Wednesday, 15 August 2012

The Nature of Language and Historical Linguistics, redux

 Historical linguistics often seems to rely on the assumption that languages are not very creative, and that words are confined to particular meanings.  This isn't necessarily so, but it is nonetheless generally the case that languages are not very creative.  Here I'll try to explain a few of my thoughts on how one can reconcile the fact that words change in meaning and form with the way in which we can find out more about the past using reconstructed languages.  I discussed the same things in my post the other day, but I believe this is a better explanation of the position.

People act on the basis of reasons whenever they do anything or perform any action - or at least, they give the illusion of doing so. This includes speaking and communicating. When people communicate, they make choices about how best to accomplish their desires (which usually include effective communication of what is on their mind, but can also lots of other desires, like the desire for status) on the basis of their beliefs about the world. With language, these beliefs are complicated, as meaning relies on nested beliefs that are necessarily recursive.

Language relies on what game theorists call common knowledge. Common knowledge in ordinary speech is something that everybody in a group knows. In game theory, however, something someone, or everybody, knows is called 'first-order knowledge'. It is something individuals know. Second-order knowledge would be something that individuals know because of their beliefs about what other individuals know; I believe that 2 + 2 = 4 because you told me so.

Common knowledge, on the other hand, is knowledge of every order. It is something that you not only believe, but something that you believe others believe. And, moreover, something that you believe the others believe that you believe that the others believe, and that they believe that, and so on, ad infinitum. When you use a word to convey an idea to somebody else, you necessarily believe that they will believe that they have the same beliefs about the word as you do, and that they believe that you have the same beliefs as they do, and so on. As you can see, with only a few orders of recursion a great cognitive burden is placed on your brain. Your brain cannot process infinity, and recursive processes are necessarily potentially infinite. But the brain seems quite good at taking potentially infinite processes and ideas and turning them into easily computable abstractions.

The Fibonacci sequence is a recursive procedure. In the sequence, you use the sum of the two prior terms to generate the next term. The first term is 1, and the second is 1 as well. The next is 2, because 1 + 1 = 2.  1 + 2 = 3, so 3 is the fourth term. The sequence goes: 1, 1, 2, 3, 5, 8, 13, 21, 34, etc. I can remember this as a recursive process ('use the two prior terms to generate the next term'), and I can use this principle to work out new terms to add to the sequence, but I can also remember it as a finite sequence of numbers that I store differently in my brain ('the Fibonacci sequence goes 1, 1, 2, 3, 5, etc').

This seems to be how humans learn words. We use the complicated recursive procedure of common knowledge to work out what other people mean ('I believe that he believes that I believe that he believes that […] “salmon” means that kind of fish'), and we can then abstract the words we have learnt as a result. If I have used that procedure several times, then I have learnt the word 'salmon', which I can then use without the cognitively-burdening recursive procedure. When I hear the word 'salmon', I don't have to think of the intention of the speaker, but simply of the fish. This is more efficient computationally, which means that my brain has less work to do in speaking and understanding. Importantly, however, when there appears to be ambiguity, we are capable of breaking these meanings down into beliefs about the intentions of the speaker. This is important because we would otherwise be constrained to always use conventional meanings without taking context into account.

When people use existing words to convey ideas to one another, they are relying on a convenient, efficient means for doing this. They are not totally constrained by words in what they can and cannot communicate, and words are clearly very flexible things. Here is an example: as I mentioned in my previous post, the word 'shark' came from Yucatec Mayan into English at some point in the sixteenth century.1 It conventionally refers to big, toothy fish of the clade Selachimorpha. If you overheard it in conversation, you would be reasonably justified in assuming that selachimorphs were being referred to rather than, say, tea. But if I tell you beforehand that I am using the word 'shark' to refer to tea, a drink made from dried, roasted, oxidised leaves of Camellia sinensis brewed in boiling water, and then I say, 'I am drinking a hot cup of shark', then you know precisely what I mean even if you find it comical.2  As the distinguished twentieth century philosopher Donald Davidson put it:
'… you cannot change what words mean […] merely by intending to […], but you can change the meaning provided you believe (and perhaps are justified in believing) that the interpreter has adequate clues for the new interpretation. You may deliberately provide those clues.'3

So humans can use language in entirely novel ways if they believe that the other humans they are communicating with will understand what they mean. The words we inherit in our languages are conventional signs that make communication efficient and relatively trouble-free because they are found to be common knowledge to the people with whom one wishes to communicate. Because the advantages of using conventional signs far outweigh whatever advantages there might be in crafting an entirely new language for yourself every time you wish to speak, humans almost uniformly use the conventional signs, or words, they come into contact with on the basis of a justified belief that most of them will have abstract, conventional meanings that are common knowledge within the community. It's important to recognise that humans can be creative with language, and are not bound by the terms they come across. They use them because they're efficient, not because the words they already know are the only way in which they could potentially communicate.

The use of conventional terms operates with such regularity that we can use cognate words – words that are related to one another in the different languages of a language family – to understand what the people who used these words thought or came into contact with.  It turns out that meanings usually change quite slowly (although there is no standard speed at which this happens). When linguists compare words in languages they believe are related, they often find words that are very similar both in form and meaning. The Indonesian word lima and the Hawaiian 'elima both mean 'five' (the number). They sound very similar. In fact, the two languages show systematic similarities of this sort, indicating that lima and 'elima are not similar due to a fluke, but are cognates.  They are related words that come from a prior word of the same meaning that can be reconstructed to a proto-language.

These systematic similarities point to the two languages being genetically related4 members of the Austronesian language family. As the terms 'elima and lima are clearly related, and come from different branches of the Austronesian language family (meaning that the ancestor of both sub-families possessed the term), we might reconstruct a term for the Proto-Austronesian language – the language from which both Indonesian and Hawaiian are ultimately derived – with the sound *lima and the meaning 'five'.5 We would be justified in believing that the speakers of Proto-Austronesian used the concept of five, and that they verbalised this concept using sounds very similar to *lima. This is called the Wörter und Sachen method, from the German meaning 'words and things'. If we can reconstruct a word for an object to a proto-language, then we are reasonably justified in believing that the speakers of the proto-language knew of the thing for which the term applies. As another example, the word for 'beech tree' is reconstructible to Proto-Indo-European, meaning that the speakers of Proto-Indo-European probably lived in an area with beech trees.

What we are relying on in using historical linguistic data is the human propensity for efficiency in satisfying their aims. It is not inevitable that languages should show such consistent changes, nor such continuity in the meaning of individual words. It is possible for words to change in meaning, potentially muddying the results of the Wörter und Sachen method. An example with Proto-Indo-European is that of Indo-European words for 'salmon'. A word can be reconstructed to Proto-Indo-European for 'salmon', but it appears that the word in Proto-Indo-European actually applied to any long fish, including trout as well as salmon (true Salmonidae). Salmon were not present in the Indo-European homeland.


The desire for efficiency in spoken communication – maximising the effect/effort ratio, as the anthropologist Dan Sperber puts it – has inevitable effects on language outside of the decision to employ conventional signs and words.  It affects how sentences are structured and the nature of the information conveyed. When I ask a friend, 'how many beers have you had?', it is inefficient for her to say, 'I have had five beers'. Instead, she will say, simply, 'five', relying on the prior mention of beers to efficiently communicate what she means. If she says, 'I have had five beers' instead of only the word 'five', then she must have another reason for doing so, like the desire for perfect clarity – an aim that she wishes to satisfy beyond merely communicating the information. Likewise, instead of saying, 'I did not', most speakers of English say, 'I didn't'. The contraction expresses the same information more quickly and thus more efficiently than the extended phrase.6 If someone says 'I did not' in place of 'I didn't', they must have another reason for doing so, as Bill Clinton did when he wanted to emphasise the fact that he had not had sex with Monica Lewinsky.  Perhaps someone might believe that the full 'I did not' phrase is more 'proper', or 'correct'. In most ordinary speech, however, it is quicker and easier to say 'I didn't', and it is this form that we find more commonly in spoken English today for this reason. 'I didn't' is very different in sound to 'I did not', and to some extent it represents a corrosion of the original phonemes7 of the phrase.

This corrosion is a natural result of the human desire for efficient communication coupled with the belief that shorter phrases with less distinct phonemes will be as easily understood by listeners as the long phrase.  This seems to be a generally unconscious decision, one you make without thinking deeply about it.  People don't tend to consider saying 'I have had five beers' when they could just say 'five', and they don't tend to think about saying 'I did not' when they're speaking hurriedly.  These unconscious choices cause changes in consistent directions, meaning that phonemes do not change randomly.  For a host of other reasons, phonemes tend to change consistently (although not necessarily predictably).  S frequently becomes h, for instance - as happened in Greek, but also in several east Polynesian languages.  S is a phoneme that often results from a prior k, as happened in branch of Proto-Indo-European (the so-called centum/satem split), and also later in Latin (where centum, pronounced kentum, became French cent, pronounced with an initial s).

Phonemes don't necessarily change predictably, because they don't always change in the same directions.  The other phonemes of the language - in addition to social concerns - can affect the change.  The th in think (a voiceless dental fricative, θ) and the th in the (a voiced dental fricative, ð) are common sounds in most dialects of English.  Pronounced by speakers of Estuary English, however, these sounds present entirely differently: 'think' becomes 'fink' (fɪŋk) and 'the' becomes 'vuh' (və).  Germans stereotypically pronounce them yet differently: 'sink' (sɪŋk) and 'zuh' (zə).  This is partly because of other phonemes in the dialects these people use, partly due to difficulty in pronouncing sounds you're not habituated to through practice, and partly to do with social pressure to use the same phonemes used by the population they want or need to imitate.  So predicting precisely which phoneme will be adopted by which group of people is very difficult indeed, if not impossible.

There is a consistency in the choice of phonemes, however.  S is unlikely to become k, f is unlikely to become p, and so on.  In the case of ð and θ, the distinction is between voiced and voiceless consonants respectively.  θ doesn't buzz when it's pronounced (try it!), and ð does.  Likewise, z is voiced and s is voiceless - z buzzes, s doesn't.  F is voiceless, v is voiced.  In both cases of English pronunciation, by Estuary dialect speakers and Germans, the consonants replacing ð and θ are analogous in respect of the voiced/voiceless distinction.  This means that we can predict roughly what categories of sounds might be used in the change in phonemes in some specific cases.  It doesn't allow us to predict precisely which phoneme will be used, though.  And there is a further problem in that some sound changes are unconditioned, meaning that they apply across the language with no exceptions, and some are conditioned, meaning that they apply in certain positions relative to certain other phonemes.

As phonemes change over time in fairly consistent but not entirely predictable ways, it is impossible to establish a relationship between two languages on the basis of similar sounds (and therefore cognate words) after a certain amount of time.  The number of possible sound changes mounts up, leading to no certain conclusions and only speculation.  A k can become a sound like an s which can become an r which can become an l, and each of these changes will have repercussions throughout the pronunciation of the language.

Over the course of thousands of years, this causes related languages to diverge considerably from one another as different labour-saving linguistic devices are employed. Given sufficient time, languages can diverge so completely that discovering their connections to one another is impossible. We are lucky with certain languages, including those of the Indo-European and Austronesian language families, because they diverged from each other recently enough that their connections have not been obscured by thousands upon thousands of years of progressive sound changes and transfers of meaning in their vocabularies. Most of these languages families propagated themselves in the last five thousand years or so. Some other languages either diverged from their parent languages too long ago for linguists to be able to find connections between them and any other languages, or the languages to which they were related have died out. These seemingly unique languages are called language isolates. Sumerian, the earliest attested human language, is a language isolate, as is Basque, today spoken in northern Spain and southern France.8

Similarly, the proto-languages reconstructed by linguists were not the earliest languages on earth. They are only the earliest languages of which we can have reliable evidence. Proto-Austronesian, Proto-Indo-European, Proto-Niger-Congo – they all came from some other earlier language or languages. Perhaps all human languages descend directly from a single language spoken in east Africa 100,000 years ago, or perhaps some significant linguistic diversity has existed since the very beginning of speech. We will probably never know, and the reason is that the desire for efficiency in communication has caused successive sound changes in all human languages such that accurate reconstruction of the ultimate proto-language of earth is impossible. It's also important to note that many language families have disappeared, and are disappearing still, through invasion, conquest, migration, and recruitment by the large and powerful language families that now dominate our planet. This means that any reconstruction of Proto-Earth created through the application of the comparative method to the proto-languages we have in our possession would be at best incomplete.

In any case, despite the potential for radical changes in meaning and sound over incredibly short periods, and despite the lack of absolute meanings for particular words and phrases, it is clearly possible not only to establish relationships between languages and to track changes in use but also to use the relationships discovered to find out more about the history and prehistory of Homo sapiens sapiens.  And I think that is quite cool.

1Prior to this, the word had apparently been 'sea-dog', as I mentioned before.
2See Keith Donnellan's "Putting Humpty Dumpty Together Again" (1968), and Alfred MacKay's response, "Mr Donnellan and Humpty Dumpty on Referring" (also 1968), both in The Philosophical Review.
3Davidson, in his fantastic article, A Nice Derangement of Epitaphs (1986).
4 Confusingly, linguists use the term 'genetically' to refer to these sorts of relationships between languages.
5An asterisk * before a term in italics indicates that it is a hypothetical, unattested word in a proto-language, reconstructed using the methods of historical linguistics.
6Assuming the speaker doesn't want to waste time, that is. If time-wasting is their intent, then using a contraction is counter-productive. The fact that language changes consistently in the direction of quicker communication indicates a general and non-culture-specific human preference for maximising the use of time, however.
7See below.
8Basque is probably the only extant survivor of the Old European languages that were spoken in Europe before the arrival of Indo-European speakers in the Bronze Age. Surprisingly, Proto-Basque can be reconstructed using internal reconstruction, as opposed to the comparative method (as an isolate, Basque has no languages to be compared with). A language ancestral to Basque has been spoken in Spain since at least the time of the Roman empire, and its speakers at that time were Roman citizens. Basque acquired some Latin loanwords as a result, and these loanwords have undergone consistent sound changes over the past 2,000 years. The sound changes displayed in these early Latin loanwords give indications of sound changes that have operated in the rest of the Basque lexicon, allowing linguists to reconstruct the Proto-Basque tongue. It's quite a remarkable feat.

No comments:

Post a Comment

You can post anonymously if you really want to, but I would appreciate it if you could provide some means of identifying who you are, if only for the purpose of knowing who has written what.