Phrase Structure and Constituency

Structural ambiguity and PP/modifier attachment

Structural ambiguity occurs when one string of words can be assigned two or more different constituent structures, so that each structure yields a different meaning. The difference lies in bracketing rather than in the words themselves, and it is revealed by drawing a tree for each reading. It must be distinguished from lexical ambiguity, in which one word form carries several senses, and from semantic role ambiguity, in which the same noun phrase can be read as bearing different thematic roles such as agent or theme. A sentence can be syntactically ambiguous and also show semantic role ambiguity, because the two readings often come from the same structural difference.

A classic case is the prepositional phrase (PP) attachment problem. A PP can attach to a verb phrase, where it functions as an adjunct or is selected as a complement by the verb, or it can attach to a noun phrase, where it functions as a postmodifier. In 'The students watched the man with the telescope', the PP may modify the man or describe the means of watching. Attachment is decided by grammar and by the meaning of the words, not by linear proximity. A verb's selectional restrictions specify the kinds of arguments it can take. For instance, a verb of cutting selects an instrument phrase such as 'with a knife', so such a PP attaches to the verb.

A useful test is to ask which readings the sentence permits. When several PPs follow one another, each one may attach to the verb or to the nearest preceding noun, and the number of possible readings grows quickly. Some readings are excluded for reasons outside syntax. A reading that places an entity in a location conflicting with the time reference of the sentence cannot be the intended meaning, so it is not a possible interpretation.

Compound nouns show the same problem at the level of word structure. In English, the rightmost noun of a compound is normally the head, so 'history teacher' denotes a kind of teacher, and the modifier to its left is a non-head. Headedness passes upward: a phrase inherits its head from its head daughter, so the head of the whole constituent is the head of the daughter that carries it. A compound such as 'Roman history lecturer' can be bracketed as [Roman [history lecturer]] or as [[Roman history] lecturer], and each bracketing gives a different head for the internal nodes.

Finally, to analyse a sentence one must first identify its constituents: count the noun phrases, the verb phrase and the PPs, and decide which PPs are adjuncts. Subjects and objects are arguments of the verb, whereas adjuncts such as temporal or locative phrases are optional and can be removed without making the clause ungrammatical. Counting determiners is a separate task from counting noun phrases, since a noun phrase may have no determiner and a possessive word also counts as a determiner.

Phrase structure rule derivations

Phrase structure rules are rewriting rules of the form A \rightarrow X\ Y that expand a category symbol into a sequence of categories. Parentheses mark optional elements, so a rule such as VP \rightarrow (Adv)\,V\,(PP) generates the verb alone, the verb with an adverb, the verb with a PP, and so on. A derivation begins from the start symbol S and repeatedly rewrites non-terminal symbols until only terminal symbols remain. Terminal symbols are the words of the language, while non-terminal symbols are category labels such as S, NP, VP and PP. Non-terminals are rewritten by productions and never appear in the final string, so a derivation ends when every symbol in the string is a terminal.

A grammar generates a sentence when some derivation yields exactly that string. To check a sentence against a set of rules, first assign each word to a category, then ask whether the category sequence can be built by expanding the rules. Recursion is possible when a category reappears in its own expansion, so a VP may contain a TP that itself contains a VP. Rules also limit what cannot be generated. If the NP rule has no postmodifying PP, a PP cannot attach to a noun. If the VP rule allows only one PP, two PPs cannot both appear as sisters of the verb. If the VP rule has no direct object NP, a transitive object cannot be generated at that position.

Garden-path sentences make the derivation of the subject NP tricky. Consider a structure like 'The boat floated down the river sank'. The early parse treats 'floated down the river' as the main predicate, but the correct parse treats it as a reduced relative clause modifying 'boat', so 'sank' is the main verb. Once the parse is corrected, the VP consists only of the final verb, and everything before it belongs to the subject NP. The size of the VP is therefore decided by identifying the main verb, not by the first verb that appears.

Morphological structure can be drawn with the same tools. A prefix such as re- attached to a verb keeps the verb category, as in 'rewrite'. A suffix may change the category: -able derives an adjective from a verb, as in 'washable', and -ness derives a noun from an adjective, as in 'happiness'. In a tree, the node formed by a prefix has the category of its base, and the top node receives the category set by the suffix.

Basic word order typology

Word order typology classifies languages by the basic, unmarked order of subject (S), object (O) and verb (V) in declarative sentences. The six logically possible orders are SOV, SVO, VSO, VOS, OVS and OSV. Among them, SOV and SVO are by far the most common among the world's languages, VSO is found in a smaller number of languages, and VOS, OVS and OSV are rare. The most frequent pairing of basic orders is therefore SOV with SVO, since these two dominate the typology in both frequency and distribution.

The basic order is the neutral order from which other orders are derived by movement of constituents or by discourse-motivated fronting. Orders obtained by movement away from the basic pattern are called marked, and they are usually licensed by information structure or by a specific grammatical operation. If an SOV language allows movement that produces OSV or VSO, these derived orders are marked relative to the default SOV. Calling an order marked does not mean it is ungrammatical. The derived order is fully grammatical but departs from the unmarked pattern.

Verb-second (V2) is a constraint in which the finite verb occupies the second constituent position of a main clause, whatever constituent precedes it. German main clauses show this pattern, for example when a time adverbial is placed first and the finite verb is followed directly by the subject. Kashmiri is widely described as a V2 language, with the finite verb following the first topicalised constituent. By contrast, Kannada, Konkani and Konyak are typically described as basic SOV languages without a V2 constraint.

Case marking and postpositions are typical of SOV languages. In a Japanese-type language, the marker -ga marks the subject, the marker -o marks the object, and -ni marks the location of existence, as with the existential verb imasu. The marker -de marks the place where an action happens, and -e marks direction. The verb comes at the end of the clause, and the object precedes it. Translation exercises require matching the verb stem to its meaning: ik- means 'go', and tabe- means 'eat'. For example, 'Taro-ga neko-o miru' means 'Taro sees the cat', with the object before the verb.