Interfaces: Linearization

I think that the essence of syntax proper is linking lexical items together in various ways. In the last post I tried to show that the dependency structures we obtain by linking lexical features together just are the tree-like structures with movement we are familiar with, disguised. I think this is important, because it reveals grammaticality to be verifying that features of lexical items were checked off against each other in the right way, which is different from how we usually think of things.

Of course, these abstract structures have very little contact with the rich empirical data that we would like to account for; they account for perhaps the number and identity of words (i.e. lexical items) in sentences. While this is already of some interest, it is not of that much interest: the the on alligator the of anaconda side road ate and the alligator ate the anaconda on the side of the road are identical w.r.t. the numerosity of their words, but one is word salad, the other an every day occurrance in Florida. Much more interesting would be if we could account for not only the numerosity, but also the linear order of words.¹

As we have shown in the previous post how to move from dependency structures to (multiple-dominance) trees and back, we will linearize our structures by specifying how to linearize trees. There have been many suggestions as to how to spell out (linearize) an unordered multiple dominance structure, or equivalently, a tree with multiple copies. As we know how to linearize an ordered tree, we can try to construct an ordered tree from our unordered MDS. To do this, we must solve two independent problems.

1. turn an unordered tree to an ordered tree

   essentially: how to order sisters

2. turn an unordered MDS to an unordered tree

   essentially: how to decide what gets pronounced in each multiply dominated position

I will begin with what I take to be the simplest answer to these questions

1. specifiers are pronounced before heads before complements²

   This should be made more precise as:

   specifiers are pronounced to the left of their sisters, complements to the right.

2. each multiply dominated expression is to be exclusively pronounced in its structurally highest position³

   In other words:

   an expression is pronounced in the position which checks its last feature

Ordering

The simple answer to the ordering problem (SPEC-HEAD-COMP) treats every lexical item (in every language) the same; you are pronounced to the left of the expression that checks your first positive feature, and to the right of all expressions that check your other positive features. Furthermore, an expression checking a later positive feature of yours will be pronounced before (i.e. to the left of) an expression checking an earlier positive feature of yours.

It is a commonplace to analyze German as having “complement-head” order in the verbal and tense domains, but “head-complement” order elsewhere. This ‘mixed’ word order property of German, if we accept it at face value, means that we cannot hold onto this simple answer to the ordering problem.⁴ Instead, German word order properties are naturally stated by reference to particular lexical items (verbal and temporal lexical items require Comp-Head order, others Head-Comp order). We could implement this lexical item-dependent word order variation at the featural level, by specifying for each positive feature on each lexical item, whether its checker should appear to the left or right of the head.

This purely interface information needs to be recorded somehow. It is easiest to attach linearization instructions to the individual features.⁵ We can write \(\bullet\hspace{-.5pt} x\) for a positive feature which wants its checker to be on its left, and \(x\hspace{-.5pt}\bullet\) for a positive feature which wants its checker to be on its right (so the dot is on the side that the checker should be linearized on). With this convention, we can write lexical items for German words that disagree about what side their complements should appear on:

Head-Comp	Comp-Head
\(\textsf{jeder} : n\hspace{-.5pt}\mathop{\bullet}. d. k\)	\(\textsf{lachen} : \bullet\hspace{-.5pt}d. V\)

Chains

The simple answer to the multidominance problem⁶ again treats every lexical item the same; it and its maximal projection are pronounced in the position checking their last feature.

There are two obvious ways of generalizing this. The first way is parallel to the way we allowed each lexical item to determine for itself the pronunciation orders of its dependents above. This amounts to allowing a lexical item to specify which of its features it will be pronounced in.

The second generalization of copy spellout takes control away from the lexical item itself, and moves it to the interaction of its governing heads. This seems most reminiscient of actual practice, where a governing head which can support pronunciation is said to have an EPP feature. The intuition underlying this generalization is that a lexical item will prefer to be pronounced as high as possible, but being pronounced in a certain position is possible only when the position in question is able to host an overt specifier.⁷ In contrast to our previous generalizations, which were stateable in terms of the properties of a particular lexical item in isolation, we cannot predict on the basis of a particular lexical item in which of its negative feature positions it will ultimately be pronounced - its ultimate pronunciation position is only determinable on the basis of properties of the structure in which it finds itself. When multiple positions compete for hosting the pronounced version of the phrase, we will say that the (structurally) highest one wins.⁸ If no position wants to host the phrase, it is pronounced in its first negative position.

We can indicate on a feature whether it is willing to support pronunciation or not by adding a tilde above the name of a feature which does not support pronunciation. Because saying ‘tilde-marked’ and ‘non-tilde-marked’ is a mouthful, I will say instead weak and strong.

Weak	Strong
•x̃, x̃•, x̃	•x, x•, x

Our first generalization (where each lexical item decides for itself which of its negative features will host its pronunciation) amounts to allowing tildes to appear on negative features only. The last strong in a feature bundle is the one where the lexical item’s maximal projection will be pronounced. If all features are weak, it will be pronounced in the first one. This generalization would be able to capture a situation where a particular position was able to host expressions of a certain type, but where some expressions of that type were always pronounced lower than that position.

Our second generalization (where the ultimate pronunciation site of a maximal projection is a global property of the structure) amounts to allowing tildes to appear on positive features only. The maximal projection of a lexical item will be pronounced at the position of the last of its features checked by a strong feature, or in the position of its first negative feature if all of its negative features are checked by weak ones. This generalization would be able to capture a situation where some words allow overt movement to their specifiers, but other words require the same expressions to move covertly. Some people think of case in English as working in this way. Clearly, all DPs in English can be pronounced in the position which assigns them case (esp. nominative case). However, many people analyze objects as being in their in-situ positions, despite being assigned case. This could be accounted for by assigning a strong positive case feature to the T head, and a weak positive case feature to (say) the v head.

Of course, we could combine these generalizations, allowing tildes to appear willy-nilly in feature bundles. The strong/weak negative features in a lexical item’s feature bundle indicate where a lexical item is willing to be pronounced. The strong/weak positive features in a lexical item’s feature bundle indicate where a lexical item is willing to overtly host something. Then the maximal projection of a lexical item will be pronounced in the highest of its strong negative features which is checked by a strong feature, or in its first negative feature position if none of its strong features are checked by strong features.

---

1. We would also like to account for some aspects of sentence meaning, but that is for next semester. ↩︎

2. A complement is the expression linked to a heads first positive feature, a specifier is an expression linked to any other positive feature other than the first. ↩︎

3. How is structural height determined? As movement is always to a c-commanding position, we could just choose the position which c-commands all the others. However, c-command might give us contradictory results in multiple dominance land. Recall that c-command can be defined as: a node c-commands its sister and everything in it. Here, ripened c-commands the DP which mango, but the DP which mango c-commands its sister, the C', and thus also ripened. In this case, it doesn’t really matter (because Spec-C asymmetrically c-commands Comp-T), but one can imagine situations where these loopy c-command relations come back to bite one on the butt. Gärtner (op. cit.) shows how this can be done correctly. We will avoid this question by focussing on the derivation of structures. ↩︎

4. Of course, we could keep this simple answer if we made our analysis more complicated, taking at least one of the two word orders to be the result of additional movements. ↩︎

5. This contaminates the representation, combining true syntactic and merely interface information. ↩︎

6. This problem is not particular to the multidominance representation. It appears in exactly the same way with multiple coindexed copies. ↩︎

7. A particularly fine-grained version of this idea comes in the form of Koopman’s complexity filters, where each position may place restrictions on not only whether it will host an overt expression but also on the syntactic and prosodic shape of this expression. ↩︎

8. That is, the one which checks the latest feature. ↩︎