In this post, I look at a construction that I often saw in drafts of documents I was reviewing. Although the construction is grammatical and concise, readers find it difficult to process.
I explain what this construction is and why it is difficult. I also summarise a published review of some of the vast linguistics literature that analyses this structure. I conclude by recommending that writers should avoid this structure.
What is this difficult structure?
A couple of years ago, I talked with our in-house editors about a construction that our staff had become addicted to. I was seeing several examples of this construction every week. Here is one of the examples:
(1) This Discussion Paper sets out, and seeks feedback on, the Board’s preliminary views on this topic.
I objected to this wording because the reader must wait until the end of the sentence to discover the object of the verbal phrase sets out. I suggested that readers could process the sentence faster and more easily if it were rewritten as in example 2:
(2) This Discussion Paper sets out the Board’s preliminary views on this topic, and seeks feedback on those views.
I suggested why I thought our staff had become addicted to this construction: a mistaken belief that cutting the number of words in a sentence always makes the sentence easier to read. (Of course, I agree that cutting the number of words sometimes makes a sentence easier to read).
My initial diagnosis
In trying to diagnose the problem, I suggested that the underlying structure of the sentence contained 2 instances of the phrase the Board’s preliminary views on this topic, as shown in example 3.
(3) This Discussion Paper sets out the Board’s preliminary views on this topic and seeks feedback on the Board’s preliminary views on this topic.
My diagnosis then derived the wording in example 1 by deleting (or perhaps just not pronouncing) the first instance of that phrase, as shown in example 4.
(4) This Discussion Paper sets out
the Board’s preliminary views on this topic, and seeks feedback on the Board’s preliminary views on this topic.
In this sentence, text in
strike through is deleted (or is not pronounced).
Literature on this construction
My attempted diagnosis was unclear to our editors. So, I looked for literature on this construction and soon found that there is a lot. The literature often calls this construction ‘right node raising’ (RNR), I explain below where that name comes from.
I have recently found a useful summary of the literature on RNR in Right Node Raising, by Barbara Citko, a chapter in The Wiley Blackwell Companion to Syntax (Second Edition, 2017). Her chapter is available at https://onlinelibrary.wiley.com/doi/10.1002/9781118358733.wbsyncom020
Citko notes some limitations on her review. It covers only material published up to 2013. And it considers only transformational generative treatments of this construction.
The rest of this post summarises some key points from Citko’s chapter.
What is right node raising?
To illustrate this construction, I use example 5, a simplified version of one of Citko’s examples.
(5) Terry praised____, and Leslie criticised, movies.
In RNR, a single element, such as movies in example 5, is understood as being ‘shared’ between 2 clauses linked by a co-ordinating conjunction, such as and. The label conjunct is used for the 2 clauses. As is common in the literature on RNR, Citko calls the shared element the pivot (in italics), and uses an underscore to mark the positions where the pivot originated.
In speech, noticeable pauses separate the conjuncts from each other and the final conjunct from the pivot. To mark this separation, Citko uses commas.
Approaches to RNR
Citko identifies 3 main types of approach to RNR:
Figure 1 illustrates how a raising approach might analyse example 5. To start with, there are two copies of the word movies (which is the pivot). One copy follows the verb praised and the other copy follows the verb criticised. Later, both copies move from their original positions and raise rightwards out of the conjuncts to a new position, which is higher in the tree than the original positions.
(Positions in syntactic trees are known as nodes.)
This analysis is the source of the name Right Node Raising. The pivot moves rightwards and raises to a new node that is higher than the original nodes.
Figure 1 shows copies of the pivot at the original location in parentheses (movies). This shows that those original positions contain an unpronounced copy of the pivot (or, in some syntactic theories, some other trace that the pivot was originally there).
An important feature of the movement approach is the final location of the pivot: it is in a position higher than both conjuncts. From that position, the pivot could affect material inside the conjuncts. If the pivot is no higher than the conjuncts, it might not be able to affect them in the same ways. In contrast, in ellipsis and multi-dominance approaches, the pivot does not move outside the two conjuncts.
Figure 2 illustrates how an ellipsis approach might analyse example 5. Initially, there are two copies of the word movies (which is the pivot). Those copies are in the same place as in figure 1. Later, depending on the precise details of the particular ellipsis approach adopted, the 1st copy (after praised) either:
- remains in the same place but is not pronounced; or
- is deleted, but leaves some trace showing that a copy of the pivot was originally at that location.
Also, the 2nd copy of the pivot (the one after criticised) stays where it is.
The parentheses around the 1st copy of movies symbolise the fact that this copy has been deleted (or is still there, but not pronounced).
In an ellipsis approach, an RNR construction behaves in a way that suggests that the pivot (or a copy or other trace of it) is still inside both conjuncts. Also, if RNR involves ellipsis, we can expect that RNR behaves like other ellipsis constructions in various respects, for example: mismatches between the features of the 2 conjuncts; ‘vehicle change’ effects; and whether the deleted (or silent) copy is the 1st copy or the 2nd copy.
Linguists often use the labels:
- mother for a node that immediately dominates one or more other nodes; and
- daughter for a node immediately dominated by its mother.
In the tree diagrams often used by linguists to depict syntactic structure, a mother sits immediately above its daughter(s), with no node intervening between mother and daughter(s).
A fundamental axiom in some syntactic theories is that each node can only have one mother. Unlike movement and ellipsis approaches, multi-dominance approaches do not adopt that axiom. In multi-dominance approaches, a node can have more than one mother.
A multi-dominance approach to example 5
Figure 3 illustrates how a multi-dominance approach might analyse example 5. There is only one copy of the pivot (movies). That copy has 2 mothers: M1 (the node directly above praised) and M2 (the node directly above criticised). The pivot is simultaneously within both conjuncts. And neither of the pivot’s mothers dominates the other mother.
Figure 3 includes:
- a line (free-from, in blue) connecting movies to its mother M1; and
- a line (straight, in black) connecting movies to its other mother M2.
Although I have drawn those 2 lines in different styles, that is only for convenience and legibility. Both lines have the same status and meaning.
In an elliptical structure, there are two copies of the pivot (each with one mother). One deletes when it is identical with the other (though perhaps with some limited mismatches). In contrast, in a multi-dominance structure:
- there is a single copy of the pivot.
- the pivot has two mothers, one in each conjunct.
- the pivot is inside both conjuncts. Thus, the pivot needs to satisfy whatever requirements apply inside each conjunct.
- within each conjunct, the pivot is still in a low position. It must satisfy whatever requirements apply to an item in that low position.
Arguments for and against the approaches
Citko presents the pros and cons of each approach, without necessarily arguing for any particular one. She points out that in her other work she advocates a multi-dominance approach.
- first, whether the pivot moves. Only movement approaches regard the pivot as moving.
- second, if the pivot does move, whether it has one copy or 2 copies. Multi-dominance approaches involve only a single instance of the pivot, whereas movement and ellipsis approaches also involve a copy of the pivot.
I provide below a very high-level summary of her discussion. Her chapter includes examples illustrating each point she discusses.
Does the pivot move?
Evidence that the pivot moves
Citko offers the following evidence that RNR behaves like movement, and so moves the pivot:
- Like movement, RNR is subject to a constraint that prevents movement either of an entire conjunct or of any element contained within one conjunct (the ‘Coordinate Structure Constraint’). Also, like movement, RNR permits simultaneous movement from both conjuncts.
- Movement constructions and RNR constructions behave in similar ways with respect to: (a) ‘parasitic gaps’; (b) extracting indirect objects or post-nominal of-genitives; and (c) verbs like allege.
- There are also some similarities in how movement and RNR with respect to ‘island constraints’ that restrict movement from the left edge of a clause and from the subject. And although there appear to be differences in how these island constraints affect them, those differences might arise from other constraints.
- If the pivot contains a quantifier (such as some or every), the quantified pivot can have scope over a different quantifier sitting within the conjuncts. On widely held assumptions, the pivot can have scope over another quantifier only if the pivot is higher than that other quantifier. To reach that higher position, the pivot must have moved. (But Citko mentions another analysis that might allow this outcome without moving the pivot.)
Evidence that the pivot does not move
Citko offers the following evidence that RNR behaves unlike movement and so does not move the pivot:
- RNR is not subject to all of the ‘island constraints’ that normally prevent movement. Constraints not applying to RNR include those on extraction from question phrases (‘WH-islands’), from complex noun phrases and from ‘adjuncts’.
- Items cannot normally move beyond the right edge of the clause containing them, but RNR can move items further.
- Normally, only constituents can move, but RNR can also apply to pivots larger or smaller than a constituent. A constituent comprises a mother and all its daughters (and their descendants). And RNR can also apply to some types of constituent that cannot move, such as ‘tense phrases’ and ‘noun phrases’.
- RNR can leave prepositions stranded behind in the original position even in languages which do not allow preposition stranding. This can apply both in languages that do not allow preposition stranding generally (eg Irish) and in languages that do not allow it in cases of rightward movement, even though they allow it generally in other cases (eg English). Similarly, RNR can leave complementizers (such as that, if or but) stranded, but movement cannot do so.
- The pivot behaves as if it has not moved and is still present within both conjuncts (a) for the ‘binding’ conditions that determine: whether the pivot contains a pronoun instead of a noun; and (b) in determining whether the pivot may or must contain items (such as any) with ‘negative polarity’. ‘Normally, ‘negative polarity items’ are grammatical only if they appear within the scope of a negative expression (for example, not) or of some other expressions with similar properties.
- The pivot must originate at the right-hand edge of the conjunct that contains it (unless a ‘heavy’ pivot is shifted first to the right before RNR takes place). Movement is not subject to such a constraint.
One copy of the pivot or two?
Evidence for multi-dominance
As evidence favouring multi-dominance approaches, Citko offers the following evidence that only one copy of the pivot is present in RNR:
- Sometimes, the form of material inside the pivot is explicable only if material inside both conjuncts dominates the pivot jointly, rather than dominating 2 separate copies of the pivot in different ways. Examples include plural agreement on the pivot when each conjunct contains a different singular noun. (Though Citko points out that sometimes the pivot agrees only with material within the conjunct that is closer to the pivot.)
- Sometimes, the pivot contains a relational term (such as same or different) and the intended meaning is available only if that relational term applies to both conjuncts jointly, rather than separately.
- Ellipsis generally deletes the 1st copy of repeated material, yet RNR deletes the 2nd copy. If RNR involves ellipsis, it is not clear why it would differ in this way from other forms of ellipsis.
Evidence for ellipsis
As evidence favouring ellipsis approaches, Citko offers the following evidence that 2 copies of the pivot are present in RNR:
- RNR sometimes permits a mismatch between a feature of the pivot in its original position and its feature in the final position. For example, it sometimes permits a verb within the pivot to be in one tense in one position and a different tense in the other position. RNR permits such mismatches in the same cases as a form of ellipsis known as Verb Phrase ellipsis (VP ellipsis) This suggests that RNR uses the same mechanism as VP ellipsis.
- Like ellipsis constructions, RNR allows ‘vehicle change’ effects.
(a) One vehicle change is when the pivot is a noun, but its (deleted or silent) copy inside the 1st conjunct must be, for example, a pronoun because another constraint blocks that copy from being a noun.
(b) Another vehicle change is when the pivot contains an item (such as any) that has ‘negative polarity’ but the corresponding item in the other conjunct (such as some) has positive polarity.
- RNR permits both ‘sloppy identity’ and ‘strict identity’ readings in cases when ellipsis constructions also permit it. Example 6 is one instance when RNR permits both ‘sloppy identity’ and ‘strict identity’ readings.
(6) John likes ____, but Bill hates ____, his father.
Example 6 can mean either:
(a) John likes his father, but Bill hates John’s father [strict reading];
(b) John likes John’s own father, but Bill hates Bill’s own father [‘sloppy’ reading]; or
(c) John likes some 3rd party’s father, but Bill hates that 3rd party’s father [3rd party reading].
Citko presents the pros and cons of each approach, without necessarily arguing in her chapter for any particular one. She points out that in her other work she has argued for a multi-dominance approach.
Citko observes that there may not be a single approach that works best for all RNR constructions. Instead, different approaches may work best for different types of RNR construction.
Right node raising is an interesting construction and many books and papers have been written on it. There still seems to be little consensus on how best to analyse it. Citko’s chapter is a useful overview of work done on this construction (within transformational generative grammar) up to 2013.
I started this post with a point about editing and writing. Right node raising creates structures that are grammatical. But, in my view, many readers find it difficult to process those structures. In some cases, readers cannot understand those structures at all, or can understand them only with great effort. That is because these structures force readers to wait until the end of the sentence before they can find the object of a verb that appeared early in the sentence.
I recommend that writers avoid those structures unless they are confident that their intended readers will understand them easily and quickly.