To build a theory of syntax, we need to look at data. We’ll start by considering the following sentences of English.

Here’s an obvious observation about these sentences: All the sentences in (1) are different. For instance, they use different words; they mean different things; some of them are in different tenses. Ask any English speaker, and they would agree: these are different sentences.

Here’s a not-so-obvious observation about these sentences: All the sentences in (1) are the same. Why? Well, they all can be described in the same way. They consist of a determiner followed by a noun followed by a verb. Stated schematically, I can say that,

What have we done? We’ve translated English into a system of variables. For whichever values (that is, words) that we choose for each of the variables (Determiner, Noun, Verb) we will generate a sentence of English.

This means that I, as an English learner/speaker, don’t have to memorize every combination of determiner, noun, and verb in English. All I have to know is this rule, and I can generate an infinite number of sentences (…as long I know the possible determiners, nouns, and verbs in English).

So far we are able to capture only a limited set of data. Obviously it is possible to say many more things in English than sentences with intransitive verbs. Let’s look at some more examples.

Our rule for Sentence_English doesn’t allow us to generate the sentences in (2), because the rule doesn’t know what to do with words like happy, terrible, new, stubborn, and stupid. So let’s make another rule.

By itself, this rule generates all the sentences in (2), but now it over-generates, because according to the rule, every sentence in English must consist of the sequence Determiner-Adjective-Noun-Verb. So according to this new rule, the sentences in (1) are not possible sentences in English because they don’t have adjectives.

So we conclude that we need two rules for English. Now our grammar fragment of English is the following

This is simply a logical step. Rule 1 and Rule 2 both correctly generate the sentences in (1) and (2), and they don’t generate any sentences that aren’t in English, like *Mailman happy slept the. Our rules predict that this sentence shouldn’t be an English sentence, because our rules do not allow this sequence of elements.

Together, Rules 1 and 2 are a grammar fragment. They just describe some of what is possible in English. To continue the process and develop a complete grammar, we’d look at more data, and determine the rules that describe the data. We would do this until we developed a set of rules that could generate all the possible sentences in English, and wouldn’t generate any sentences that weren’t in English. We would then have a complete grammar of English.

However, there are a couple of things we seem to be missing in this grammar fragment. Foremost, these rules don’t really capture the relationships between the words. For instance, intuitively, we have the sense that the happy mailman “goes together,” like it’s a unit, more than mailman slept. For instance, I can replace  the happy mailman with he and the sentence means the same thing: He slept.  I cannot replace mailman slept with anything to get an equivalent sentence: *The happy he. (This is called substitution; we’ll talk about it in the next chapter.) Thus, if we want our rules to truly represent English, then we want to represent that the happy mailman goes together to the exclusion of slept.

We can actually re-write our grammar fragment to capture this grouping. We’ll use the term “Determiner Phrase” (abbreviated DP) to refer to the sequence Determiner-Adjective-Noun or Determiner-Noun.

By adding the term Determiner Phrase, we can now correctly represent that the words the happy mailman go together. Putting a determiner phrase together with a verb makes a sentence in English. I can choose either one of the DP rules, Rule 2 or Rule 3, and put it into Rule 1 to make a sentence. Importantly, our revised grammar fragment does all the things that our first attempt did, that is, it gets all the words in the right order, and does not generate sentences of English that don’t exist.

This level of abstraction raises two issues.

This topic is addressed in the chapter titled Constituency and so we’ll put it aside for now. 

The second question is asking about syntactic categories. A noun has the syntactic category of N. A determiner has syntactic category D. A verb has the syntactic category of V, an adjective has the syntactic category of A, etc. A noun phrase  is a group of words that “acts like” a noun. A determiner phrase is a group of words that “acts like” a determiner, etc.

The names are essentially arbitrary. We’re simply saying that there are classes of things that behave similarly. Table, dog, teacher, brick, honesty, …, all behave in a similar fashion (in some respects). Lick, sit, own, jump, describe, believe, …, all behave in a similar fashion (in some respects).  When we say that something is a “noun” and has category N all we’re saying is that that thing “acts like” the other noun-words. Likewise, when we say that something is a verb, we’re simply saying that that thing “acts like” the other verb-words. 

What does it even mean to “act like noun” or “act like a verb?” In what way are “nouns” differentiated from “verbs” or “adjectives?” In defining syntactic categories, it is tempting to rely on what you were probably taught in school: nouns are “people, places, or things” and verbs are “actions,” adjectives “describe,” etc. That is, we can attempt to define syntactic category using meaning.

This turns out be wrong for a number of reasons. For instance, we can all agree that idea is a noun, but it’s debatable whether it’s a person, place, or thing. Or what about the noun destruction? That’s an action! The destruction of the city by Godzilla describes an event—it’s clearly not a person, a place, or a thing. 

Likewise, not all verbs are actions. Tasha owns a car contains the verbs owns.  Is owning an action? Only under a very loose definition of “action.” What about the verb seems in  Abigail seems tired. Is seem an action?  Doubtful…

Perhaps most problematic for the idea that we can use meaning to define syntactic category comes when we look at languages other than English. For instance, if you’ve studied a Romance language, you’ve learned that certain concepts which are expressed in English using adjectives are expressed in Romance using verbs and nouns. This happens to be a robust strategy found across the world’s languages, occurring beyond Romance languages, like, e.g., Swahili.

If we (as English speakers) were to use meaning to define word category, we might be tempted to call words like hambre, faim, and njaa adjectives meaning “hungry.” This would clearly be wrong, because in fact they’re nouns, as the literal translations suggest. So we simply cannot use meaning to determine syntactic category; it doesn’t help us to define categories within a language, and it doesn’t help us define categories across languages.

So how actually do I know that idea, table, hambre, faim, and njaa are nouns if I can’t use meaning? Or stated differently, how can I uniquely define a class of words in any particular language? There are two kinds of reliable evidence we can use to classify syntactic categories.

Morphology refers to pieces of words. In any particular language, every word class may have its own specific morphological pattern. In English, only nouns can take a plural suffix: ideas, tables, children, etc. Adjectives cannot inflect for plural in English: *reds, *sads,  etc. Only verbs in English can inflect for (be marked with) tense and aspect information. walked, walking,  but *childed, *childing. Only adjectives in English can inflect for the comparative suffix: sadder, *childer. 

Importantly, morphological properties are going to be language specific. For instance, in Estonian, nouns inflect for number (singular or plural), just like in English and also for case (which indicates the role the noun plays in the sentence).

Thus, in Estonian, if a word inflects for both number and case in the same way, then we have evidence that it’s a noun. Just like in English, if a word inflects for tense and aspect, then we conclude that it’s a verb. If a word in English can take the comparative morpheme -er, then we conclude that it’s an adjective.

The second piece of evidence we can use to determine syntactic category is distributional evidence. Every word class goes in a particular position with respect to other word classes. In English, the only thing that can appear in the underlined space is a noun: The ____ is… . That is, only nouns can appear between a determiner and copula. Or stated differently: only certain words can appear between a determiner and copula in English, and we call those words “nouns.” In French, the only thing that can appear in the underlined space in a comparative construction is an adjective.

Or stated slightly differently, the only kind of word that can appear in the configuration above is a word like grande “tall,” or intelligente, “smart,” and we give those words the label “adjective.” (Notice that adjectives in French also have a morphological distinction: they inflect for masculine and feminine gender, in addition to singular and plural.)

Again, distributional properties of syntactic categories are going to be language-specific. We can’t make general statements like “adjectives always precede nouns” because there are many languages in which they come after nouns. Similarly we can’t say something like “verbs always come after subjects” because there are many languages in which the verb comes after the object, or even before the subject. To distinguish syntactic categories for any particular language, you have to look at the properties of that particular language.

Categories in Swahili

Here are some grammatical sentences in Swahili, with their English translations.

From this data, we would conclude that mtoto contributes the meaning ‘the child,’ amefurahi contributes the meaning ‘is happy,’ and amechelewa contributes the meaning ‘is late.’ What are the categories of these words? Well, to answer that we need to look at other words in the language, and determine what else has the morphology and distribution of these words.

Let’s start with mtoto. Is there morphological evidence for its part of speech? Well, we find that there are many nouns in Swahili that start with m-. Importantly, everything that starts with m- is singular, and to form the plural, we just replace this part with wa-. So there is a systematic morphological pattern that characterizes this group of words. (Note that there are hundreds of words with this pattern.)^[1]

So mtoto ‘child’ falls into the category of these words. What category is that? A good guess is that they’re nouns! Remember that the name of the category is somewhat arbitrary; we’re simply saying that there is a group of words that acts similar in Swahili.

What about distributional evidence for mtoto? Well, it turns out that everything in the table above can be appear in the same place in the sentence; so this is distributional evidence that they belong together.

So have a pretty good idea that mtoto belongs together with the words that generally get called “nouns.” What about amefurahi and amechelewa? Well, based on their English translations, we might be tempted to assume that these are adjectives. Why? Because in English, happy and late are unambiguously adjectives. (Feel free to verify this by looking at their morphological and distribution properties.) In Swahili, though, when we look at the evidence, we would probably come to a different conclusion. For instance, consider the morphological evidence in (7).

Here, we’re showing that we can change the tense of amefurahi and amechelewa. By replacing me- with ta-, the statement is now a description of a future event. Is this something that adjectives (in Swahili) do? Well, we can compare. Here are some other sentences that in English are translated with an adjective, in this case mzuri ‘beautiful.’

At the very least, we find that morphologically speaking, the words amefurahi and amechelewa don’t act like mzuri, in that mzuri doesn’t seem to inflect for tense. Moreover, when we look at what else inflects for tense, we see that it is mostly what might call “verbs.”

In other words, the words -chelewa and -furahi (in a more abstract form, taking away the tense and agreement morphology) act like verbs in that they inflect for tense. This is something mzuri, presumably a “true” adjective (though we need to prove this), doesn’t do.

And what about distribution? Our example sentences in (8) already give us a hint. After the subject mtoto the next word is the copula (a form of the verb meaning ‘be’). So if ‘be’ is a verb, then so must amefurahi and amechelewa be verbs.

 Where we’re going

Now that we can identity parts of speech (using distribution and morphology), we can use that information to write our abstract rules about language, as demonstrated earlier. That is, we can translate any language into a system of variables.

We can also determine the category of groups of words. For instance, suppose I wanted to know what category the group of words happy mailman is. From distributional evidence, we would conclude that it behaves like a noun, not like an adjective (or a verb, adverb, etc). For instance, we can’t use this phrase to modify another noun (11a), and it can go in the location that only nouns can go (11b).

We’re going to use this information to help us determine how to “chunk up” the sentences we find, and in turn how to write the rules that are in our grammar fragment.

Advanced

Changing categories

One of the complications that frequently crops up when we’re considering category is that words can change category. Consider the English word cut.  If I asked you what category cut was you could say, “It’s a verb.” Or you could say, “It’s a noun.” Both answers would be correct.

One question that this raises is, for any particular word, which category (if any) is the  “true” form? That is, is the word cut fundamentally a verb which can be turned into a noun? Or is it a noun which can be turned into a verb? A third option is that it’s fundamentally neither a noun nor a verb! It’s simply some abstract semantic information which can be used either as a noun or a verb. And a fourth option is that there are simply two words pronounced as cut: one’s a verb and one’s a noun.

It’s also interesting to think about  the limit of possibilities For cut, there is an upper bound on the category. It can be a verb, it can be a noun, and it can also be an adjective (technically a participial form): the cut bread. That’s probably it. But other words can definitely have more categories. Consider the word down:

Given the abundance of forms for down, why are some “alternations” not allowed? For instance, unlike cut, the noun chop doesn’t naturally have similar noun form: Mary chopped the wood. ??It was a deep chop. Why not? Why are some words restricted in “changing” categories, and others less so?

Morphological complexity in word formation

When a word is morphologically ambiguous between different categories, we call it conversion. So the difference between cut as a verb and cut as a noun is descriptively an instance of conversion.

Sometimes, in fact often, there are morphological reflexes of changing category. Consider the (famous) example of the verb destroy, which has the noun form destruction. Unlike cut, the nominalization of destroy involves at least addition of the suffix -tion, plus a change in the stem of the verb from destroy to destruc-

One question we can ask about such processes is how “productive” they are. By this, we mean, “How many words exhibit a similar change?” For instance, what other verbs  form nouns with -tion? If we find that it’s a lot of verbs, we might say that this kind of nominalization is generally productive. If it only happens with a few idiosyncratic verbs, then we might say that it isn’t fully productive.

English has a few fully productive category changing processes. Consider –ness affixation. Basically any adjective can be suffixed with -ness to form a noun.

Interestingly, such morphological affixation “stacks,” meaning that I can add multiple affixes to a word, each one changing the category.  Consider the following productive affixation processes:^[2]

Other languages have a much richer process of category alternations. In Dumi, a Kirati (Tibeto-Burman) language spoken in Eastern Nepal, adjectives can be made from verbs by putting –sa at the end.

In K’iche’, a Mayan language spoken in Guatemala, verbs can be suffixed with -ib’al to forms nouns meaning “location for VERBing.”

Overt realizations of a category changing affixes introduce a number of interesting theoretical questions. For instance, what actually is a category? Is it just the affix that attaches last? That is, since -ness always “makes” a noun, can we just say that -ness is the category noun? This is schematized below.

But if that’s true, does that mean when we have conversion (e.g., cut_Noun versus cut_Verb), do we have to say that there’s a null affix, corresponding to a noun or verb?

Another question concerns the stacking noted earlier. Take the word breakability. This consists of three distinct morphemes, each associated with a different category. Break is a verb; -able (here spelled –abil) attaches to verbs and makes adjectives;  -ity attaches to adjectives and makes nouns. It’s clear that there is an order in how the elements stack. -Able has to attach to break before -ity can. What about when there is a suffix and a prefix? Is there also an ordering?

Consider the word unbreakable. This also consists of three morphemes: un-break-able. At first glance, it might not seem possible to determine the order of attachment of un- and -able. But in fact, it must be the case that -able attaches first, to form the word breakable, and then un- attaches after that, to form the word unbreakable. If it happened in the opposite order, we would first make the word *unbreak—but this isn’t a word!

More importantly, the idea that there is a determined order of “introduction” for each piece of a word—and that order might not be the same as the linear order—will be a central observation when we start looking not at individual words, but at groupings of  words, otherwise called constituents. What is remarkable is that the same processes that underlie our understanding of syntax also underlies our understanding of morphology (for the most part). These topics are discussed more in depth in LING 527/727.