Practical Syntax

Practical syntax

Richard Hudson, May 1995

1. Practical syntax?

Some people - perhaps most people who are likely to read this paper - have a practical interest in sentence-structure (the subject-matter of syntax). Here are some practical reasons for such an interest:

P When learning a foreign language you have to learn how to use words in phrases and sentences; the more clearly you understand the patterns you are learning, the more successful you will be.

P The same is true when you teach a foreign language; again, the clearer your understanding of sentence structure, the more you will be able to help your students.

P If your responsibility is to help young children to develop their language skills, you need to understand the skills they already have as well as those that are still beyond their grasp. This applies as much to parents or grandparents as to professional first-language teachers.

P Some unfortunate individuals lose, or never acquire, the normal skills; speech and language therapists who try to treat them must understand exactly what it is that is missing before they try to provide a remedy.

P When you communicate about anything which is at all complicated you have to select sentence-structures so that they convey the intended message with the minimum of pain to the hearer or reader. There is never a single way to communicate a given message - there are always alternatives. (As I type this, I am writing two or three versions of most sentence before I find one that I can't improve.) The better we understand sentence-structure, the better we can make these choices.

P In these high-tech days an increasing number of people are trying to program computers to understand what humans write or say, and to produce human-like writing and speech. This requires a profound understanding of sentence-structure, and this understanding must be severely practical because it has to be translated into detailed instructions for the computer.

Fortunately there is a large measure of agreement on how sentences are structured. For example, consider this example sentence:

(1)We drank the coffee.

1

We grammarians would all agree on a rough classification of the words (personal pronoun + past-tense verb + definite article + singular common noun), using traditional terms which (in some cases) date back to the Romans or even to the Ancient Greeks. We could also agree that the coffee `hangs together', as a phrase, because both words help to identify the same bit of meaning; and that we and the coffee have distinct `roles' or `grammatical functions' in the sentence, called `subject' and `object'. Furthermore, the same kind of classification can be applied to sentences in other languages, like the Turkish translation of sentence (1).

(2)Kahveyi içtik.

In this case of course the details are different: only the common noun and the verb survive the translation, as the meanings of we and the are expressed differently (by suffix-like elements) in Turkish. But any grammarian would describe kahveyi as the object of içtik (or of the whole sentence), and would accept that ik shows what the subject would have been if there had been one.

I think the agreement extends to the following generalisations about sentence structure:

P Sentence structure is closely linked to meaning, but the same meaning can be expressed by different sentence structures (especially when we consider different languages).

P At the heart of sentence structure are the relations among words. These are of two types: the grammatical functions (e.g. subject and object), and the links which bind words into larger units - into phrases, into clauses and ultimately into sentences.

P Sentence structures are controlled by the rules (which constitute the grammar of the language concerned). The rules control the ways in which words are combined and the meanings which the various combinations express; they refer to the relations among words, but also to the ways in which these are reflected in the order of the words concerned and in the word-classes (e.g. verb, common noun, article, singular, past tense) to which they belong.

However, alongside this agreement there is a great deal of disagreement, both about general principles and about the details of how particular kinds of sentence should be analysed. This disagreement is largely a product of the last 40 years, which have seen a great expansion in research on sentence structure and a proliferation of alternative theories as well as of grammars of particular languages[1].

1

This expansion is a mixed blessing for someone in need of practical help. On the one hand, I feel sure that we all understand at least some parts of sentence-structure better now than we did 40 years ago; but on the other hand the choice is overwhelming. Unfortunately, the choice matters; some approaches are much more helpful than others. Different users no doubt have different needs, but in my opinion the following list covers the most important things in a truly `practical' approach to syntax:

P Rather obviously, it must be as near as possible to `true', in the sense of being compatible with all the relevant facts. It is no help to have a grammar which is simply used, but also simply wrong.

P It must be as complicated as the facts require, but no more so. Simple sentences should have simple analyses, and complicated sentences correspondingly complicated analyses.

P It should offer a usable diagramming system. Even a sentence as simple as We drank the coffee has a structure which is hard to describe in prose, but it should be easy to present as a diagram. To be usable a diagramming system should be visually clear and reasonably economical of space.

P It must be easy to learn cumulatively, so that each new item can be used as soon as it is learned.

P The more comprehensive it is, the better. The ideal is a grammar which allows us to handle every word in every sentence that we try to analyse. No such grammar exists, but it is reasonable to look for a grammar which handles the vast majority of words - 99% coverage is a realistic target.

The aim of this paper is to present a theory of sentence-structure which, in my opinion, satisfies all these criteria as well as any others that are available, and better than some. It is called Word Grammar and has been described in two books (Hudson 1984, 1990). For several years I have been teaching it to undergraduates who learn enough in one term to be able to do an almost complete syntactic analysis of any ordinary text in English. The analysis they learn is as near to the truth as I have been able to make it (after 30 years of work in this area); it is as simple or as complicated as the patterns being analysed; there is a simple diagramming system; and they can apply it from the first week.

2. Word Grammar: an overview

We start with the Word Grammar (henceforth WG) analysis of our sample sentence:

1

Diagram 1

This diagram is just as simple as the structure it represents. The sentence contains four words, so the diagram shows precisely four units (one per word) and their interrelations. These relations are shown by the arrows, which point towards the word which bears the grammatical function shown by the label (where `s' stands for `subject', `o' for `object' and `c' for `complement'). The vertical arrow points at the one word which has no grammatical function, drank. (The reason for it will become clear below.) The fact that the coffee is a phrase is shown by the arrow which links the directly to coffee.

The simplicity of the diagram isn't the result of ignoring a lot of detail which is uninteresting or unimportant. It represents a scientific claim that there really is nothing else to say about the syntax of this sentence, a claim which could be defended against counter-claims. It may be wrong, of course, but I don't know at present of any reason for doubting it. However, it is important to make clear exactly what I am claiming.

I would certainly agree that a complete analysis of every aspect of this sentence would go far beyond the analysis shown. On the one hand we can say a great deal about its meaning - who the word we refers to, the time reference of drank, the semantic relations between the coffee and the drinking, and so on. On the other hand, we can also say something about the internal structures of the words (described in terms of consonants, vowels and syllables). The meaning involves a semantic structure, and the sounds are part of a phonological structure; both these structures would be part of a complete linguistic analysis of the sentence, but not part of a syntactic analysis (which is concerned, as I said at the outset, with sentence-structure - how the words are combined with each other).

Nor am I claiming that the diagram shows the whole of the syntactic analysis. The diagram contains three kinds of labels: labels for word-classes, for word-features, and for grammatical functions; but some of these labels imply one or two other more general labels as well. For example, We is labelled `pN', short for `pronoun'; but (according to WG, and most other theories) pronouns are a kind of noun, and (of course) nouns are a kind of word, so `pN' implies `N' (for `noun') and `word' as well. Since these extra labels are completely predictable there is no need to include them in the diagram. I shall say more later about the grammatical-function labels.

A single example gives a `taste' of how WG works - what the diagrams look like, and the kind of information they contain. As we consider further examples below I hope to show that they are as complex as required, but no more so; and that they show everything that a syntactic analysis should show. I can report that the system can be learned cumulatively, that it is usable, and that it is reasonably comprehensive. My evidence for these claims is that my undergraduate students do learn it cumulatively (adding a new part of the grammar each week), that they do learn to apply it to ordinary bits of English (or other languages), and that a term's teaching brings them to the stage where they can probably deal confidently with about 97% of the words in an average text which they themselves have chosen.

1

But what about truth? Nothing in this life is perfect, and the same is certainly true of every theory of grammar, including WG. In syntax even some of the facts are unclear and in such cases it is perhaps unwise to talk of truth; but there are many parts of syntax where the facts are clear, and truth really is relevant. Anyone who builds a theory of syntax must confront these facts honestly as a test of the theory, and reject (or at least modify) the theory when it fails. I have spent decades doing precisely this, and the result is a good approximation to the truth. Unfortunately the world is full of other linguists who can say precisely the same, but who have come to very different conclusions from each other (and from me). The study of syntax is a central part of the science of linguistics, and is notoriously frought with controversy.

Some of the differences between theories are certainly not important, but equally certainly others are; so any `consumer' of syntactic theories should treat any theory (including WG) with caution. Behind every theory lies a large body of arguments, decisions and mind-changing which may have been debated in the technical journals and monographs but which the consumer may not have time to explore. The result is that some parts of any analysis are to some extent controversial.

Even our simple sentence contains a controversial point: the analysis of the. According to Diagram 1, the object of drank is the, not coffee; whereas traditional analyses would reverse this relationship. Linguists are divided on this question. Some would agree with me (technically, they recognise `determiner phrases'), and others would disagree, believing that the should be treated as part of a phrase whose main word is coffee, a `noun-phrase'. It isn't just a matter of taste; there are relevant facts and arguments for and against. For example, in English the belongs to a group of words called `determiners' which includes a, this, that, some, any, his and some others. A word such as book needs a determiner (we cannot say **I bought book, but any determiner before book makes the sentence grammatical); but most determiners can be used without a following common noun (I bought this/that/some/any/his). This suggests that the determiner is the obligatory part of a phrase consisting of a determiner plus a common noun, so it should be the determiner, rather than the common noun, that carries the external relations (e.g. the object link to drank). (Another controversial question is how determiners fit into the system of word-classes; according to Diagram 1 they are a type of pronoun, which makes them also a type of noun, but many linguists treat them as a completely separate word-class. Once again there are relevant facts, and it is possible to debate the issue.)

One advantage of taking the rather than coffee as the object of drank is to reduce the differences between English and other languages. Take the Turkish translation of this sentence, for example: Kahveyi içtik. The ending -i on kahveyi gives two important pieces of information about this word: it is definite, and it is the object of the verb içtik. The definiteness is conveyed in English by the determiner (in this case, the), but our analysis gives the English determiner the object function as well. It has often been suggested that the word-parts of languages such as Turkish could easily be recognised as separate syntactic words, which are combined into single words only for phonological purposes; so maybe kahveyi could also be written kahvey + i (with `+' to show that the two words belong together in phonology). There may be good arguments against this analysis, but if we accept it the WG analysis might be as in Diagram 2.

Diagram 2

1

I have labelled -i simply `OD' (for `object definite'), since I have no reason for thinking it is a pronoun (in contrast with English the). However classification is less important than the `geometry' of the diagram. If we compare this diagram with the one for We drank the coffee, the two main differences between English and Turkish stand out clearly: the object follows the verb in English but precedes it in Turkish, and there has to be a separate subject in English, but not in Turkish. But more subtly, in both languages the word for `coffee' is only indirectly related to the verb, by a little word that marks definiteness and which in Turkish, but not in English, can only carry one grammatical function, `object'.

3. Dependencies

The most important part of a syntactic analysis is what it tells us about the relationships among the words - how they combine to make a meaningful sentence. (In contrast, the classification of the words is much less less relevant to the meaning, and correspondingly less important.) This information is carried in a WG analysis by the labelled arrows, so we need to consider these a little more carefully. They are also probably the most controversial part of the theory, which is another reason for explaining them. What I have said so far about the arrows is that each one points towards the word that has the grammatical function indicated by the label; the arrow which carries the label `s' (for `subject') points at we, the one labelled `o' (for `object') points at the (coffee), and so on. However, we can now take a step further: each of these arrows shows a dependency between two words. The grammatical function is always the function of the dependent in relation to the other word, so the arrow always points towards the dependent word. In our example, both the subject and the object depend on the verb, and (according to my analysis at any rate), the common noun (coffee, kahvey) depends on the determiner or object marker (the, -i). The main idea behind this notion of dependency is that each word depends on the word which links it to the rest of the sentence, and which explains why it is used. The latter is often called the `head' of the former, so in we drank, we is a dependent of drank and drank is the head of we. Table 1 lists some of the most common types of dependency found in English, and Diagram 3 shows how a more complicated English sentence is analysed.

Diagram 3

1

head word / dependent word / function / order / example
verb / noun / subject / we drank
noun / adjective / preadjunct / nice coffee
noun / noun / " / London hospital
adjective / adverb / " / really nice
verb / noun / object / drank it
verb / adverb / particle / fended off
preposition / noun / complement / in London
determiner / common noun / " / the coffee
any / preposition / adjunct / lived in (London)

Table 1

This kind of dependency-based analysis allows some very simple generalisations about word order. From my very limited knowledge of Turkish I gather that most dependencies are head-final - that is, the head word follows the word which depends on it. For example, verbs follow their objects (as well as their subjects), postpositions are used instead of prepositions, and so on. English word order tends to be the other way round, with heads preceding their dependents (e.g. verbs precede their objects and we have prepositions instead of postpositions); but this is only a tendency, as there are a significant number of dependencies with head-final order. In Table 1 the first four dependencies are head-final, while the last five are of the more typical head-initial type.

1

This kind of analysis is very controversial in modern linguistics because it treats phrases as secondary by-products of the dependencies. In short, (with one major exception which I shall discuss below) all the units of syntax are single words. For example, the analysis of We drank the coffee recognises just four units, one per word, and does not recognise the coffee as a separate unit which could be classified or bear a grammatical function; and similarly for our second example in Diagram 3, where its London hospital closure plans is not treated as a single unit, but as five separate words. Most modern theories of syntax put `phrase structure' at the centre of syntax, and derive the relations between individual words from their relationships to the phrases. In contrast, the WG analysis treats the word-word relationships as basic, and leaves the larger units implicit. For example, the fact that London depends on hospital means that these two words `belong together', in contrast with its and London which are not related in this way (although they are physically next to each other in our sentence). The reason for ignoring phrases in this way is quite simple: once the word-word dependencies have been analysed, the phrases are simply redundant. A phrase is just a group of words which all depend, directly or indirectly, on one single word, so adding phrase-units to the analysis would add no extra information whatever. This is a testable claim on which WG rests, and may (of course) be wrong; but so far I know of no good evidence to the contrary. The exclusive focus on word-word relations has a very important practical consequence: the whole system is much easier to learn and to diagram than one which uses phrases as well as words.