Andrew Radford - Linguistics An Introduction [Second Edition]

(481)

John helped the students who lost out

This sentence again has two interpretations, paraphrases of which are given below: (482) a.

John supported the students who lost out

b.

John helped out the students who lost

The first interpretation (482a), in which the preposition out is associated with the verb lose, is clearly preferred by most listeners, but the second interpretation (482b),

in which out is associated with help, is also perfectly grammatical. Ambiguities such as those in (481) are less likely to occur in spoken language comprehension, as different stress patterns lead hearers to prefer certain interpretations; for example, if the speaker introduces a noticeable pause after lost in (481), then the particle out is likely to be understood as linked to the main verb helped. Obviously, such cues are not available in written language comprehension. Notice also that the separation of a particle from the verb is actually quite common in English and does not normally produce any processing difficulties. Compare, for example, (481) and (483): (483)

Betty put the big Persian cat out, before she left the house.

372

senten ces

We can roughly indicate the structural ambiguity of (481) by the different bracketings in (484a, b):

(484) a.

John helped [the students who lost] out

b.

John helped [the students who lost out]

But why do listeners prefer the bracketing in (484b) to that in (484a)? Given that both structures are equally grammatical, we have to look beyond mere structural descriptions to find an answer to this question. Recall that syntactic theory accounts for the existence of certain types of structural ambiguity by deriving them from different structural representations, as, for example, in (480a, b). But syntactic analysis itself cannot explain how it is that people resolve such ambiguities in the way they do on specific occasions, nor why they often prefer one structure over another in a manner independent of context, as, for example, with (481). In short, we need to find out what additional strategies or principles listeners employ when they parse sentences.

The fact that listeners prefer interpretation (484b) over (484a) is indicative of a fairly general property of sentence processing: the idea is that as the parser builds a structure, whenever there is a choice between a local and a distant attachment possibility, as in the case of out in (481), it favours the more local one. Put differently, listeners prefer to construe any given word as part of the constituent being processed at that time, rather than as part of a different constituent. With respect to (481), this means that the preposition out is construed as a constituent of the nearest VP, which is the VP headed by the verb lose, rather than of the VP headed by help, which is further away from out. The structure in (485) illustrates this:

(485)

CP

C

TP

ϕ

DP

T'

John

T

VP

ϕ

V

DP

helped

D

NP

the

N

CP

students

D

C'

who

C

TP

ϕ

D

T'

who

T

V

ϕ

lost out?

Sentence processing

373

The structure in (485) is based on the assumption that syntactic parsing is like reading or speaking in that it proceeds from left to right and that it is done on-line, i.e. whenever the parser comes across a new word in its left-to-right journey through the sentence, it has to incorporate the word into the tree which is available at that point. In (485), we have got to the point where the parser encounters out, and the options for the attachment of this item are indicated in (486a, b) – we assume that out adjoins (see section 23) either to the verb lost (486a) or to the verb phrase helped the students who lost (486b):

(486)

a.

V

b.

VP

V

P

VP

lost

out

V

DP

P

helped

the students who lost

out

The listener’s grammar provides the information that out can adjoin to either lost or helped. However, attaching out to the verb lost as in (486a) is a local attachment and is therefore preferred. The alternative of attaching out to the verb phrase helped the students who lost as in (486b) involves the parser in looking back earlier in the sentence and reconsidering the structure of the higher VP (a procedure known as backtracking). In general, the parser will avoid backtracking and rearrangement of constituents as much as possible. With (486a), no such backtracking is required, and this parse is consistent with the local nature of human parsing.

Another type of grammatical sentence which does not involve structural ambi-

guity but which yields considerable processing difficulties is one which includes centre-embedding. Compare the sentences in (487):

(487) a.

The pen the author the editor liked used was new

b.

The pen which the author whom the editor liked used was new

c.

The editor liked the author who used the pen which was new

These sentences are paraphrases of each other, with no significant meaning

differences, and none of them violates any grammatical requirements of

English. But on several processing measures (e.g. RTs, accuracy of paraphrasing, etc.), (487a) proves more difficult to parse than (487b), and (487b) proves more difficult than (487c). Notice that this holds despite the fact that in terms of the number of words involved, (487a) is actually shorter than both (487b) and (487c).

How do we explain these processing differences?

The main factor distinguishing (487a) from (487c) is that, in the latter, parsing can proceed locally, whereas this is not possible in the former. In (487c), the three basic clauses (the editor liked the author, the author used the pen and the pen was new), separated by commas, can be straightforwardly parsed from left to right.

Thus, when the parse for the first clause is closed, it can be cleared from short-term memory, as can that for the second clause. In (487a), however, the three basic clauses must all be kept in short-term memory until the end of the sentence is

374

senten ces

reached; only at this point can the listener attach the appropriate verb to each of the three sets of arguments to form the three basic clauses. A very fundamental finding in research on short-term memory is that its capacity is severely limited (see how many digits you can remember in the sequence in which they are presented to

you), so we might plausibly suppose that one difficulty with (487a) is that its processing overloads short-term memory – importantly, this is not a linguistic difficulty. Another aspect of (487a) that prevents local parsing decisions being taken is that the sequence the pen the author the editor does not contain any cues as to the grammatical function (subject or complement) fulfilled by these DPs in the various clauses of the sentence. Compare this with (487b). In this case, the relative pronouns which and whom provide cues which allow the parser to assign grammatical functions to these elements. As the parser can make some decisions early on in (487b), which cannot be made in (487a), parsing (487b) is more local than parsing (487a), and hence again is less burdensome for short-term memory

(exercise 4).

We consider finally a phenomenon touched on briefly in our introduction

(p. 10), that of garden-path sentences or syntactic illusions, as this also requires a processing explanation. In syntactic illusions, a certain decision about interpreting a sentence which is locally tenable leads to the (incorrect) conclusion that a grammatical structure is ungrammatical. Consider, for example, (488a) which is a perfectly grammatical sentence as can be seen from inserting the relative pronoun which and the auxiliary verb was in the appropriate positions (488b):