Comparing languages

This chapter outlines different ways of comparing languages. It discusses the sources of shared features, then sketches the assumptions and methods of comparative historical linguistics, and the reconstruction of the proto-language from which daughter languages developed. It also looks at further ways of extending knowledge of past language stages.

Estimates as to the number of different languages in the world vary considerably, partly because of problems in defining the word ‘language’ (Chapter 10). The figure most usually quoted is somewhere between 4,000 and 8,000. Some linguists carry out detailed studies of individual languages. Many, however, are involved in comparing pairs or groups of them. Sometimes they compare them in order to pinpoint dissimilarities (contrastive linguistics) and sometimes to identify similarities, which may be due to universal, genetic, areal or typological factors.

[]() Contrastive linguistics

The comparison of languages in order to find dissimilarities is known as contrastive linguistics. It is carried out mainly by applied linguists, the name usually given to people who look at the application of linguistic principles to the field of language teaching. It is useful to know in advance where someone learning a language is likely to have []()difficulties, and these often arise in areas of the ‘target’ language which are different from one’s own. For example, in Hindi, negation is simple for the most part. A single negative word is placed before the verb, which is at the end of the sentence:

• []()Bill hindustani nah <Image src="../images/f299-01.jpg" alt="Image" /> hai.

• []()Bill Indian not is

• []()‘Bill is not Indian.’

Because of this, Indian learners of English sometimes have difficulties with the English preference for bringing negatives to the front, and they tend to produce sentences such as:

• []()All of these pens don’t work.

where a British English speaker would prefer:

• []()None of these pens works.

Contrastive linguists, therefore, make detailed comparisons of pairs of languages in order to pinpoint dissimilarities. This enables them to predict difficulties likely to be experienced by learners, which will in turn influence the preparation of teaching materials.

[]() Language similarities

In a broad sense, almost all linguists are looking for language similarities, since the search for language universals is one of the major tasks of linguistics. Many linguists, however, study characteristics shared by groups of languages, rather than all of them. Genetic, areal and typological factors are the three main causes behind these shared features. Genetically based similarities occur when []()languages are descended from a common ancestor. Areally based similarities are due to contact between neighbouring languages. And typologically based similarities occur when languages belong to the same overall ‘type’. Let us look at each of these.

[]() Genetic similarities

The search for genetically related languages, and the reconstruction of the hypothetical parent language from which they were descended, was considered to be the most important task of linguistics in the nineteenth century (Chapter 3). These days, comparative historical linguistics is a branch of historical linguistics (Chapter 13). It enables us to follow through the development of a language from an early stage, and to distinguish inherited features from recent innovations.

It is often not immediately apparent which languages are related. At first glance, Welsh, Spanish and Russian look quite different, yet these are all Indo-European languages. We need to look for systematic correspondences between the languages, rather than similar-looking words, which can be misleading. For example, it is mere chance that German haben ‘have’ resembles Latin habere ‘have’. And Turkish plaz ‘beach’ only sounds like French plage ‘beach’ because it was borrowed from French. On the other hand, beef and cow are (perhaps surprisingly) related, and so are the words paradise, dough and fiction, which can be traced back to a Proto-Indo-European word meaning ‘make, mould, build’.

Two basic assumptions underlie our search for systematic correspondences. First, linguistic symbols are essentially arbitrary. As explained in Chapter 2, there is no connection between the sound of a word and the thing it symbolizes, except in the case of occasional onomatopoeic words. Therefore consistent similarities between languages which cannot be explained by borrowing may be due to common origin. The second assumption is that sound changes are for the most part regular. If one sound changes, then all similar sounds in the same phonetic environment and geographical area change also. On the basis of these two assumptions, we may draw up reliable []()and systematic correspondences between the various related languages.

[]()Insight

Two basic assumptions underlie comparative historical linguistics: first, that the connection between the sound and meaning of a word is arbitrary; second, that sound changes are for the most part regular: if one sound changes, so do all similar sounds in the same linguistic environment and geographical area.

The correspondences between languages which we look for can be found either in the sounds, or, more reliably, in the morphology, since it is rare (though not impossible) for one language to borrow another’s morphology. shows some English and German examples:

<Image src="../images/f0213-01.jpg" alt="Image" />

[]()Figure 14.1.

These systematic sound correspondences between words with the same or similar meaning are the first clue that the languages may be related. The evidence is cumulative. The more correspondences, the more likely the languages are to be related. In the German–English example above, the words are fairly similar, but (as already noted) this is not essential. For example, we can recognize the relationship between English and Latin on the basis of correspondences []()between words such as:

Here Latin p consistently corresponds to English f.

However, correspondences must never be accepted uncritically. We might be dealing with a series of loanwords which diverged in development after being borrowed. For example, there is a superficial correspondence between:

But these are all words borrowed from French at the time of the Norman invasion. More reliable are morphological correspondences, such as those shown in .

<Image src="../images/f0214-01.jpg" alt="Image" />

[]()Figure 14.2.

Such correspondences definitely []()prove that German and English are related. And in the twentieth century, Hittite was established as an Indo-European language on the basis of morphological correspondences, in spite of the fact that its vocabulary consists mainly of non-Indo-European words.

[]() Building a family tree

Once we have established that a number of languages are related, then we have to form a hypothesis as to exactly how they are linked. If we find three related or cognate languages, say German, English and French, then we have to decide whether they should form three separate branches from the parent, or whether (as is in fact the case) two of them diverged from one another at a later stage ().

This would affect the reconstruction, because we would then have to reconstruct the ancestor of German and English before moving on to the next stage, that of reconstructing the overall ancestor.

[]()

<Image src="../images/ch14fg01.jpg" alt="Image" />

Figure 14.3.

[]() Reconstructing the parent language

When we have set up a family tree, we can begin to reconstruct. We do this by looking first of all at the ‘majority verdict’. That is, we look at the sounds found in daughter languages of a similar age, and, as a first hypothesis, suggest that those on which the majority agree might represent the original sounds. For example, among the Indo-European languages, []()we find Sanskrit sapta, Greek hepta, Latin septem, all meaning ‘seven’. This group of words suggests that the parent language word was perhaps \[septa].

Second, our preliminary hypothesis must be checked to see if the developments we have proposed are phonetically probable. We are assuming, for example, that in Greek, \[s] changed to \[h], that in Sanskrit, \[e] changed to \[a], and that in Latin \[a] changed to \[em]. Are such changes possible or likely? In this case, the answer is ‘yes’ as far as \[s] → \[h], and \[e] → \[a] are concerned, but ‘no’ to \[a] → \[em], which is highly unlikely. \[em → a] would be more probable. Can we propose an original \[septem]?

On checking further, we discover that Greek and Sanskrit do not normally lose all trace of \[m] at the end of words, so something must be wrong somewhere. A more plausible hypothesis is that the final syllable was originally rather like the sound sometimes heard at the end of English madam, in which \[m] appears to be behaving in a vowel-like manner. This vowel-like \[m] (sometimes written \[m]) commonly becomes either \[a] or \[em], so our final reconstruction for the word ‘seven’ is \[septm].

[]()Insight

When reconstructing the forms of a proto-language, our main guidelines are majority verdict, and phonetic probability.

As we gradually build up a picture of the proto-language, we need in addition to check whether, in the light of our knowledge of languages in general, we have reconstructed a possible proto-language. If it looks totally unlike any language we have ever seen, then we should be suspicious of our conclusions.

[]() Unreliability of reconstructions

It is, unfortunately, most unlikely that we shall succeed in reconstructing an accurate representation of the parent language. For a start, there are always enormous gaps in the evidence available. In the reconstruction of []()Proto-Indo-European, linguists rely overmuch on Greek, Latin and Sanskrit because of the extensive written records which have survived. Similar written records of Albanian or Armenian might dramatically change the picture. Second, it is not always possible to deduce the actual pronunciation from written texts, yet our reconstructions are to a large extent based on these texts. Third, no parent language is ever a single, homogeneous whole. Every language has dialectal variations within it, so reconstructions are likely to be hotch-potch forms made up from several dialects. Fourth, daughter languages sometimes undergo independent, parallel developments which can falsify the picture of the parent language. If we possessed only English, Russian and Italian, we might wrongly deduce that Indo-European had a stress accent. But stress developed independently in all three languages after the break-up of the parent language. Fifth, borrowing from neighbours can distort the picture.

In conclusion, we realize that reconstructions merely represent the best guesses we can make about the parent language in the light of current knowledge. No one nowadays has the confidence of the nineteenth-century scholar who attempted to translate one of Aesop’s fables into Proto-Indo-European! Above all, reconstructed forms provide a convenient summary of possible inherited features, so allowing linguists to distinguish long-standing characteristics from recent innovations.

[]() Linguistic areas

When similarities are found between adjacent languages, so-called borrowing should be suspected as a possible source. Languages which come into contact with one another often take over some of the linguistic features of their neighbours. Borrowed vocabulary items are particularly common: English has adopted numerous French food words such as courgettes, aubergines, pâté, for example. Borrowing of constructions is more likely to occur if the languages are structurally similar. But even dissimilar languages can, over time, gradually absorb features from one another. If some particularly striking characteristic has []()spread over a wide range, linguists sometimes talk about a linguistic area.

The reason for studying areal characteristics is twofold: on the one hand, knowledge of how languages can affect one another extends our understanding of language change. On the other hand, it is important to isolate shared features caused by borrowing, so as not to confuse them with genetic and typological similarities.

Areal features can involve any aspect of the language. For example, Chinese, Thai and Vietnamese are all spoken in the Far East, and they are all tone languages, something which has apparently come about through contact. And in India, languages with quite different origins have all developed a particular type of sound, known as a ‘retroflex’, in which the tongue is curled backwards against the palate.

Several Balkan languages show similarities which appear to be due to proximity. Albanian, Bulgarian and Romanian all have the so-called ‘definite article’ the attached after the noun. For example, Romanian has munte-le ‘mountain-the’, a construction which has clearly been borrowed from its neighbours, since languages to which it is more closely related show the reverse order, as in the historically related French equivalent le mont ‘the mount’. The same three languages, as well as another neighbour, modern Greek, all say the equivalent of: ‘Give me that I eat,’ when one might have expected them to say ‘Give me to eat,’ judging from other European languages. These particular features seem to have spread during the centuries when Byzantine culture was a unifying force in that part of the world.

[]()Insight

A linguistic area is one in which similar linguistic features have spread via borrowing.

Features which are borrowed from another language seep in slowly. This has led to a search for wider-ranging, more ancient borrowings. Linguistic characteristics shared over more extensive parts of the globe might shed light on prehistoric population movements, an approach known as population typology.

For example, some languages distinguish between []()two types of we: inclusive we which indicates the people in the conversation, and exclusive we, which does not:

• []()Yesterday we (inclusive) arrived. = I and others present.

• []()Yesterday we (exclusive) arrived. = I and others not present.

Hardly any European languages have this distinction, quite a lot of south and east Asian ones have it, and so do most Australian languages. This suggests that it might be a very old feature which has spread slowly westward in the course of centuries.

[]() Language types

Parallel structures in languages may occur because the languages are of a similar type. Just as one can divide human beings into different racial types on the basis of characteristics such as bone structure, skin colour, blood group and so on, so one can divide languages into different groups.

The recent interest in linguistic typology has arisen in part out of the failure to find large numbers of language universals. Absolute universals, characteristics shared by all languages, proved hard to identify, and those attempting to list them were driven back onto vague statements such as: ‘All languages have the means of asking questions.’ When people tried to pin these statements down further, such as querying how questions were asked, it became clear that certain devices recurred in human languages, though different languages favoured different constructions.

Of course, the observation that different languages use different constructions is by no means new. What is newer is the interest in implicational universals and implicational tendencies. That is, if a language has a particular construction, it is also likely to have further predictable characteristics. Just as one can say that, if an animal has feathers and a beak, it is also likely to have wings, so one can make statements of the type: ‘If a language has a basic pattern of subject, verb, object, it is also likely to have []()prepositions (rather than postpositions).’

[]() Morphological criteria for language classification

What criteria should form the basis for language classification? There is considerable controversy about this. The earliest work on the topic, in the nineteenth century, was based on the way in which morphemes were handled.

The number of morphemes per word varies from language to language – so does the way in which morphemes are combined within a word. In the nineteenth century, scholars tried to use such criteria for dividing languages into different types. They recognized at least three different morphological types.

An isolating (or analytical) language is one in which words frequently consist of one morpheme. This is often the case in English:

• []()Will you please let the dog out now?

An agglutinating language (from the Latin word for ‘glue together’) is one in which words can be divided into morphemes without difficulty. Turkish and Swahili are well-known examples. But agglutination is also used to a limited extent in English:

A fusional language is one such as Latin which fuses morphemes together in such a way that they are not easily recognizable as separate elements. For example, -us on the end of taurus ‘bull’ indicates that it is masculine, singular, and the subject of the sentence, but these three aspects cannot be disentangled. Occasional examples of fusion occur in English:

• []()went = go + past tense

At one time it was thought that languages followed []()a fixed pattern of development. The first stage was an isolating one, the second agglutinating, the third fusional. Greek and Latin were spoken of in sentimental terms as representing the highest and best of language types. Everything else was regarded as an aberration, or a symptom of decline and decay. The fallacy of such a belief is pointed out vividly by the American anthropologist and linguist Edward Sapir: ‘A linguist that insists on talking about the Latin type of morphology as though it were necessarily the high-water mark of linguistic development is like the zoologist that sees in the organic world a huge conspiracy to evolve the race-horse or the Jersey cow.’

The main flaw in the type of classification outlined above is that no language is a ‘pure’ morphological type. A few languages fit into one category rather than another, but many have mixed morphological processes. So nowadays, most linguists use other criteria for dividing languages into different types.

[]() Word-order criteria

English uses word order as a basic syntactic device. In linguistic terminology, it is a configurational language (Chapter 7). Perhaps for this reason there has been an enormous amount of interest in word order as a typological characteristic. Among the possible word orders, only a limited number are commonly used, and each of these is likely to possess certain predictable characteristics.

The most usual preliminary classification is in terms of subject, verb, object. In theory, there are six possibilities:

In practice, the ones on the left []()(subject first) are considerably more common than the ones in the middle (verb first), whereas the ones on the right (object first) are extremely rare. In fact, no sure example of OSV has ever been found, and the few examples of OVS are clustered together in South America.

Examples of languages which fit each of these types, with the literal order in which they would express a sentence The dog killed the duck are:

[]()Insight

The best-known work on typology has tried to classify languages into the order in which they express subject, object and verb.

This preliminary classification is useful, but it also presents some problems. The most obvious difficulty is that a number of languages do not fit easily into one of these categories, for various reasons. In some languages, such as the Australian languages Dyirbal and Walbiri, it seems to be impossible to identify a ‘basic’ word order. These appear to be genuine non-configurational languages: their word order is extremely free and flexible. In other languages, the word order seems to be not fixed, but mixed. For example, German has SVO order in main clauses, but SOV in subordinate clauses. It says in effect:

• []()The dog killed the duck (SVO, main clause).

• []()I heard that \[the dog the duck killed] (SOV, subordinate clause).

Furthermore, in several languages, it is extremely difficult to identify the ‘subject’ of the verb. Take the sentences:

• []()The dog killed the duck.

• []()The dog ran away.

In English, the dog would be regarded as the subject of both these sentences. But in some languages, such as Inuit, an Eskimo language, the dog in the first sentence would []()be given a different inflectional ending from the dog in the second sentence. The rationale behind the Inuit situation (somewhat simplified) is that there is a standard ending put on most nouns, but this is changed in cases where there are two nouns in a sentence, in which case the more active participant, the ‘agent’, is given a special ending. Situations such as this make it difficult to make reliable decisions about what is a ‘subject’.

In addition, so-called pro-drop languages cause problems. These are languages which can omit pronouns, usually the subject pronoun. In Latin, for example, can <Image src="../images/f333-01.jpg" alt="Image" /> ‘sing-I’ was commoner than eg <Image src="../images/f333-01.jpg" alt="Image" /> can <Image src="../images/f333-01.jpg" alt="Image" /> ‘I sing-I’, where the pronoun was added only if extra emphasis was needed. In these languages, the order of verb and object when the pronoun is dropped is not necessarily the same as that of verb and object when S, V, O are all present.

These problems show that word order classifications are not entirely trustworthy. However, statistically, certain probabilities emerge. For example: an SVO language is likely to have auxiliaries preceding the verb, prepositions rather than postpositions, and genitives following the noun, whereas an SOV language is likely to have auxiliary verbs after the verb, postpositions rather than prepositions, and genitives preceding the noun. The English examples on the left would be likely to be represented in an SOV language by the order on the right:

Because language is always changing, there are very few languages which are ‘pure’ types, in the sense of being a perfect example of the statistical probabilities. []()Most languages have some inconsistencies, and some doublets (double possibilities). English, for example, can say Sheba’s queen as well as queen of Sheba.

[]()Insight

A list of statistical probabilities is only the first stage in sorting out language types. The next step is to discover why these probabilities occur.

Finding out why these probabilities exist is a more important stage. The answer is still under discussion, and there may be several interacting explanations. One suggestion is that in languages there is a principle of cross-category harmony. That is, different linguistic categories such as nouns, verbs and prepositions, all behave somewhat similarly to one another: the main word or head in a phrase is likely to be in a similar position throughout the different types of phrase. For example, if a verb normally occurs at the beginning of the verb phrase, as in English _eats_ peanuts, then a preposition is likely to be at the front of its phrase, as in _on_ Saturday, and an adjective at the front of its phrase, as in _red_ in the face, and a noun at the front of its phrase, as in _father_ of the family. Interestingly, the conclusion that languages behave in this way has also been arrived at independently by theoretical linguists trying to describe sentence patterns (X-bar syntax, Chapter 7).

Implicational probabilities can also, with a certain amount of caution, be used to reconstruct probable earlier states, as a supplement to other types of reconstruction in historical linguistics (Chapter 13). If we found traces of an old language which had verbs after objects and postpositions, then we would also be able to say that it was statistically likely to have genitives preceding nouns, for example.

At the moment, there is still an enormous amount more to be done in relation to typological characteristics for classifying languages, and the ensuing implicational relationships. Even Chomsky and his followers have started to take an interest in this type of work. Some of these ideas will be discussed later in the book.

[]()[]()THINGS TO REMEMBER

• []()Linguists compare languages for different reasons.

• []()Language teachers and applied linguists often try to find differences between languages.

• []()Other linguists try to find similarities between languages.

• []()Historical linguists try to identify genetically related languages, and to reconstruct aspects of the parent language: this is known as comparative historical linguistics.

• []()Such reconstructions are inherently unreliable, and simply summarize best guesses about the parent language in the light of current knowledge.

• []()Features can spread via borrowing over different languages; the space within which this happens is known as a linguistic area.

• []()Languages can be divided into different types.

• []()The earliest work on typology tried to divide languages into different morphological types.

• []()Better known these days is an attempt to divide languages into those which have a similar basic word order, even though few languages are ever ‘pure’ types.

• []()Interesting work is taking place on implicational universals: if a language has X, then it is also likely to have Y.