Re: Pattern recognition

Date: Mon Oct 06 2003 - 17:46:07 MDT

Someone in this group suggested that it might be desirable to create a
grammar so comprehensive that misunderstandings would no longer happen (at least
that's how I remember it, and there have been too many new posts of late for me
to recall which post it was). Seems this isn't possible.
I came across an interesting brief at relating to this
In essence, a speaker prefers fewer words, a listener more, and all human
tongues occupy a 'sweet spot' in the middle.

Language evolved in a leap

Conflicting needs may have driven rapid development of communication.
22 January 2003 PHILIP BALL

Language probably leapt, not crept, from squeaks to Shakespeare, two
physicists have calculated. Human communication, they propose, underwent a 'phase
transition', like solid ice melting to liquid water.

The richness of human languages is a fine-tuned compromise between the needs
of speakers and of listeners, explain Ramon Ferrer i Cancho and Ricard Solé of
the Universitat Pompeu Fabra in Barcelona. Just a slight imbalance of these
demands prevents the exchange of complex information, they argue.

So languages between those of present-day humans and the limited signalling
of some animals cannot really exist. There must, at some point, have been a
switch from rudimentary to sophisticated language.

This contrasts with some linguists' view that language evolution was a
gradual affair in which new words accumulated steadily.

Greek or grunt

A language that conveyed all information unambiguously, say Ferrer i Cancho
and Solé, would have a separate word for every thing, concept or action it
referred to. Such a language would be formidably complicated for the speaker: the
green of grass, for example, would be represented by a totally different word
to the green of sea, an emerald or an oak leaf. But it would be ideal for the
listener, who wouldn't have to work out any meanings from a word's context.

Ideal for the speaker is a language of few words, where simple, short
utterances serve many purposes. The extreme case is a language with a single sound
that conveys everything that needs saying. Some might suggest that teenagers
prefer this kind of minimal-effort tongue that forces others to figure out what
their grunts actually mean.

Ferrer i Cancho and Solé have devised a mathematical model in which the cost
of using a language depends on the balance between these conflicting
preferences1. They calculate the properties of the lexicon that requires minimal effort
for different degrees of compromise, from exhaustive vocabularies to one-word

They find that the change from one extreme to the other does not happen
smoothly. There is a jump in the amount of communication, from very little to
near-perfect, at a certain value of the relative weightings of speaker and hearer

Human languages, say the duo, seem to sit right on this sudden change. When
it happens, the frequency of word usages develops a distinctive mathematical
form, called a power law. The power law disappears on either side of the
communication jump.

It has been known since the 1940s that human languages do indeed show just
this kind of statistical distribution of word usage - the social scientist
George Kingsley Zipf spotted the power-law behaviour. But it has never been
satisfactorily explained before, although Zipf himself speculated that it might
represent some kind of "principle of least effort".


Ferrer i Cancho, R. & Solé, R. V. Least effort and the origins of scaling in
human language. Proceedings of the National Academy of Sciences USA, published
online, doi:10.1073/pnas.0335980100 (2003). |Article|

© Nature News Service / Macmillan Magazines Ltd 2003


This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:42 MDT