STRATEGIES IN THE COMPREHENSION OF RELATIVE CLAUSES
Language and Speech, 1975, 18, 3, 204-212
Historical note: to my knowledge this is the first experiment using the Accessibility Hierarchy in Second Language Acquisition research, preceding the well-known paper by Comrie and Keenan 1977 (which cites it - better time-travelling than Dr Who) - I'd got the idea from a London talk by Keenan.
experiment is described to test the hypothesis that errors in the
comprehension of relative clauses in English are caused by perceptual
strategies resorted to when the normal capacity of the processing channel is
exceeded. Native children, foreign adults, and native adults were asked to
show Subject and Object relations in sentences that were read to them. The
same type of error pattern was found in native children and foreign adults
with single embedded clauses, in native adults with double embedded clauses,
and, to a lesser extent, in native adults with single embedded clauses. The
dominating strategies appeared to be that the first Noun Phrase was the
Subject and the Noun Phrase following the verb the Object. These results
suggest that the difficulty of relative clauses is indeed due to the load they
put on the processing system and that an overload is reached with a single
embedding for children and foreign adults, and with a double embedding for
most native adults.
has often been claimed that plurally embedded relative clauses in English such
as "The boy the girl the dog bit saw kicked the ball" are
grammatical but difficult to comprehend. Several explanations have been
offered for this. Yngve suggested that their regressive structure tends to
place an excessive load on the temporary memory used for speech processing (Yngve,
1960). Miller and Chomsky agreed that memory limitations are involved but
disagreed about the importance of regressive structure (Miller and Chomsky,
1963). Miller and Isard suggested that the difficulty is caused by
interruption of a sub-routine for interpreting relative clauses; when
confronted with a plurally self-embedded sentence, the hearer starts using
this subroutine; halfway through it he has to interrupt it to start again;
and so on for each embedding (Miller and Isard, 1964). Blumenthal claimed that
self-embedded sentences are perceived as ungrammatical and that the hearer
simply searches for the nearest grammatical equivalent; thus "The manager
whom the designer whom the typist whom the receptionist encourages interests
consults phoned the producer" would be perceived as "The manager
that the designer encourages, that the typist interests, and that the
receptionist consults, phoned the producer" (Blumenthal 1966). Finally
Bever has argued that the perceptual strategy that any Noun Verb Noun sequence
tends to be interpreted as actor-action-object and the general perceptual
principle that "A stimulus may not be perceived simultaneously as having
two positions on the same classificatory dimension" both contribute to
the difficulty: thus in a sentence such as "The dog the cat the fox was
chasing was scratching was yelping," "The dog " is
simultaneously Subject of " was yelping " and Object of "was
scratching," "the cat" is at the same time Subject of "was
scratching" and Object of "was chasing" (Bever, 1970). This
would suggest that the presence of relative pronouns such as "that"
would make the sentences easier to comprehend by revealing the structure of
the sentence rather than more difficult, as a straightforward explanation in
terms of memory for words would suggest; this has been confirmed by some
experiments (Fodor and Garret, 1967; Hakes and Foss, 1970) but not by others
(Foss and Lynch, 1969). In addition it has been shown that the difficulty of
the sentence is increased by the presence of a verb that can take a
complement, such as "know," rather than one that can only take an
object, such as " meet" (Fodor, Garrett, and Bever, 1968).
The variety of these explanations is matched by the variety of the tasks employed in the related experiments. The commonest task is paraphrase: the subjects are asked "to say in their own words what it meant" (Hakes and Foss, 1970) or "to break it into its component clauses and to write each clause as a simple sentence" (Stolz, 1967); then their errors are calculated by one method or another. A second task is recall: the subjects repeat the sentences aloud and the number of trials necessary for accurate repetition is counted (Miller and Isard, 1964); accuracy is measured by the number of words in the correct order. Further tasks have measured the response latency before the subject starts to paraphrase (Fodor and Garret, 1967) or before he identifies a particular phoneme (Hakes and Foss, 1970). These tasks all have indirect, though plausible, links with normal speech comprehension. Paraphrase involves production as well as perception; the ability to recode a sentence in another form is not necessarily the same as the ability to comprehend it. Furthermore the instructions in a paraphrase task may be open to misconception; it does not appear surprising that Blumenthal's subjects treated relative clauses as ungrammatical when they were instructed to "rewrite each sentence to render it more acceptable or understandable" (Blumenthal, 1966). Recall tasks also may have little directly to say about speech comprehension; they may show something about the effectiveness of rehearsal or about the task of repetition but not necessarily anything about normal speech comprehension. Phoneme monitoring similarly involves problems of interpretation particularly since it has been argued that phonemes are recognized subsequently to syllables (Savin and Bever, 1970). The variety of evidence and explanations for the comprehension of relative clauses seems to suggest that a variety of mental processes are being tapped that are only indirectly linked to normal comprehension. This is supported by the fact that all the experiments described have dealt with the comprehension of plurally embedded relative clauses rather than single embedded clauses; it is admitted that the plurally embedded clause is rare in normal speech and Stolz has indeed argued that it represents a new grammatical structure that has to be learnt (Stolz, 1967).
research to be described attempted to investigate how people tackled a task
that asked them, more or less directly, how they comprehended relative
clauses, both single and double embedded. Earlier work had suggested that
foreign adults and native children made the same type of errors in the
repetition and comprehension of relative clauses (Cook, 1973). The hypothesis
that suggested itself was that errors in the comprehension of relative clauses
are caused by perceptual strategies employed when the normal capacity of the
processing system is exceeded. Evidence
in favour of this would be: (i) while the overall level of errors would be
different for native adults, native children, and foreign adults on sentences
with single embedded relative clauses, they would show the same general
pattern; (ii) though the level of difficulty would be higher, the same pattern
would manifest itself for native adults with double embedded sentences.
types of subject were used: (i) native English-speaking children from primary
schools in East London aged from 4 to 9 years, ten at each year (n
(ii) four groups of foreign students from mixed language backgrounds studying
English as a Foreign Language at Ealing Technical College (w = 52); (iii)
eight groups of native English-speaking adults, all students in higher
= 111). ENDNOTE
The subjects were required to identify the syntactic relationships of sentences including relative clauses. There were two variations of the task:
were tested in groups and were given an answer sheet with the instructions
"After each sentence has been read to you, please show who is biting or
pushing and who is being bitten or pushed by means of a circle and an arrow as
in the examples." This was followed by two sentences, "The dog
pushes the horse " and "The man is bitten by the horse "; each
of these sentences had the words "cat dog man horse " written below
them. In the first case "dog" was ringed and the arrow went to
"horse"; in the second case " horse" was ringed and the
arrow went to "man". Then on the answer sheet followed numbers for
each sentence with columns of "cat dog man horse"; for the single
embedded sentences "man" was omitted. The instructions were also
given orally and illustrated on a blackboard. Then each sentence was read
twice, sufficient time being given for everyone to mark the answer on their
Children were tested individually. A set of toys was placed in front of them
consisting of two cats, two dogs, and a horse. They were told what the animals
were; then asked to "show me some things that they do."
same practice sentences as for the adults were used except that, for some
children, additional sentences with the same structure were given till they
understood the task. Then the test sentences were each read twice and the
result noted on a score sheet. If the child's action was unclear, he was asked
specifically "Show me who is biting/ pushing." Some older children
preferred to reply orally rather than to demonstrate with the toys.
1 gives the sentences that were used, one set having single embedded relative
clauses, the other double embedded clauses differing from the first set by
having "sees" and "the man" in the appropriate
Table 1 The test sentences.
cat that likes the dog bites the horse.
dog that pushes the cat likes the horse.
cat what likes the dog pushes the horse.
dog what bites the cat likes the horse.
dog that the cat likes pushes the horse.
cat that the dog bites likes the horse.
cat what the dog likes bites the horse.
dog what the cat pushes likes the horse.
dog the cat likes pushes the horse.
cat the dog bites likes the horse.
cat that likes the dog bites the horse.
dog the cat likes pushes the horse.
(1) The cat that likes the dog that sees the man bites the horse.
(2) The dog that pushes the cat that sees the man likes the horse.
(3) The cat what likes the dog what sees the man pushes the horse.
(4) The dog what bites the cat what sees the man likes the horse.
(5) The dog that the cat that the man sees likes pushes the horse.
(6) The cat that the dog that the man sees bites likes the horse.
(7) The cat what the dog what the man sees likes bites the horse.
(8) The dog what the cat what the man sees pushes likes the horse.
dog the cat the man sees likes pushes the horse.
cat the dog the man sees bites likes the horse.
(11) The cat that likes the dog that sees the man bites the horse.
dog the cat the man sees likes pushes the horse.
The sentences were all marked for intonation using the system described in Cook (1968); the intonation pattern used consisted of a high fall on "horse" throughout preceded by a low rise on the last word of the last embedded clause. Sentences 1 to 4 had the Subject of the main clause acting as Subject of the relative clause; Sentences 5 to 10 had the Subject acting as Object of the relative clause. Sentences 1, 2, the relative pronoun "that"; sentences 4, 5, 7, 9 had the relative pronoun "what", believed to be grammatical in the variety of English spoken by the children; sentences 9, 10 had zero relative pronoun. Sentences 1, 3, 5, 7, 9 had "bites" or "pushes" in the main clause; sentences 2, 4, 6, 8, 10, in the relative clause. In addition the following safeguards were built in: (a) sentences 11 and 12 were identical to sentences 1 and 9; (b) two alternative versions of each sentence were used in which the nouns "cat" and "dog" were interchanged; (c) two different randomized orders of presentation were used. Taking all these into account, the test had eight different forms.
The children were given the four sets of single embedded sentences in rotation. Each group of foreign adults were given a different set of single embedded sentences. Four groups of native adults heard the different sets of single embedded sentences (w = 57); four groups heard the double embedded sentences (n = 54).
The answer sheets were analysed in terms of errors made in choosing the Subject and Object of the verbs "bite" and "push." The total errors for Subject (S) and Object (O) for all groups are shown in Table 2.
differences between the sentences for the native children, the foreign adults,
and the native adults with double embedding reach statistical significance (S
errors for the children, p <0.05; S
errors for foreign adults and native adults with double embedding, and O
errors for all three, p <0.01). When
the results were analysed for order effects, significant differences emerged
for sentences 7 and 8, those
in which "what" acted as Object (S errors for 7, p <0.01; S
errors for 8, p <0.05). The
results for these two sentences are not included in later tables, though their
pattern is similar to that for the other sentences.
Table 3 Subject and Object errors grouped by sentence pairs
Table 3 shows
the pattern more clearly by combining pairs of sentences that differ only in
the relative pronouns they contain. For the adult groups most S errors
occurred on 6 and 10, those
in which the Subject of the main clause is the Object of the relative clause
and in which the relative clause contains "bites" or
"pushes": the next most difficult pair were 5 and 9, those
in which the Subject of the main clause is the Object of the relative clause
and in which the main clause contains "bites" or "pushes".
The children found these two pairs equally difficult but substantially more
difficult than the other sentences. For all groups except the native adults
with a single embedding, the easiest pair were 2 and 4, those
in which the Subject of the main clause is also Subject of the relative clause
and in which "bites" or "pushes" is in the relative
clause. Most O errors occurred on 6 and 8; the
next most difficult pair were 6 and 8, particularly
for children. The differences between all groups are statistically significant
(p <0.01). There
were no statistically significant differences between sentences having "
that" (5 and 6) and
zero relative (9 and 10).
Table 4 Subject and Object errors grouped by relative pronoun
us first take the S errors on single embeddings. Bever's perceptual strategy
that NVN = actor-action-object would predict that 1, 3, 5, 9 would be easier
than 2, '4, 6,
10; while this is true for all groups but the children, it does not go very far
towards explaining the pattern in Table 3. An alternative strategy for S is that
the first Noun Phrase in the sentence is the Subject of both main and relative
clauses; this predicts that 6 and 10 would be difficult, as they are. The
residue of unaccounted errors are those on 5 and 9; these are precisely the
sentences that would be affected by Bever's general perceptual principle that a
stimulus cannot simultaneously have two functions. Taking these together we have
then accounted for the order of difficulty found, namely 6/10, 5/9, 1/3, 2/4,
with the exception of (a) the ease of 1, 3, 5, 9 for the adults with a single
embedding and (b) the equal difficulty of 5, 9, 6, 10 for the children. Turning
now to O errors on single embeddings, Bever's strategy predicts that 6 and 10
would be the most difficult, which is true, and does not distinguish between the
rest. To account for Table 3 an additional strategy is needed, namely that the
last Noun Phrase in the sentence is the Object: this predicts that 2, 4, 6, 10
would be more difficult than 1, 3, 5, 9, thus explaining the comparative
difficulty of 2 and 4. However, if Bever's perceptual principle is interpreted
as affecting the location of Objects, it would wrongly predict that 5 and 7
would be more difficult. So, for both S and O errors, dominating strategies can
be found that are similar to those proposed by Bever; nevertheless, additional
strategies have to be postulated to account fully for the results.
regard to the double embedded sentences, again the same strategies seem to be
employed. S errors follow the same pattern as for the children and foreign
adults and reveal more effects of Bever's perceptual principle than for native
adults with single embeddings. O errors are again similar, showing the influence
of Bever's proposed strategy but no evidence of the last Noun Phrase = Object
strategy. The absence of a relative pronoun had surprisingly little effect.
There was, however, the problem with "what" as Object mentioned above,
the sentences with this having been excluded from all tables except Table 2. The
order effect found here seems to show that the assumption that "what"
as Object was grammatical was incorrect; if it occurred early in the test
hearers did not know what to make of it; if they heard it later in the test they
assumed that it fitted the pattern of the other sentences: thus in a sense
"what" as Object had to be learnt during the test.
experiment confirms in general that hearers make use of definite perceptual
strategies in the comprehension of relative clauses and that the strategies
found in a type of task intended to be closer to normal speech were similar to
those that had been found in the other tasks that have been employed. The common
S strategy seems to be that the first Noun Phrase is the Subject; this is
similar to Bever's perceptual strategy, could account for Blumenthal's
paraphrase results, and would fit into the Accessibility Hierarchy of
relativization suggested by Keenan (1972). The common O strategy seems to be
that the Noun Phrase following the verb is the Object with some subjects relying
on the strategy that the last Noun Phrase in the sentence is the Object. The
same strategies occur markedly in children, in foreign adults, and in native
adults with two embeddings; they appear still to some extent in the
comprehension of single embeddings by native adults. It seems then true that
they are resorted to more and more depending on the load on the processes of
speech perception; these processes are overloaded for children and foreign
adults by a single embedding, and are overloaded for most native adults by
double embeddings. This would suggest that the ability to comprehend relative
clauses will increase gradually as the learner's capacity grows. The results
also provide a warning against exaggerating the powers of the adult native
speaker. It seems largely to have been taken for granted that native adults have
no problems in comprehending single embedded relative clauses; yet these results
show that one adult in five misinterprets certain types of relative clause.
Perhaps the same may also be true of other types of sentence that have been
assumed to be too easy to warrant
ENDNOTE I am
grateful to the following people for allowing me to carry out these tests on
their pupils and students: the head teachers and staff of Five Elms Infants
School, Godwin Infants School, and Upton Cross Infants School; the Principal
Lecturer in charge of English as a Foreign Language at Ealing Technical College;
the staff of the Education Department, North-East London Polytechnic, St. Mary's College of Education and South Midlands College of Education.
T. G. (1970). The cognitive basis for linguistic structures. In John R. Hayes
and the Development of Language (New
A. L. (1966). Observations with self-embedded sentences. Psychonomic
Cook, V. J. (1968). Active
V. J. (1973).
The comparison of language development in native children and foreign adults. IRAL, 11, 13.
J. A. and Garrett,
Some syntactic determinants of sentential complexity. Percept.
J. A., Garrett, M. and Bever,
T. G. (1968).
Some syntactic determinants of sentential complexity II:
Verb structure. Percept.
Psychophys.. 3, 453.
Foss, D. J. and Lynch,
R. H. (1969). Decision processes during sentence comprehension: Effects of surface
structure on decision times. Percept.
Psychophys., 5, 145.
D. T. and Foss, D. J. (1970). Decision processes during sentence comprehension: Effects of surface
structure reconsidered. Percept.
Psychophys., 8, 413.
E. L. (1972).
On Semantically Based Grammar. Ling.
Inq., 3, 413.
G. A. and Chomsky,
N. (1963). Finitary models of language users. In R. D.
Luce, R. Bush
and E. Galanter (eds.), Handbook
of Mathematical Psychology (New
G. A. and Isard,
S. (1964). Free recall of self-embedded English sentences. Inform.
T. G. (1970). The non-perceptual reality of the phoneme. J. verb.
verb. Behav., 9,
W. S. (1967). A study
of the ability to decode grammatically novel sentences. J. verb.
Learn, verb. Behav., 6, 867.
V. H. (1960). A model
and an hypothesis for language structure. Proc.
Amer. Philos.Soc, 104, 444.