Tense mismatches in Korean gapping and bare ko-coordination: an experimental study

1 Introduction

The sentence in (1) is an example of English gapping (the term coined by Ross 1970), which is forward ellipsis in that a gapped clause follows its antecedent clause.

The sentence in (2) is the Korean counterpart, which seems to be backward ellipsis in that a gapped clause precedes its antecedent clause. According to Ross (1970), this elliptic phenomenon is classified as an instance of gapping:^[1]

Many linguists have observed that ellipsis permits agreement mismatches with respect to voice, category, and inflections (Bošković 2004; Citko 2018; Lasnik 1999; Sag 1976). For instance, Ross (1970) observes that Russian allows phi-feature mismatches in gapping as follows:

In (3a) the missing verb has a masculine ending, whereas the antecedent verb has a feminine ending, and vice versa in (3b).

With respect to agreement mismatches in ellipsis, there have been considerable debates as to whether Korean gapping allows tense mismatches, as illustrated in (4).

In (4) the two conjuncts are mismatched in tense; the first conjunct includes the temporal adverb caknyeney ‘last year’, which triggers a past tense interpretation of the missing verb, whereas the second conjunct includes naynyeney ‘next year’, which is compatible with a future tense verb.

Regarding tense mismatches in gapping, Ahn and Cho (2006) and others (Kim 2007; Kim and Cho 2012; Park 2009) claim that conjuncts in gapping need not match in tense, which predicts (4) to be acceptable. On the other hand, Chung (2005a) and others (Choi 2019; Kim 2019) claim that gapping does not permit tense mismatches, predicting (4) to be unacceptable. Importantly, all these claims were based purely on informal judgments. More recently, however, Park and Kim (2021) claim that tense-mismatched gapping is unacceptable via a formal experiment.

Further, the issue of whether gapping permits tense mismatches has been discussed with reference to bare ko-coordination (Kim and Cho 2012):^[2]

Until now, there has been no consensus in the literature regarding the level of acceptability of the bare ko-coordination in (5) and its acceptability when aiming for a tense-mismatch interpretation. One group of researchers (Chung 2005b; Kang 2013; Lee and Tonhauser 2010) argues that bare ko-coordination permits the tense mismatch between conjuncts, which predicts (5) to be acceptable; i.e., the first conjunct depicts a past event and the second conjunct depicts a future event. The other group of researchers (Kang 1988; Park 1994; Yoon 1996; Yoon 1997) argues that only the tense-match interpretation is allowed in bare ko-coordination, which predicts (5) to be unacceptable. Let us suppose, for the time being, that the bare ko-coordination in (5) allows tense mismatches; the first conjunct contains a bare verb without an overt tense morpheme, whereas the second conjunct is suffixed with an overt tense morpheme. This difference sets apart bare ko-coordination from full coordination, as in (6), where the verb in the first conjunct is suffixed with the past tense morpheme (y)ess.

Such a mismatch between form and meaning is one of the reasons why bare ko-coordination has attracted a lot of attention.

In addition, there has been a controversy regarding the role of temporal (i.e., sequential vs. reverse) order between conjuncts on the acceptability of Korean gapping and bare ko-coordination. For example, Chung (2005b) and Lee and Tonhauser (2010) claim that bare ko-coordination with tense mismatches is allowed regardless of its temporal order. On the other hand, Park and Kim (2021) claim that the acceptability of tense-mismatched bare ko-coordination improves significantly when it is presented in the reverse order.

The rest of this paper proceeds as follows. In Section 2, we offer a brief history of tense mismatches regarding Korean gapping and bare ko-coordination. In Section 3, we report a formal experiment of tense mismatches with regard to three types of coordination: full coordination, bare ko-coordination, and gapping. In Section 4, we discuss experimental findings and propose a syntactic analysis of Korean gapping and bare ko-coordination. Finally, we conclude in Section 5.

2 Two types of tenseless coordination in Korean

English does not allow coordination in which one conjunct drops an overt tense morpheme as follows:

Given this, it is interesting to note that Korean tense inflection may be optional in the first conjunct of bare ko-coordination, although Korean verbal tense is usually encoded via an overt tense morpheme (Sohn 1995):^[3]

The first conjunct in (8), even without the past tense morpheme, may receive a past tense interpretation. That is, the two events described in each conjunct may refer to the same time reference, solely depending on the verbal tense of the second conjunct.

Taking this into consideration, Yoon (1997) and others (Kang 1988; Park 1994; Yoon 1996) argue that the tense morpheme in the second conjunct determines the temporal interpretations of both conjuncts in bare ko-coordination. Specifically, Yoon (1997) proposes that the tense morpheme takes the entire coordinate structure as its complement, rather than being suffixed only to the final verb, as schematized in (9).

Under this analysis, both conjuncts are equally under the scope of the same tense morpheme, predicting only tense-matched interpretations.

Meanwhile, Chung (2005b) and others (Kang 2013; Lee and Tonhauser 2010) observe that conjuncts in bare ko-coordination may receive temporally independent interpretations. For instance, Chung observes that the following examples are acceptable:

Specifically, he proposes that a phonetically null tense morpheme exists in the first conjunct, as schematized in (11).

As shown in (11), although the first conjunct does not have an overt tense morpheme, its tense does not rely on the tense morpheme in the second conjunct, predicting either tense-matched or tense-mismatched interpretations.

In sum, there have been two different views on the status of bare ko-coordination with respect to temporal interpretations. One view is that only the tense-matched bare ko-coordination is acceptable. The other is that either tense-matched or tense-mismatched bare ko-coordination is acceptable.

Turning to a brief survey of cross-linguistic studies of gapping, it has been observed that gapping does not tolerate tense mismatches. For example, gapping in English, French, Romanian, Turkish, Hungarian, Polish, and Japanese does not allow tense mismatches (Abeillé et al. 2014; Bartos 2001; Citko 2018; Griffiths and Lipták 2014; İnce 2009; Kato 2003, etc.):

In this light, we will settle a controversy regarding tense mismatches in Korean gapping. Among others, Chung (2005a) claims that Korean gapping does not allow tense mismatches:

In particular, Chung claims that (19) is unacceptable because the non-past tense morpheme n(ta) in the second conjunct is not compatible with the past temporal adverbial ecey ‘yesterday’ in the first conjunct.

Meanwhile, Ahn and Cho (2006) and others (Kim 2007; Kim and Cho 2012; Park 2009) claim that Korean gapping may allow tense mismatches, as shown in (20) and (21).

In particular, Park (2009) claims that Korean gapping allows tense mismatches insofar as the antecedent conjunct observes the proximity condition between an antecedent verb and its clause-mate temporal adverbial as in (22).

As for the grammaticality of (22a), Park argues that since the tense information of the temporal adverbial olhayey ‘this year’ and that of the tensed verb tut-nunta ‘takes’ in the full clause match, a tense-mismatch interpretation is available in Korean gapping. Park judged (22b) as ungrammatical under the interpretation of John is taking statistics this year due to the tense mismatch between the temporal adverbial olhayey ‘this year’ and the tensed verb tul-essta ‘took’.^[4]

Besides the TENSE (i.e., matches or mismatches) predictor, another potential predictor to be considered affecting the acceptability of Korean coordination is temporal ORDER (i.e., sequential or reverse). To be specific, Chung (2005b) and Lee and Tonhauser (2010) argue that conjuncts in bare ko-coordination can be in any temporal order without affecting acceptability. Recall that the conjuncts of (10a) are in the sequential order (i.e., past–present), while those of (10b) are in the reverse order (i.e., present–past). On the other hand, Park and Kim (2021) claim that reversely tense-mismatched bare ko-coordination is significantly more acceptable than sequentially tense-mismatched bare ko-coordination.

Since linguistic theories are ideally built on robust empirical foundations, the data merit rigorous verification. In this regard, we will launch an empirical investigation of tense mismatches in the two types of tenseless coordination (i.e., bare ko-coordination and gapping) with respect to full clause coordination in Korean.

3 Experiment

3.4 Results

The responses of 240 tokens for each of the six conditions (N = 1,440 in total) were analyzed. Figure 1 presents the mean of the z-scored ratings for the six experimental conditions. The zero represents the overall mean acceptability rating; positive z-scores indicate that conditions are rated towards being acceptable, while negative z-scores indicate that conditions are rated towards being unacceptable. The SYNTAX effect is represented by a vertical separation between the lines, and the TENSE effect is represented by the downward slope of the lines:

Figure 1:

Mean acceptability of 24 sets of experimental conditions (error bars indicate standard errors).

As for our main interest, we ran a 2 × 2 mixed-effects model for each of the two types of tenseless coordination, [Bare] and [Gapping]. First, to test whether bare ko-coordination allows tense mismatches, we ran a 2 × 2 mixed-effects model with SYNTAX [Full vs. Bare] and TENSE as fixed effects and the maximally convergent random effects structure (i.e., by-participant and by-item intercepts and a by-participant slope for each of the two fixed factors). The results of the model are summarized in Table 1.

Table 1:

Fixed effects (SYNTAX × TENSE) summary for bare ko-coordination.

	Estimate	SE	t	p
(Intercept)	0.775	0.048	12.045	***
SYNTAX [Full vs. Bare]	−0.200	0.057	4.105	***
TENSE	−0.012	0.063	−4.883	0.769
SYNTAX:TENSE	−0.295	0.086	−9.560	***

1. *p < 0.05, **p < 0.01, ***p < 0.001.

There was a significant effect of SYNTAX [Full vs. Bare] but no significant effect of TENSE. However, there was an interaction, which revealed that the difference in acceptability between the tense-match and tense-mismatch cases in the [Bare] condition was greater than that in the [Full] condition. Post-hoc tests confirmed that the tense-mismatched [Bare] condition suffered a separate penalty: [Full | Match] versus [Full | MisMatch] (β = 0.012, SE = 0.040, t = 0.295, p = 1.000) and [Bare | Match] versus [Bare | MisMatch] (β = 0.306, SE = 0.063, t = 4.858, p < 0.001). This suggests that tense mismatches inflict a degradation in bare ko-coordination.

Second, to test if gapping allows tense mismatches, we fit a model for SYNTAX [Full vs. Gapping] crossed with TENSE. The results of the model are summarized in Table 2.

Table 2:

Fixed effects (SYNTAX × TENSE) summary for gapping.

	Estimate	SE	t	p
(Intercept)	0.775	0.032	24.526	***
SYNTAX [Full vs. Gapping]	0.034	0.038	0.905	0.369
TENSE	−0.012	0.038	−0.309	0.758
SYNTAX:TENSE	−1.121	0.091	−12.281	***

There was neither a significant effect of SYNTAX [Full vs. Gapping] nor a significant effect of TENSE. However, there was an interaction, which revealed that the difference in acceptability between the tense-match and tense-mismatch cases in the [Gapping] condition was greater than that in the [Full] condition. Post-hoc tests confirmed that the tense-mismatched [Gapping] condition suffered a separate penalty: [Full | Match] versus [Full | MisMatch] (β = 0.012, SE = 0.038, t = 0.309, p = 1.000) and [Gapping | Match] versus [Gapping | MisMatch] (β = 1.133, SE = 0.084, t = 13.435, p < 0.001). This result indicates that tense mismatches degrade gapping.

Further, a visual inspection of the plot in Figure 1 suggests that the penalty due to tense mismatches was greater in gapping relative to bare ko-coordination. We thus ran another 2 × 2 mixed-effects model with SYNTAX [Bare vs. Gapping] and TENSE. The results of the model are summarized in Table 3.

Table 3:

Fixed effects (SYNTAX × TENSE) summary for bare ko-coordination versus gapping.

	Estimate	SE	t	p
(Intercept)	0.575	0.048	12.045	***
SYNTAX [Bare vs. Gapping]	0.234	0.057	4.105	***
TENSE	−0.306	0.063	−4.883	***
SYNTAX:TENSE	−0.826	0.086	−9.560	***

There were significant effects of SYNTAX [Bare vs. Gapping] and TENSE.^[6] The interaction was also significant, indicating that the difference in acceptability between the tense-match and tense-mismatch cases in the [Gapping] condition was greater than that in the [Bare] condition. Post-hoc tests confirmed that the tense-mismatched [Gapping] condition suffered a separate penalty: [Bare | Match] versus [Bare | MisMatch] (β = 0.306, SE = 0.063, t = 4.883, p < 0.001) and [Gapping | Match] versus [Gapping | MisMatch] (β = 1.133, SE = 0.084, t = 13.513, p < 0.001).

Another goal of this study was to examine whether temporal order affected the amount of penalty on acceptability due to tense mismatches in bare ko-coordination and gapping. As planned, we added ORDER (Sequential vs. Reverse) as another predictor in the analyses. Figure 2 presents the mean acceptability of the six experimental conditions split by ORDER: the left panel for the results of the [Sequential] ones (i.e., the first 12 sets) and the right panel for that of the [Reverse] ones (i.e., the latter 12 sets).

Figure 2:

Mean acceptability of 12 sequentially ordered sets (left panel) and 12 reversely ordered sets (right panel) of experimental conditions.

Overall, the two plots of Figure 2 show very similar patterns. A visual inspection of the plots suggests that ORDER did not make much influence on the TENSE effect (i.e., the penalty due to tense mismatches) in both the [Bare] and [Gapping] conditions. While there seems to be no TENSE effect in the [Full] condition in general (i.e., their mean acceptability z-scores seem to overlap), both the [Bare] and [Gapping] conditions seem to show a clear difference between [Match] versus [MisMatch] in both the [Sequential] and [Reverse] conditions.

In order to examine the role of temporal ORDER through statistical modeling, we ran two additional sets of a 2 × 2 mixed-effects model for the effect of ORDER on the tense-mismatch penalty in the [Bare] and [Gapping] conditions. Table 4 presents the summary of the 2 × 2 mixed-effects model (TENSE × ORDER) within the [Bare] condition, and Table 5 summarizes the 2 × 2 mixed-effects model (TENSE × ORDER) within the [Gapping] condition.

Table 4:

Fixed effects (TENSE × ORDER) summary for bare ko-coordination.

	Estimate	SE	t	p
(Intercept)	0.582	0.056	10.349	***
TENSE	−0.353	0.081	−4.349	**
ORDER	−0.015	0.058	−0.250	0.803
TENSE:ORDER	0.093	0.090	1.032	0.305

Table 5:

Fixed effects (TENSE × ORDER) summary for gapping.

	Estimate	SE	t	p
(Intercept)	0.815	0.045	18.206	***
TENSE	−1.117	0.094	−11.908	***
ORDER	−0.013	0.059	−0.222	0.825
TENSE:ORDER	−0.032	0.083	−0.383	0.704

In both bare ko-coordination and gapping, the effect of TENSE was significant, but the effect of ORDER was not. Also, the effect of the two-way interaction TENSE × ORDER did not reach significance. Together, the results confirmed that temporal order did not affect the amount of penalty due to tense mismatches in both bare ko-coordination and gapping.

4 General discussion and syntactic analysis

There were two main findings from the experiments reported above. First, the acceptability of both bare ko-coordination and gapping was degraded by tense mismatches but that of full coordination was not. However, the tense-mismatch penalty was different between bare ko-coordination and gapping. As given in Table 3, tense mismatches induced a significantly greater amount of degradation in gapping than in bare ko-coordination. As a result, tense mismatches made bare ko-coordination moderately acceptable (mean: 0.268), but gapping with tense mismatches was quite unacceptable (mean: −0.324). This suggests that tense mismatches in gapping inflict a separate penalty relative to those in bare ko-coordination. This finding argues against the claim that gapping permits tense mismatches (Ahn and Cho 2006; Kim 2007; Kim and Cho 2012; Park 2009), defending the claim that gapping requires tense matches (Choi 2019; Kim 2019; Park and Kim 2021). Furthermore, the finding that tense-mismatched bare ko-coordination was moderately acceptable undermines the view that only tense-matched bare ko-coordination is allowed (Kang 1988; Park 1994; Yoon 1996; Yoon 1997), confirming the view that either tense-matched or tense-mismatched one is allowed (Chung 2005b; Kang 2013; Lee and Tonhauser 2010).

Second, temporal order, whether it was sequential or reverse, did not significantly affect the acceptability of bare ko-coordination and gapping. This finding defends the claim that bare ko-coordination can be presented in any temporal order without affecting acceptability (Chung 2005b; Lee and Tonhauser 2010), casting doubt on Park and Kim’s (2021) finding that reversely tense-mismatched bare ko-coordination is significantly more acceptable than sequentially tense-mismatched bare ko-coordination.

As pointed out by a reviewer, one might wonder why our experimental finding is different from Park and Kim’s (2021) with respect to bare ko-coordination. We want to point out that Park and Kim’s experiment had one important issue that differentiates it from the experiment reported in this paper. While the experimental design was similar, Park and Kim split their 2 × 2 × 3 design (TENSE × ORDER × SYNTAX in our terms) into two sub-experiments, treating TENSE as a between-subject factor. In other words, they ran two instances of a 2 × 3 (ORDER × SYNTAX) experiment with two different groups of participants: one group saw only the tense-matched conditions and the other group saw only the tense-mismatched conditions, in addition to fillers. In their experimental setting, the distribution of acceptability ratings could be very different between the two groups: while the responses from the group with only tense-matched conditions would be predominantly of high acceptability, the responses from the group with only tense-mismatched conditions would be predominantly of low acceptability, except for the full-tensed-coordination conditions. This issue has been referred to as an equalization strategy in response (cf. Sprouse 2009): participants tend to balance their responses throughout an entire experiment between, for example, high ratings and low ratings. As attested by experimental findings from both Park and Kim and the current study, tense-matched conditions turned out to be significantly more acceptable than tense-mismatched conditions with respect to bare-ko coordination and gapping. It was thus very likely that the overall acceptability ratings of experimental items from the two participant groups were influenced by the different proportions of high acceptable items and low acceptable items, respectively. In particular, the high acceptability of the [Reverse | Bare | MisMatch] condition (raw-score mean: 5.41) in Park and Kim’s results, which was even slightly higher than the [Reverse | Bare | Match] condition (raw-score mean: 5.35), could have been due to the participants’ equalization strategy in response. Since the participants in this group needed to give a low acceptability rating for the four conditions (i.e., the [MisMatch] conditions in [Gapping | Sequential], [Gapping | Reverse], [Bare | Sequential], and [Bare | Reverse]) out of the six conditions, which is significantly more frequent than the conditions of the other group, there is a high chance that they might have made a conscious effort to achieve balance among all of their ratings, and it might have distorted the ratings on the [Reverse | Bare | MisMatch] condition towards an unfairly higher acceptability.^[7]

With the above discussion in mind, we propose that the experimental stimuli in (23) are derived in the following way:

The [Full | Match] condition in (24a) and the [Full | MisMatch] condition in (24b), both of which involve coordination of full clauses, are acceptable because they have an independent tense projection, i.e., TP, in each coordinated clause. Since there is no difference in acceptability between the two conditions, it can be concluded that tense mismatches in full coordination do not affect acceptability, as expected.

Considering the acceptability of [Bare | Match] in (24c) and [Bare | MisMatch] in (24d), we propose that the conjunct size in bare ko-coordination is TP, similar to full coordination. One minimal difference is that the first conjunct in (24c) and (24d) may have a null tense morpheme in T (Chung 2005b), which is independent from the overt tense marker in the second conjunct. This null tense is interpreted, via the congruence with a temporal cue (i.e., the year before last), as [Past] in both (24c) and (24d). This analysis thus supports the claim made by Chung (2005b), Kang (2013), and Lee and Tonhauser (2010). Since bare ko-coordination has missing tense information that the parser needs to recover for a successful interpretation, the cost of such a process will be reflected in the acceptability.^[8] Regarding the slightly degraded acceptability of [Bare | Match] compared with the acceptability of [Full | Match], we suggest that identifying the null tense information via a temporal adverbial causes a processing difficulty.^[9] The tense-mismatched situation will present an additional challenge to the parser. In this regard, we thus suggest that the tense-mismatch penalty further lowers the acceptability of [Bare | MisMatch] compared with that of [Bare | Match].^[10]

However, tense mismatches significantly decrease the acceptability ratings of [Gapping | MisMatch] in (24f). The low acceptability of [Gapping | MisMatch] suggests that the missing tense of the gapped clause in gapping, which is coordinated at the vP level, is determined via a structural notion like c-command by the higher overt tense-marker of the antecedent clause, thus defending the claim made by Chung (2005a) and others (Choi 2019; Kim 2019; Park and Kim 2021).

Before concluding this section, we wish to answer the question, raised by a reviewer, of why tense-matched gapping (= [Gapping | Match]) was more acceptable than tense-matched bare ko-coordination (= [Bare | Match]). Although there was a statistically significant difference of acceptability ratings between [Gapping | Match] and [Bare | Match] (β = −0.234, SE = 0.054, t = −4.301, p < 0.01), it seems that the difference should be addressed from the perspective of processing, given that both were judged as well-formed (mean: [Gapping | Match] 0.809 vs. [Bare | Match] 0.575). To be specific, we suggest that [Bare | Match] places an extra processing challenge relative to [Gapping | Match], which would explain the difference in acceptability. In the case of [Gapping | Match], the tense information of two coordinated vPs was simultaneously decided by the tense marker in T of the antecedent clause, as schematized in (24e) and (24f). Thus, there was less leeway for the parser to intervene. Meanwhile, in the case of [Bare | Match], the parser needs to interpret the null tense morpheme in the first conjunct, as illustrated in (24c) and (24d), with other clause-internal temporal cues (i.e., the temporal adverbial). This would cause a processing difficulty.

5 Conclusions

In this experimental study, we showed that tense mismatches greatly lower the acceptability of bare ko-coordination and gapping, but not that of full coordination, regardless of temporal order. We thus defended the claim that tense-mismatched gapping is not acceptable in Korean (Choi 2019; Chung 2005a; Kim 2019), contra Ahn and Cho (2006), Kim and Cho (2012), and Park (2009). We also showed that the influence of temporal order on the acceptability of bare ko-coordination is not significant, contra Park and Kim (2021). In addition, unlike the previous empirical studies such as Park and Kim (2021), we provided a theoretical analysis of tense mismatches of the constructions at issue: full coordination, bare ko-coordination, and gapping. We accounted for the gradable tense-mismatched phenomena of the three constructions via the difference of conjunct sizes: For full coordination and bare ko-coordination, the conjunct size is TP since each conjunct has an independent (overt or covert) tense T. To be more specific, for full coordination, there are two independent overt Ts. For bare ko-coordination, there is a null T in the first conjunct and an independent overt T in the second conjunct. For gapping, the conjunct size is vP; there is only one T, which c-commands both conjuncts but is located only in the second conjunct. In sum, the tense-mismatch effects in three types of coordinate constructions are derived from the conjunct size (i.e., height). While full coordination and bare ko-coordination allow the tense mismatch between conjuncts, gapping does not.