A Science Of Human Language - Part #10

In this part of the series, commenced here, I give some concrete examples from various languages of how words can cue the category from which other words were, or are to be selected.
"Can the Saussurean definition of grammar as a structured system of SIGNs be reinterpreted as a structured system of code + information?"
Huang, Chen and Gau1, Institute of Linguistics,
Academia Sinica.

I suggest that the naturally-evolved grammar of every human language is, indeed, a system of codes and information.  Phrases and sentences in all spoken language, I suggest, consist of information which is intended for transmission as content, and information which is intended to prevent or correct errors in transmission.  In this article, I focus mainly on the information encoded as grammatical gender as an example of how this error-handling mechanism works.

Grammatical Gender As Error-Handling Code

Grammatical gender has little to do with biological gender.  Except for words which refer to humans and animals, for the most part grammatical gender is an entirely arbitrarily evolved classification scheme.

Just because English and other languages have three genders, it is often assumed that only three genders exist.  In fact, there is no theoretic limit to the number of genders possible.  The Tova-Tush2, or Bats language has eight genders.

It might be better to think of genders as purely arbitrary word categorisation schemes.  Suppose a language with three genders.  Let the class specifiers or determiners be li, lo and lu.  let us imagine a rule which says that words used with li end with i, and so forth.  Let the three groups of words be equal in size.  A person hearing the word lo is primed to expect a sub-set of words, the words ending in -o.  If the next word does not end in -o then an error has been detected.  Normally, other information in the linguistic or environmental context will still further reduce the range of candidate words so that the hearer can make a well-informed guess about the speaker's intent.

In general, word categorisation schemes in languages serve to reduce the range from within which other words may be selected.  I suggest that sentence formation operates primarily at a mental model or nuon level.  Once the semantic content has been selected, other words and parts of words, quons, are added to the data stream as error-handling codes before final output as speech.  These error codes repeat some of the semantic content in the nuons, and thus narrow down the lexical domain within which the original words may be found.

The hallmark of quons is that they cannot easily be defined in terms of referents in our environment.  We can readily point to instances of objects, actions, attributes of objects and methods for actions.  In the case of words such as 'the', and 'to' when used with a verb, there is no referent.  Even pronouns such as 'he, 'she', 'it' can only be demonstrated by use.  Their application is so wide that, unless used within a predefined context, they are entirely ambiguous.

Child, to parent: "She hit me."
Puzzled parent, to child: "Who is she?"

Many languages share a code of gender conformity: a set of rules which cause various words to conform to a pattern according to the class or grammatical gender.  The pattern may be detected by a computer program using stochastic processes, regardless of the language.  The advantage of a grammatical gender system, I suggest, is that it adds greatly to the error-handling capabilities of a language.  Some examples follow which illustrate something of this code-pattern similarity.

his old brown horse was eating hay.
il suo vecchio cavallo marrone stava mangiando il fieno.
su caballo marrón viejo comía el heno.
son vieux cheval brun mangeait le foin.
sein altes braunes Pferd aß Heu.

his old brown cow was eating hay.
la sua vecchia mucca marrone stava mangiando il fieno.
su vaca marrón vieja comía el heno.
sa vieille vache brune mangeait le foin.
seine alte braune Kuh aß Heu.

In languages which have masculine and feminine genders the question arises as to whether or not the use of gender-specific classes affects a language user's perception of inanimate objects.  In other words: do such language users consciously or unconsciously feel that inanimate objects can be masculine or feminine in a psychological sense?  Experiments using the Italian3 language suggest that the answer is 'no'.

In the Chinese language, there are many classifiers4.  These words serve to restrict the category or class from which a following word may be selected.  Equivalent examples in English might be 'one stick wood', 'one slice bread'.  These words seem to perform the same task as grammatical genders: they act as a cue or cross-check so as to confine any mis-heard word into a fairly narrow category.  The more narrow a word category, the more accurate will be any guess as to what the mis-heard word was.

Taken together: the use of error-handling codes, or quons, the conformity to social norms of pronunciation and writing, conformity of topic to other language users' expectations, all of these confine any guesswork to a very narrow domain for each wrongly heard word.  In most instances, I suggest, the confinement is so narrow that the error-correction is made at an entirely subconscious level.  A hearer may even deny that there was any error.


[1] Huang, Chen and Gau, Noun Class Extraction From A Corpus Based Collocation Dictionary http://cwn.ling.sinica.edu.tw/churen/1998%20noun%20class%20extraction.pdf

[2] The Bats or Tsova-Tush, a forgotten Caucasian people

[3] Vigliocco, Vinson and Paganelli, Grammatical Gender And Meaning

[4]  Wikipedia, Chinese classifiers http://en.wikipedia.org/wiki/Chinese_counters

If you have enjoyed this article, you may enjoy other articles, mostly about language, in my  blog The Chatter Box