We will suggested that we can assign words in sentences to categories on the basis of their grammatical properties. However, it should be pointed out that simply specifying what category a particular word in a particular sentence belongs to does not provide a full description of the grammatical properties of the relevant word. For example, categorizing he as a pronoun in (72) doesn’t tell us in what ways he differs from other pronouns like e.g. I/us/you/her/it/them – i.e. it doesn’t tell us about the (third) person, (singular) number, (masculine) gender and (nominative) case properties of he. In other words, there is a great deal of additional grammatical information about words which is not represented by simply attaching a category label to the word – information which provides a finer level of detail than relatively coarse categorial descriptions. This information is generally described in terms of sets of grammatical features; by convention, features are enclosed in square brackets and often abbreviated (to save space). Using grammatical features, we can describe the person/number/gender/case properties of the pronoun he in terms of the features [3-Pers, Sg-Num, Masc-Gen, Nom-Case] i.e. ‘Third-Person, Singular-Number, Masculine-Gender, Nominative-Case’. Each of these features comprises an attribute (i.e. a property like person, number, gender or case) and a value (which can be first/second/third for person, singular/plural for number, masculine/feminine/neuter for gender, and nominative/accusative/genitive for case).

An adequate description of syntax also requires us to specify the selectional properties of individual words (e.g. what kinds of complement they can take). We can illustrate the importance of selectional information by considering what kinds of word can occupy the position marked by — in the sentences below:

A categorial answer would be ‘A verb’. However, we can’t just use any verb: for example, it’s OK to use verbs like go/fly, but not verbs like find/stay. This is because different verbs select (i.e. ‘take’) different types of complement, and verbs like go/fly select a to-expression as their complement but verbs like find/stay do not. But the story doesn’t end there, since each of the structures in (75) requires a different form of the verb: in (75a) we can use the infinitive form go, but not other forms of the verb (cf. He might go/∗going/∗gone/∗goes/∗went to Paris); in (75b) we can only use the progressive participle form going (cf. He is going/∗go/∗gone/∗goes/ ∗went to Paris); and in (75c) we can only use the perfect participle form gone (cf. He has gone/∗go/∗going/∗goes/∗went to Paris). This in turn is because the auxiliary might selects (i.e. ‘takes’) an infinitive complement, the progressive auxiliary is selects a progressive participle complement, and the perfect auxiliary has selects a perfect participle complement. In other words, a full description of the grammatical properties of words requires us to specify not only their categorial and subcategorial properties, but also their selectional properties. It is widely assumed that the selectional properties of words can be described in terms of selectional features. For example, the fact that progressive be selects a progressive participle complement might be described by saying that it has the selectional feature [V-ing] – a notation intended to signify that it selects a complement headed by a verb carrying the -ing suffix.

As far back as his 1965 book Aspects of the Theory of Syntax, Chomsky argued that all the grammatical properties of a word (including its categorial properties) can be described in terms of a set of grammatical features. In work in the 1970s, he argued that the categorial distinction between nouns, verbs, adjectives and prepositions can be handled in terms of two sets of categorial features, namely [±V] ‘verbal/non-verbal’ and [±N] ‘nominal/non-nominal’. More specifically, he suggested that the categorial properties of nouns, verbs adjectives and prepositions could be described in terms of the sets of features in (76) below:

What (76) claims is that verbs have verbal but not nominal properties, adjectives have both nominal and verbal properties, nouns have nominal but not verbal properties, and prepositions have neither nominal nor verbal properties. This analysis was designed to capture the fact that some grammatical properties extend across more than one category and so can be said to be cross-categorial. For example, Stowell (1981, p. 57 fn. 17) notes that verbs and adjectives in English share the morphological property that they alone permit un-prefixation (hence we find verbs like undo and adjectives like unkind, but not nouns like ∗unfriend or prepositions like ∗uninside): in terms of the set of categorial features in (76), we can account for this by positing that un- can only be prefixed to words which have the categorial feature [+V]. Likewise, as the following example kindly provided for me by Andrew Spencer shows, in Russian nouns and adjectives inflect for case, but not verbs or prepositions:

Thus, the nouns and adjectives in (77) carry (italicized) case endings (-a is a nominative suffix and -u an accusative suffix), but not the verb or preposition. In terms of the set of categorial features in (76) we can account for this by positing that case is a property of items which carry the categorial feature [+N].

An obvious drawback to the system of categorial features in (76) above is that it describes the categorial properties of a number of substantive/lexical categories, but not those of functional categories. Each functional category seems to be closely related to a corresponding lexical category: for example, auxiliaries appear to be related to verbs, determiners to adjectives, and the complementizer for to the preposition for. One way of handling both the similarities and differences between substantive categories and their functional counterparts is in terms of a functionality feature  , with functional categories carrying the feature [+F], and substantive categories carrying the feature [−F]. On this view, main verbs would have the feature specification [−N, +V, −F] whereas auxiliaries would have the feature specification [−N, +V, +F]; likewise, the complementizer for would have the feature specification [−N, −V, +F], and the preposition for would be specified as [−N,−V,−F]. We shall not speculate any further on this possibility here: for an attempt to motivate such an analysis, see Radford (1997a, pp. 65–68 and p. 84).

Although many details remain to be worked out, it seems clear that in principle, all grammatical properties of words (including their categorial properties) can be described in terms of sets of grammatical features. (See Ramat 1999 on categories and features.) However, in order to simplify our exposition, gradually introducing specific features where some descriptive purpose is served by doing so.