Having a syntactic choice is not always better: the effects of syntactic flexibility on Korean production

Abstract

Sentence production requires speakers to select lexical items and a structural frame necessary to communicate a message. The present study examines how Korean speakers choose between alternative syntactic structures. Following the methodology of Ferreira, we conducted a series of production studies investigating the effects of syntactic flexibility in Korean numeral quantifier constructions and active/passive constructions. Two models of sentence production make different predictions regarding the effects of syntactic flexibility. The competitive model predicts that syntactic flexibility should cause production difficulties (e.g. longer production latencies and more errors) because alternative structures compete for selection, restricting one another's availability. In contrast, according to the incremental model, syntactic flexibility should facilitate production (e.g. shorter production latencies and fewer errors) because it allows more accessible lexical items to be accommodated sooner. Ferreira's results support the incremental model in English. Our results, however, show that Korean speakers produced utterances more slowly in the flexible condition, which provides support for the competitive model. We suggest that the different findings in English and Korean are related to how they assign grammatical functions.

Keywords:

• Korean
• sentence production
• syntactic flexibility

Acknowledgements

We thank Jeung-Ryeul Cho, Sung Hak Jung, Taehoon Kim and Jaehong Ko for their help with data collection for Experiments 2 and 3.

Notes

1. Ditransitive sentences in Korean are not suitable for our purposes because, although they allow two different word orders similar to English (give the book to the boy vs. give the boy the book), these two word orders in Korean have identical case marking (give book-ACC boy-DAT vs. give boy-DAT book-ACC). As a result, in Korean where a syntactic structure is contingent on case marking, it is difficult to manipulate syntactic flexibility with ditransitive constructions in Korean (i.e. hard to force Korean speakers to produce a certain ditransitive structure).

2. In Korean, the realisation of the plural morpheme -tul is not obligatory, unlike English plural marking -s.

3. In English, the active/passive and the ditransitive alternation (the two structures investigated by Ferreira in his seminal 1996 paper) are commonly regarded as conveying the same core meaning, although they are not always semantically fully equivalent. For example, Every boy kissed at least one girl does not entail that At least one girl was kissed by every boy, and John taught the students Spanish implies that students learned Spanish, whereas John taught Spanish to the students does not. Similarly, Korean pre-nominal and post-nominal structures are not always fully equivalent; unlike pre-nominal structures, post-nominal structures can convey a part-whole relationship as seen in (1) (see Han, 1999; Y.-H. Kim, 1983; Shin, 2007).

(1) Twu tay-uy olaytoyn cha-ka kocangnassta.

two CL-GEN old car-NOM broke down

‘Two old cars broke down’.

(2) Olaytoyn cha twu tay-ka kocangnassta.

old car two CL-NOM broke down

‘Two old cars broke down’.

‘Two of the old cars broke down’.

4. Other factors such as list and the order of presentation did not have any effect on the error rates or the production latencies, as shown by model comparison.

5. Overall, the onset latencies that we observed for Korean in our experiments are longer than those found by Ferreira (1996) for English. While this could be due to Korean speakers adopting a more deliberate attitude towards the task (considering their options more explicitly than Ferreira's English speakers), it is not clear what would be driving this kind of cross-linguistic difference in level of deliberation. In fact, this difference may well be due to the fact that the average syllable length of the initial nouns in our studies is longer than in Ferreira's. It is well known that utterance latencies correlate with the length of the initial noun phrases (e.g. Levelt & Maassen, 1981). In Ferreira (1996), the average syllable number of the initial nouns in canonical sentence structures ranges from 1 (Experiment 1) to 1.95 (Experiments 2 and 3), whereas in our study, the average syllable number (including the subject case particle) ranges from 3.3 (Experiments 2 and 3) to 4 (Experiment 1).

6. One may wonder whether the faster production latencies in the non-flexible conditions could simply be due to speakers not needing to retrieve the case-markers, since the case-markers were already given/present in the non-flexible conditions. However, this explanation is not likely. If the overt presence of a case-marker were the reason for the faster production latencies in non-flexible conditions, English speakers should also produce prepositional structures faster in the non-flexible condition with the overt preposition than in the flexible condition without the preposition, by the same reasoning. But Ferreira (1996) found that English speakers produced prepositional structures faster in the flexible condition despite the absence of the preposition (i.e. not needing to retrieve the preposition did not speed things up). Although English speakers did not need to choose between ‘to’ and another preposition, they still needed to choose whether to use the preposition ‘to’ or not. Korean speakers needed to choose between two case-markers, but they did not need to choose whether to use a case-marker or not – since case marking was obligatory in the experiments (the experimental stimuli without case-markers sound unnatural). That is, both English and Korean speakers were given two choices in the flexible condition. Thus, Korean speakers’ faster production in the non-flexible conditions cannot be attributed to the simple presence of case-markers in the non-flexible condition or the availability of two case-markers in the flexible condition.

7. It is important to note that Myachykov et al. (2013)'s results go against Ferreira's original (1996) findings because Myachykov et al. argue that syntactic flexibility is costly in English. However, due to differences in the methodologies in these two studies, one should take care when comparing their results. Because our method is closer to Ferreira's, we focus on comparing our findings with his. Furthermore, as will become clear later in this section, we do not claim that incremental processing plays no role in Korean, or that competitive processing plays no role in English. Instead, we suggest that a particular language can employ both incremental and competitive mechanisms and that the grammar of a language is a factor that modulates the extent to which the two mechanisms influence sentence production.

8. Korean speakers may omit case-markers in colloquial speech (i.e. case ellipsis, e.g. T. Kim, 2008; Sohn, 1999). This may seem to challenge our motivation for the competitive production for Korean. Crucially, however, whether case is phonologically realised or not does not change the degree of flexibility in function assignment. For example, despite case ellipsis, Korean speakers could nonetheless assign the initial noun to the subject or the object as seen below (in contrast to English speakers).

(a) Ne ku yenghwa boassni?

you that movie watched

(b) Ku yenghwa ne boassni?

that movie you watched

‘Did you watch that movie?’

Thus, in our opinion, the phenomenon of case ellipsis does not pose a challenge to our account of Korean production. We assume that (1) case-markers are represented at constituent assembly stage as affixes are in English (e.g. Ferreira & Slevc, 2007), and that (2) Korean speakers may omit case-markers for ‘economic reason’ at the level of phonological encoding. Similarly, we also suggest that whether case is abstract or morphological per se does not impact how a speaker plans a particular sentence (e.g. ‘Peter saw John’ vs. ‘Peter saw him’). We suggest that what matters might be the typological properties of the language as a whole, in particular, how flexible a language is in the mapping of word order and grammatical function.

9. Another way of looking at these issues is through the lens of Uniform Information Density (UID). According to UID, the choices speakers have to make are at least partially determined by information density: if one way to convey a message leads to more uniform information density than another way to convey the same message, the variant with a more uniform distribution of information should be preferred (Jaeger, 2006; Levy & Jaeger, 2007). Indeed, speakers’ productions have been shown to be consistent with a UID strategy (e.g. rates of complementiser that use vs. omission of that, Jaeger, 2006; Levy & Jaeger, 2007). English speakers are also likely to use full forms instead of contractions (e.g. you are vs. you're) at points of high information thereby extending the time during which the high information element is uttered (Frank & Jaeger, 2008). However, the questions of (1) whether the syntactic variants described in this paper vary systematically in terms of their information density and (2) which elements in the sentence are more versus less informative, are not straightforward. Thus, in our opinion, it is not clear how to apply UID to the findings of the current study, although it provides an interesting perspective for a future production study.