can bootstrap from knowledge of the basic categorical elements of language (phonemes, represented here by letters) to the discovery of words (statistically prominent sequences of phonemes) and other, more complex constructions. As noted earlier, all it takes is for the learner to align and compare the stream of data to shifted versions of itself. This task requires memory for temporarily holding data and a mechanism for matching the held data to incoming sequences, both very straightforward functions that can be performed by rather simple neural circuits.
If you scan the Greek text for places where it matches shifted versions of itself, you will indeed discover sequences of characters that appear more than once. One such sequence is
αὶ, which happens to be the Greek for “and” (see if you can find any others). Of course, a baby learning Greek has no access to such privileged information and so must rely on a statistical significance test to decide whether to admit
αὶ into its lexicon or discard it as a fluke.
αὶ, which happens to be the Greek for “and” (see if you can find any others). Of course, a baby learning Greek has no access to such privileged information and so must rely on a statistical significance test to decide whether to admitαὶ into its lexicon or discard it as a fluke.A partial rather than perfect match between two sequences that is nevertheless statistically reliable signifies a construction with a slot, in which some variation is to be expected. In the present example, the last two lines end, respectively, with the sequences ἤματαπάντα and ἦμαρἰδέσϑαι. Guessing that πάντα and ἰδέσϑαι are words in their own right because they appear elsewhere in the passage (indeed, these are the words for “all” and “to see”), we are led to conclude that ἤματα matches ἦμαρ. A Greek scholar will tell you that these strings are in fact related by inflection: they are the plural and the singular, respectively, of the word for “day.”
30
30
Other books
From Baghdad To America by Jay Kopelman, Lt. Col. USMC (ret.)
house of women by Yelena Kopylova
No Returns by Rhonda Pollero
Winner Takes All by Dragon, Cheryl
El salón de la embajada italiana by Elena Moreno
Sweet Persuasion by Banks, Maya
The Silken Edge (Silken Edge 1) by Paige, Laci
Queens' Play by Dorothy Dunnett
The Hand of Christ by Nagle, Joseph
The Gathering by Lily Graison