short title 1

Pied-piping in Word Grammar[*]

richard hudson (13.5pt)

Abstract

Pied-piping, as in the book in which I found it, is a serious challenge for any theory of syntax because the preposition (in) takes its position from its complement (which) rather than, as expected, from the word on which it depends (found). Theories based on phrase structure, whether transformational or not, explain this by allowingthe wh-feature of which to ‘percolate’ up to the whole PPin which, but cannot explain the absence of the ‘wh-prepositions’ that we might expect to develop diachronically, nor the non-wh semantics of the supposed wh-PP. Word Grammar (WG) offers a completely different alternative which focuses directly on the one exceptional characteristic of pied-piping: word order. Since 2007, WG has treated word order in terms of the ‘landmark’ relation, which is separate from dependency structure; this allows the default rules for deriving landmarks from dependencies to have exceptions. One such exception is that the pied-piped preposition simply has no landmark at all. This account explains the peculiarities of pied-piping as a purely surface matter of word order without implications for classification or semantics.

1 Introduction

This paper is about the analysis of sentences like (1).

(1)This is the book in which I found it.

The challenge is to explain why in stands before which rather than in its expected place after found it, as in (2) and (3).

(2)This is the book which I found it in.

(3)I found it in this book.

The problem is that the rules for front-shifting wh-pronouns such as which do not, in themselves, allow the front-shifting of prepositions. As we all know, the only reason why in stands at the start of the relative clause in which I found it is that it has been ‘pied-piped’[1] into this position by which; the preposition itself has no grammatical reason to be there.

Moreover, this is a very exceptional case of a head word taking its position from its dependent, a reversal of the usual pattern which moves the problem out of mere descriptive linguistics onto the plane of general linguistic theory. But notice that the only issue, from a theoretical point of view, is the order of words (and in particular, the position of in). Nothing else is exceptional in (1) – in fact, apart from the position of in,(1) has exactly the same syntactic structure as (2), paired with exactly the same semantics and morphology.

Another challenge is the difference between interrogative and relative clauses. Interrogative wh-pronouns allow a different range of pied-piping patterns from what relative wh-pronouns do (Huddleston and Pullum 2002:910, 1040). Huddleston and Pullum (who use the term ‘upward percolation’ rather than pied-piping) give the following examples. I have italicized the phrase whose position is affected by this percolation.

(4)In which drawer do you keep the bank statements?

(5)*The president of which country was she?

(6)What size shoes do you take?

(7)How big a hole did it make?

These interrogative examples show that (in Huddleston and Pullum’s terminology) the wh-feature can percolate up from the interrogative pronoun to a preposition (4), but no further (5); and that it can also percolate up to a pre-modified noun (6) or through an adjective to the determiner a(7).

This pattern is quite different from the one found with the relative pronouns in the next examples:

(8)the curtain behind which Kim was hiding

(9)She’s just sat her final exam, the result of which we expect next week.

(10)They will be involved in several projects, one of the most important of which will be to find ways to use the new superconductor in chips that can provide the brains of a new generation of supercomputers.

(11)The many varieties of mammalian skin secretions perform a wide range of functions, prominent among which is sexual attraction.

(12)I became disturbed by a ‘higher criticism’ of the Bible, to refute which I felt the need of a better knowledge of Hebrew and archaeology.

(13)They take a rigorous examination, passing which confers on the student a virtual guarantee of a place at the university.

These examples show that (as in interrogative clauses) the relative pronoun’s influence can percolate up to a preposition (8). This is the only possibility in a restrictive relative clause, but in non-restrictive relatives there are many other possibilities: percolation up through a preposition and noun to a determiner (9), through a chain of six prepositions, nouns and determiners (10), through a preposition to an adjective (11), through a verb to the infinitival to(12), or to a gerund (13). This freedom is further illustrated by the invented examples used by Ross to illustrate the process of pied-piping, (14):

(14)reports, the height of the lettering on the covers of which the government prescribes

As an aside on Word Grammar (WG), it is worth pointing out the consequence of taking determiners as heads (as in these books, where books depends on these rather than the other way round). This analysis means that Huddleston and Pullum’s examples such as Which car did she take? do not involve percolation because the head of which car is which. Similarly, in what size shoes it is what, rather than size, that depends directly on shoes, which produces a shorter percolation path than Huddleston and Pullum envisage. Nevertheless, most of their examples are as challenging for WG as for any other theory.

The discussion so far has established two facts about pied-piping (or upward percolation):

  • It is theoretically exceptional in allowing a phrase’s position to be determined by a word other than its head.
  • The details of the phrases influenced in this way by a wh-word vary according to whether the wh-word is an interrogative or a relative, and whether the relative clause is restrictive or non-restrictive.

In short, the solution must combine general theory with descriptive flexibility.

Given that pied-piping really is exceptional, any theory of syntax needs a sub-theory which explains exactly how this exceptionality can be accommodated in the general theory. All the major theories offer some kind of account of pied-piping (Falk 2006), but because of other differences between the theories they are hard to compare and evaluate. As just one example of an established analysis, consider the one offered by Head-driven Phrase Structure Grammar (Pollard and Sag 1994), which exploits the very rich and flexible feature structures which are part of that tradition and which allow features of daughters to percolate up to the mother node. Feature-percolation is a fundamental and general property of phrases, but normally the percolated features are those of the phrase’s head; in WG terms, the phrase and its head word are identical (which is one of the reasons why I argue that the phrase is redundant). However, pied-piping is exceptional in allowing percolation of features of a non-head wh-pronoun; for instance the phrase behind which in (8)has not only the regularly percolated features of behind (which make it a PP), but also the wh-feature of which.

The weakness of this analysis is that the only evidence that in which has the wh-feature is its position. Neither the phrase as a whole, nor its head, has any of the expected morphology or semantics of a wh-word which might give a form such as *whehind which or a meaning such as ‘x such that x is a position behind y’ (where ‘y’ is the referent of which). To pursue the semantic point, if behind which had wh-type semantics in the curtain behind which Kim was hiding, then behind which would refer to the curtain, but in fact it refers to a place behind the curtain. In short, the wh-feature on behind which has different consequences from the one on which, so it cannot be the same feature and percolation is the wrong machinery. Whatever alternative machinery is needed must separate the consequences for word order from all the other consequences of the wh-feature.

The purpose of this paper is to propose a WG analysis of pied-piping which does separate word order from other characteristics. The next section introduces the parts of WG theory that are relevant to this analysis, which section 3 applies to the English facts sketched above.

2 Word order in WG

The explanation for word order in WG, as in every other theory, builds on the abstract dependencies that hold words together. These dependencies are, of course, relations, but in WG relations are concepts (‘relational concepts’) which, like entity concepts, exist in order to bring together a bundle of properties. To take an entity example, the concept ‘bird’ combines propertieswhich correlate strongly as properties of the same entity, such as having two legs, flying and having a beak; and similarly, a relational concept such as ‘on’ combines properties such as ‘aboveness’ and ‘support’. In the domain of syntax, dependencies are the means for bringing together diverse relations such as syntactic valency, semantic valency, morphological selection, agreement and (of course) word order. The classic case is the dependency ‘subject’, which typically combines a dozen or more characteristics (Keenan 1976).

In this view of concepts, each concept, including relational concepts, may be defined by its relations to other concepts; but it is separate from these other defining concepts. For example, take the concepts associated with a typical Western cooked meal: starter, main course and dessert. Each of the courses is an entity with various properties determining the amount and kind of food that it consists of, but one of their properties is their sequential ordering, with the starter before the main course and the dessert after it. Sequential ordering is a relation, and must be represented mentally as a relation link even though our linear writing allows us to show it on the page as a left-right sequence; so in the case of a meal we must recognise two relations between the main course and the starter: the ‘starter’ relation, which recognises that this starter ‘belongs’ to this main course, and the ‘before’ relation (a subdivision of Langacker’s ‘landmark’ relation - Langacker 2007, Hudson 2007). This typical arrangement is shown in Figure 1, where the oval shapes distinguish relational concepts from entities, and the landmark relations are shown above the entities (a new notational convention for WG, but one which I shall exploit when we return to syntax). When applied to a particular meal, we can predict that the main course’s starter will be before it, and its dessert will follow it. However, thanks to default inheritance, exceptions are allowed, so we can imagine a meal where the starter is accidentally delayed, so the eaters decide to leave it till after the main course; in that case, its other properties will justify classifying it as a starter, but it will have an exceptional sequential order. In short, landmark relations are separate from other ‘dependencies’ (illustrated here by the ‘starter’ and ‘dessert’ relations).

Figure 1: Sequential order in a meal

Word order is a particular case of the same sequential order that we find in meals, and uses the same landmark relations. The ‘before’ relation of cows to moo is an example of just the same relation as we find between starters and main courses. When applied to dependencies and word order, the principle of separating order from other relations means that a dependency such as ‘subject’ between a verb and its subject is separate from the order of these two words; for instance, in Cows moo we can recognise not one but two syntactic relations between cows and moo: a subject relation and a ‘before’ relation. The grammar of English allows the latter to be predicted from the former, but the two relations are nevertheless distinct.The analysis is shown in Figure 2, which uses the same notation as in Figure 1, with the addition of a vertical arrow which conventionally picks out the root word, which has no landmark.

Figure 2: Dependency and landmark relations combined in Cows moo.

Dependencies are highly relevant to landmark relations, because any word-order rules that the grammar may contain will start from dependencies; for instance, the English rules which allow Cows moo but not *Moo cows require the verb’s subject to be before it, with a range of exceptional cases such as Do cows moo? and Here comes John, where the subject is after the verb. By default, a word’s dependents all take that word as their landmark; indeed, one of the reasons why dependencies are fundamentally asymmetrical is that the dependent takes its position from the parent (the word on which it depends), and not the other way round. This fundamental principle is what is at stake in pied-piping, where a dependent wh-word determines the position of its parent; so in which takes the position required by which rather than by its head in.

Another fundamental principle of word order requires what we might call ‘phrasal integrity’: each of the phrases defined by dependency structure should be uninterrupted – i.e. it should form a continuous chain of words which does not contain any other words. Just as in phrase-structure analysis, a word’s phrase consists of that word plus the phrases of any dependents that it may have. For instance, in I like red wine, the phrase defined by wine consists of red wine, and in this sentence the phrase has integrity because it is not interrupted by words from any other phrase. But in *I red like wine, the phrase has lost its integrity because red and wine are separated by like, which is not part of that phrase. However intuitively obvious this principle may be, it is surprisingly hard to formulate in terms of dependencies, and especially so when rich dependency structures are allowed as in WG. In simpler versions of dependency grammar it is expressed as the requirement that structures should be ‘projective’ (Robinson 1970), in the sense that dependency lines should not cross (as they would in any analysis of *I red like wine). WG has seen a series of slightly different analyses, variously called ‘adjacency’ (Hudson 1984:98, Hudson 1990:117), ‘the no-tangling principle’ (Hudson 1998:20), ‘landmark transitivity’ (Hudson 2007:139) and ‘the best landmark principle’ (Hudson 2010:171). Leaving aside these important theoretical details, the principle means thatwords which share the same landmark stick together even in languages with ‘free word order’. And, of course, in languages like English the before/after direction is determined by the dependencies concerned.

However, not every word has a landmark; most obviously, a non-dependent word such as moo has none, but this is as expected if landmarks follow from dependencies: no dependency, no landmark. But in principle, the separation of landmarks from dependencies allows exceptional dependencies to exist without landmarks. This possibility will provide the basis for the WG explanation for pied-piping which I develop in the rest of this paper.

3 Wh-words and extraction

Pied-piping is always triggered by wh-words, so we start by considering the special properties of wh-words. As we have already seen, pied piping has very different constraints in interrogative, restrictive relative and non-restrictive relative clauses, so eventually we shall have to distinguish the wh-words found in these clauses, but they all share the following special characteristics.

Syntactically, they are special because of their valency requirements, which allow them to face in two directions at the same time. Like typical pronouns, they can be used as dependents of other words, seen most simply in elliptical examples like (15).

(15)(Someone has borrowed it, but) I don’t know who.

This allows them to be part of a ‘higher’ clause, but they are also part of the ‘lower’ clause (the relative or interrogative one) which follows them. This clause is rooted in the verb which is the wh-word’s complement. Although this verb is optional for interrogative pronouns (as in (15)), it is obligatory for relative pronouns, even when it could be guessed from the context.

(16)Someone has borrowed it, but I can’t identify the person who *(borrowed it).

The possibility of a complement verb is shared with many other kinds of subordinate clauses, but what is special about wh-words is that they are also treated as dependents of the following clause; in other words, their relation to the complement verb is one of mutual dependency. This can be seen most easily in an example like (17).

(17)Who came?

On the one hand, came depends on who as the latter’s optional complement (optional because it can be omitted by ellipsis, as in the one-word question Who?); but on the other hand, who depends on came as the latter’s subject. This is why wh-words can be described as facing in two directions at the same time: towards the lower clause on the right, and towards the higher clause (typically) on the left. For example, Figure 3shows the structure for I know who came, using the notation introduced above in which the arrows above the words show landmark relations and those below the words show dependencies; the labels ‘s’, ‘o’ and ‘c’ distinguish subjects, complements and objects. Notice how the landmark relations, even in a simple example like this, are to some extent out of step with the dependency relations, because the mutual dependency of who came is not matched by a mutual landmark relation.