This paper discusses the effect of processing complexity on the English comparative alternation. The reported experiments show a processing advantage of the synthetic comparative in perception, but a preference of the analytic comparative in sentence production if the base adjective is cognitively complex. These results imply that perceptual complexity and complexity in production have diverging effects on the English comparative alternation. More generally, the paper calls for a fine-grained look at the role of processing complexity in areas of morphosyntactic variation.
Most English comparatives are formed using
either a synthetic form (e.g. easier) or an analytic
form (e.g. more important). While most
adjectives clearly prefer either the synthetic or the
analytic comparative, there is a considerable
number of adjectives which frequently take both forms,
e.g. more friendly vs. friendlier. The decision
for either form is influenced by several
phonological, morphological, syntactic and semantic factors.
For example, the probability of analytic
comparatives increases with the number of morphemes in
the adjective base. It is also higher if the
comparative is in predicative than in attributive
position, and it decreases with an increasing
comparative/positive ratio
Mondorf (2009) argues that these factors are all part of a more general, audience-oriented compensatory mechanism called more-support: if the cognitive complexity of the adjectival base or its environment increases, speakers prefer the analytic comparatives, because they have a processing advantage over the corresponding synthetic form. For instance, an adjective that is morphologically complex is assumed to be also cognitively more complex than a simplex adjectives, and in order to compensate for this increased cognitive complexity, speakers may prefer the analytic comparative over the synthetic alternative.
Yet, there is only little psycholinguistic research that investigated this assumed processing advantage of analytic forms. A notable exception is Boyd (2007, ch. 2) who conducted a self-paced reading experiment to investigate processing differences between synthetic and analytic comparatives. Indeed, he reports shorter reaction times for the sentences containing analytic comparatives, but due to the experimental design, this evidence is only indirect and allows for alternative interpretations. As yet, then, there is only limited empirical evidence for the assumption that analytic comparatives are easier to process than synthetic comparatives. In addition, as pointed out by Mondorf (2014, 201), it is still an unresolved issue whether more-support is a response to increased processing loads in production or in perception.
This paper addresses these two issues. First, it presents the results from a perception experiment which tested whether analytic comparatives are indeed easier to process for listeners. Contrary to this hypothesis, the reaction times show that analytic comparatives have a processing disadvantage in perception. Then, a production experiment is discussed which elicited spoken sentences containing a comparative construction. The analysis reveals that the processing complexity is a significant predictor of the comparative alternation: with increasing complexity of the base adjective, the probability of analytic comparatives increases. Thus, the paper argues that speakers and listeners process the English comparative variants differently, and that it is the speaker who benefits from a compensatory use of more comparatives.
Comparative variation in perception
31 native speakers of Canadian English
participated in an auditory decision task in which they
had to decide whether the acoustic stimuli was an
existing English form. The set of stimuli contained
the analytic and synthetic comparative form for 60
adjective types with at least 5 attestations for both
forms in the Corpus of Contemporary American
English
Alongside the 2 × 60 = 120 synthetic and analytic comparatives, the set of stimuli also included 360 distractors. Some of the distractors combined more with non-existing words, others combined the adjective bases with the illegal suffix -ic. In addition, the set of distractor items contained nonexisting words ending in -er as well as existing words and complex words. Examples of the test stimuli are given in (1a), and distractor examples are given in (1b). (1) a. colder, happier, yellower
more cold, more wealthy, more yellow b. ∗coldic, more ∗gorsty, ∗rilker
on wire, chasting 2.2
The density estimate suggests that reaction
times are, in general, higher for analytic
comparatives than for synthetic comparatives. This
visual interpretation is supported by a linear
mixedeffects regression model with reaction times as
the dependent variable
After removal of insignificant predictors, the final model reports significant interactions between stimulus Class and Preceding RT, PLD20, Number of phonemes, Synthetic frequency, and Analytic frequency. Figure 2 displays the partial effects for these interactions. The vertical axis shows the transformed reaction times; higher values correspond to longer reaction times.
In agreement with figure 1, the partial effects reveal significantly lower estimates for the synthetic stimuli (solid lines) than for the analytic stimuli (dashed lines). This is true even in the most adverse conditions (e.g. in cases in which the synthetic comparative of a comparative is attested only very rarely in a linguistic corpus, left edge of lower right panel in figure 2).
3.1 41 native speakers of Canadian English participated individually in a spoken sentence completion task. The task used the same set of 60 adjectives as in the perception experiment above, but none of the participants in the production experiment had also participated in the previous task. Participants were first shown a context sentence containing the adjective in the positive. After a key press, an incomplete target sentence containing a blank and one or more target words appeared also on the screen. The participants were instructed to use the target words to fill the blank in the sentence. If necessary, they could also use additional words to complete the sentence. The sentences were constructed in such a way that a comparative construction was the most likely target for completion, but participants were not explicitly instructed to use comparatives. The structure of the incomplete sentences was the same in all trials. The subject was a simple noun phrase, followed by a copula verb. The blank to be filled followed in predicative position. This design ensured that the context-dependent factors reported in the literature such as the increased probability of analytic comparatives in predicative position were held constant for all adjectives. Example (3) shows the experimental trial for the target adjective wealthy. (2) The duke is wealthy.
Yet, the king is WEALTHY .
The experiment also contained 105 distractor trials that had a similar structure, but which did not contain adjectives as the target words. 3.2
In order to be able to investigate the effect of the processing complexity of the base adjective on the preferred comparative variant, the same 41 speakers first participated in a visual lexical decision task that gathered reaction times for the 60 target adjectives, as well as 150 other existing and nonexisting distractor items. The participants were not informed about the purpose of this task, and there were at least 14 days for each participant between the lexical decision task and the production experiment. The reaction times obtained in this task were pooled for each adjective, and the median was calculated. 3.3
For most of the adjectives, the completion task was successful in obtaining comparative responses from the 41 speakers. However, two participants produced hardly any comparative in the task, and were therefore excluded from the data set. 6 out of the 60 adjectives were excluded because the responses contained almost exclusively synthetic or analytic comparatives, or because the context sentence did not elicit a considerable number of comparative responses. 747 out of the remaining 39 × 54 = 2106 responses contained a synthetic comparative (35 %), 843 contained an analytic comparative (40 %). The remaining 516 responses (25 %) did not contain a comparative construction, and were discarded. There was notable variation between the two variants both across and within items, which indicates that English comparative variation is indeed a highly non-deterministic field that is apparently affected by both speakerdependent and adjective-dependent factors.
Logistic general additive mixed-effects models
Two models were fitted: a null model which contained only a random effect for speaker, and
Discussion and conclusion a model with an additional smooth term for the effect of the median RTs. If processing complexity has a notable effect on speaker responses, the smooth term should turn out to be statistically significant, and the predictive accuracy of the model should improve by the addition of the term. As table 1 shows, this is indeed the case. While the null model has a total predictive accuracy of about 69 %, the addition of the smooth term for median RTs increases the accuracy by 5.6 %. There is a larger increase of predictive accuracy for analytic responses than for synthetic responses (7.1 % vs. 3.9 %).
This work was supported by the Deutsche Forschungsgemeinschaft (grant KU 2896/1-1). I wish to thank Ben Tucker (University of Alberta, Edmonton) for making available to me the facilities of the Alberta Phonetics Laboratory for the experiments reported in this paper.