LI Jie

This study conducted an eye-tracking experiment on processing different patterns of Chinese familiar metonymy in sentential contexts. It analyzes five eye-tracking measures concerning the processing of metonymy. The results indicate that different patterns of metonymy experience different processing processes under a sentential-context condition， and results in prototype effects. The main finding is that Spatial Part & Whole metonymy is more prototypical than other three patterns of metonymy， i.e.， Container and Contained， Location and Located， Entity and Adjacent Entity， and that the effect of metonymy pattern on the processing is stable and observable. It concludes that contextual information facilitates the processing of non-prototypical metonymy， but restrain the processing of prototypical metonymy.

Keywords： eye-tracking， pattern effects， familiar metonymy， processing， sentential contexts

1. Introduction

This paper examines how various patterns of Chinese metonymy are processed back-grounded in a sentential context. Different from relevant studies （Frisson & Pickering， 1999； Fass， 1997； Gibbs， 1999； Rapp， 2011； Joue et al.， 2018） that were mainly focused on the categorization process or the processing of metonymic senses， concerning the conceptualization of metonymy， this survey investigates to what extent the type of substitution relationship （based on different types of contiguity） affects the processing of metonymy. The central issue under discussion is whether there exits possible prototype effects generated by metonymic patterns during processing ， which has not been explored by experiments so far.

Previous experimental research on metonymy scattered over a few issues. Some of them look into matronymic sense. For example， Gerrig （1989） distinguished the processing of familiar sense and unfamiliar sense of metonymy by measuring sentence reading time in highlighted contexts： Unfamiliar sense cost longer time to process than familiar sense. Others are interested in the syntactic ambiguity resolution in processing metonymy， like Pickering and Traxler （1998）， who carried out an eye-tracking study in which participants were asked to read a sentence （including metonymic expressions） in context and then a syntactically ambiguous target sentence. They found that plausible metonymic interpretation was obtained under an available contextual condition and that the process of syntactic ambiguity resolution is thus affected. Still， some others investigate contextual effects on metonymy processing. Unlike studies by language philosophers （Grice， 1975， 1989； Searle， 1979） maintaining that the processing of figurative language is first interpreted as literal ones in a given context and activated after the first interpretation fails， experimental results （Pynte et al.， 1996； Joue et al.， 2018） indicated that figurative language is processed as fast as literal language if supported by strong contextual information， i.e.， with closely relevant context. Yet， few has discussed the type effect in processing figurative expressions.

Several typical models have been proposed to account for figurative language processing. Some typical ones like literal-first （literal sense is processed first）， figurative first （figurative sense is processed first） and parallel model （with fully specified and underspecified versions）. Frisson and Pickering （1999） compared the time course of the processing of metonymic expressions with the literal ones in two eye-tracking experiments to figure out whether people rapidly access a familiar metonymic interpretation for a noun and whether the processing of nouns that are ambiguous between a literal and a metonymic sense is informative about figurative language processing （1999， p. 1367）. Their findings show that the two patterns of metonymy （place-for-institution and place-for-event metonymy） overall tend to support an underspecified account of the parallel model， though they differ slightly in the time course. Inspired by this finding， this study investigates the four patterns of metonymy constructed as a prototypical category （Peirsman & Geeraerts， 2006） to find out whether different patterns of metonymy experience the same processing process.

2. Methodology

2.1 Assumption

Peirsman and Geeraerts （2006） constructed a category of metonymy based on their analysis of the notion of contiguity， and structured the Spatial Part and Whole relations （henceforth SW ） as the core of the category. Deviating from the core， they （2006） framed other relations of contiguity in that category and classified other three metonymic patterns， i.e.， Container and Contained （CC）， Location and Located （LL）， Entity and Adjacent Entity （EA）， as the members of the prototypical structure. This experiment attempts to figure out whether prototype effects of metonymy processing is observed in different metonymic patterns under a sentential-context condition. If yes， it means metonymic meaning of some pattern（s） is more prototypical than the rest ones. It assumes that there are no significant differences among the four patterns of metonymic processing in both familiar and unfamiliar metonymic types under context condition.

2.2 Materials

38 （6 for practice and 32 for experiment） sentences including familiar metonymy are used as experimental materials， including four sub-categories （matching the four patterns of metonymy）： Spatial Part and Whole， Container and Contained， Location and Located， Entity and Adjacent Entity. In each sub-category there are four groups of metonymic sentences， each of which is paired with a literal sentence. In total， there are eight sentences in every sub-category： fourare metonymic and the rest literal. All the materials are presented by a software in a random order.

The materials are selected according to the results of a pretestso that the frequency， and revised from the original sentences； the length ranges from 41 to 45 words （Mean = 44）. The metonymic construction are included in a four-word phrase and occur in the middle of the sentence. The four-word phrase is one of the three regions in the experiment. After this region is identified， 32 native-Chinese graduate students majoring in Modern Chinese are asked to identify the other two four-word phrases that are most related before and after that region as the other two regions for exploration. According to the results， the rest two regions for observation are identified， for example，

白宮 （White House）—总统（President）

（1） /近年以来/Region①，/美国白宫（1总统/2法律）/ Region ②很繁忙，重要事件接连发生，都得/亲自处理/ Region ③。

Translation： /Since recent years/Region ①， / the White House of the United States （1 president / 2 law）/Region ② has been very busy. Important events take place one after another and /he has to handle them in person/Region ③.

（2） /会议期间/Region ①，/美国白宫（1楼房/2战争/Region ②很热闹，重要人物接连进去，都是/亲自参加/Region ③。

Translation： /Since the meeting/Region ①， /the White House of the United States （1 building / 2 war）/ Region ②has been very busy. Important people come in one after another and/they attend it in person/Region ③.

Sentences offering contextual information are constructed in balanced grammatical， semantic， and plausibility conditions and matched for word frequency， semantic association length and syntactic structure. They arealso chosen to exemplify a wide variety of metonymic readings. （c.f. Filik &Moxey， 2010）

2.3 Research Methods

Design： The experiment is designed in a one-factor framework. The pattern factor includes four levels： Spatial Part & Whole， Container & Contained， Location & Located and Entity & Adjacent Entity.

Participants： 36 （16 male， 20 female） native Chinese-speaking volunteers are recruited to participate the experiment. All the participants are undergraduate students with agerange from 21 to 23 years old， right-handed， and have normal or corrected-to-normal vision. None of them has psychological or brain disorders.

Instrument： An EyeLink 1000 eye-trackeris used to record the eye movements. It is connected to a PC computer with a 21” LCD monitor for stimulus presentation hardware and running at 1000 Hz sampling rate. Desktop Mount for the EyeLink 1000 system sits just below the display that the participant is looking at. Binocular recording model is used in the experiment， and only the data from the right eye is further analyzed. A keyboard is used for response. SR Research Experiment Builder and Data Viewer are used respectively for programming and data collection and processing.

Procedure： The presentation of materials and recording of latencies are controlled by the software Experiment Builder for Eyelink 1000. At the beginning of the experiment， all participants are presented with a 13-point grid in order to calibrate their eye movements with the locations on the screen. If the initial calibration fails，participants are presented with a 9-point calibration grid instead. Recalibration repeats throughout all the experiment whenever necessary.The instruction is first presented and then the practice trial. Participants are told to read sentences silently at their own pace to comprehend the sentence， and allowed to practice many times until they fully know how to response. The experimental task is to choose an appropriate meaning for metonymic or literal phrase in bold. The participants viewed the one-line sentence binocularly on a LCD monitor from a distance of 71 cm， but only the participants dominant eye is tracked. The experimental items are presented in a random order； judgment is made in the end of the sentence.

Participants are set up individually in the eye tracker and are asked to respond as accurately as possible after they read the experimental materials. Before that， they are instructed to put their index fingers respectively onto the “F” and “J” buttons on the keyboard to choose the correct meaning for the phrase in bold. After the participants read the instruction part， they can press any key to enter the practice trial， which includes six sentences and is arranged in the same way as the experiment trial. The trial begins with an annulus fixation point in the centre on the screen and lasts 500 ms to attract the participants attention. It then follows the sentence and the two choice items for the phrase being tested. Once the participants respond， the stimuli disappear immediately. It take approximately 20 minutes to finish the whole experiment.

2.4 Measures

Data from 35 participants are collected and one participant is excluded due to the failure in collecting the data. The data from the practice trial are first filtered. The accuracy rate for the all the subjects is above 90%. Further statistic analysis is carried out to test the hypotheses for this experiment. Several eye-tracking measures（Clifton et al， 2007） were analyzed to detect various possible effects of metonymy type in the processing： First Fixation Duration （Mitchell et al， 2008） is employed to analyze cognitive processes concerning word frequency， lexical access （Reichle et al， 2003） and some gross syntactic anomalies （Pickering et al， 2003）； First-pass Time is counted as one of the “early” measures （Mitchell et al， 2008）； Regression Rate is crucial to distinguish first-pass regressions from later measures of regressions； Regression Path Duration is calculated to detect higher level processes， such as semantic integration， discourse processes， and even pragmatic processing； Total Time is the sum of all fixations on a word or in a region （Clifton et al.， 2007）. Under different circumstances， these measures are interpreted differently to cover different cognitive processing.

Results and Analysis

The five standard eye-tracking measures in different patterns are first analyzed to work out if there exits prototype effects of metonymic patterns during the processing. The eye-tracking data of the four patterns of metonymy， i.e.， Spatial Part & Whole， Container & Contained， Location & Located and Entity & Adjacent Entity， are treated as the four levels of the within-subject factor. Pairwise comparisons is also carried out among the four to probe the source of the difference if there is a one.

The results from the First Fixation Duration show that only region 3 differ significantly from the rest three patterns： F （3，102） = 3.4， p = .020 < .05. The results from the pairwise comparisons show that significant mean difference exists between metonymy of Spatial Part & Whole and Container & Contained： Mean difference（SW-CC） = 30.8， SE = 8.9， Sig. = .009 < .01. The significant difference of the measures on region 3 signifies the difference of lexical access to the following context for the two patterns of metonymy： the Container & Contained metonymic pattern are faster in lexical access than those following the Spatial Part & Whole.

On region 1 the mean differences of theFirst-Pass Time between Spatial Part & Whole and Location & Located， and Spatial Part & Whole and Entity & Adjacent Entity metonymy are significant： Mean difference（SW-LL） = -64.3， SE = 19.1， Sig. = 0.01 < 0.05； mean difference （SW-EA） = -62.9， SE = 21.5， Sig. = .04 < .05. On region 2 the difference is from Spatial Part & Whole and Entity & Adjacent Entity metonymy， mean difference（SW-EA） = -114.7， SE = 33.0， Sig. = .01< .05. On region 3 mean difference （SW-LL） = 110.5， SE = 27.1， Sig.= .00< .05. Theseresults show that the preceding context for Spatial Part & Whole metonymy is much easier to process than Location & Locatedand Entity & Adjacent Entity metonymy. Much less time is spent on Spatial Part& Whole metonymy than on Entity & Adjacent Entity metonymy. It demonstrates that metonymy of Spatial Part& Wholepattern is more prototypical in comparison with Entity & Adjacent Entityin the metonymic structure. It supports Pearsman and Geeraerts （2006） finding that Spatial Part & Whole metonymy is more prototypical.

The results of Regression Rate reveals significant differences in two regions： on region 1F （3，102） = 9.2， p= .00 < .05， and on region 2， the target region， F （3，102） = 6.3， p = .00< .05，. Results from the pairwise comparisons in region 1 show that mean difference （SW-EA） = 0.2， SE = 0.1， Sig. = .01< .05， which means that preceding context for Spatial Part & Whole metonymy is more difficult to process than that for Entity & Adjacent Entity metonymy. Similarly， the result that mean difference （CC-LL） = -0.2， SE = 0.1， Sig. =.01<.05， indicates that preceding context for Container & Contained metonymy is easier to process than that for Location & Located metonymy.

On region 2 more processing difficulty occurs in Spatial Part & Whole metonymy in comparison with Container & Contained and Entity & Adjacent Entity metonymy： Mean difference （SW-CC） = 0.1， SE = 0.05， Sig.= .043 < .05， and mean difference （SW-EA） = 0.1， SE = 0.04， Sig. = .003 < .01. No other significant differences are found among the rest patterns of metonymy.

Significant differences among the four patterns of familiar metonymy are also probed in the first-pass time： F （3，102） = 4.0， p = .009 < .01 on region 1， F（3，102） = 4.8， p = .004 < .01 on region 2 （target region）， and F （3，102）= 6.3， p = .001 < .01 on region 3. We further locate the source for the difference， as shown in table 1.

The results of the three measures above show that early effects of lexical access and processing occur much more frequently in familiar metonymy， and imply that metonymic patterns have a stable and obvious effect on processing familiar metonymy. The results also show that in a very early beginning of processing familiar metonymy， participants are able to locate the most relevant contextual information to facilitate the metonymy processing.

With this mechanism， they even consciously realize to fixate at some relatively closed region in their first fixation time. It might demonstrate that processing familiar metonymy involves some psychological anticipation from the contextual predictability （Frisson， Rayner， & Pickering， 2005）—which is believed to be triggered by some metonymic pattern—that directs their cognitive sources to that particular region. Metonymic processing is a process governed by subjective factor.

Late measures such as Regression Path Duration reflect cognitive cost spent on overcoming the difficulty during the course of late processing （Rayner et al.， 1989）. Results in Table 2： Onthe target region， region 2， it takes much less time to process Spatial Part & Wholemetonymy than the other three patterns ofmetonymy； mean differences of Spatial Part & Whole metonymy differ significantly from other patterns ofmetonymy. Region 3 is the interest area of the following context for the metonymy. Theresult shows that participants spend much more time on this region when they process the Spatial Part & Whole metonymy， and indicates that metonymy of Spatial Part & Whole pattern is more prototypical than other patterns of metonymy.

To combine the results on the two regions （i.e.， region 2 and region 3） of Spatial Part & Wholemetonymy， a clear picture of processing this type of metonymy emerges： Participants do not encounter too much difficulty in processingmetonymy of Spatial Part & Whole pattern， or they could access it at an early stage.However， they needed to spend more time to get contextual information to confirm their first interpretation for the Spatial Part & Whole metonymy， which accounts for the much more time spent on region 3. Thisfinding is supported by the significant differences of the regression rate. The“confirming” Process is similar to the process of “testing hypothesis” In understanding meaning （Wilson， 2004， 2006）， but it emphasizes the dynamicity of on-line meaning processing. Theconfirming process is a result for semantic， pragmatic and some other higher processing. It is also a kind of late effect rather than an early effect in the meaning processing.

Analysis of Total Time （see Table 3） for the four patterns of both familiar metonymy still shows that Spatial Part & Whole metonymy takes longer time to process than Container & Contained metonymy on Region 2 and than Entity & Adjacent Entity metonymy on Region 3. Total Time effect largelyreflects later reinspection of the region with difficulty （Mitchell et al.， 2008）， and the results mean that familiar SW metonymy is much more difficult to process when set in superfluous context. This supports the viewpoint again that SW metonymy possesses the prototypical core of contiguity （Peirsman & Geeraerts， 2006）： it is so prototypical and highly contextualized as a conventional model that no excessive contextual information is required. When extra context is offered， it triggers a process of interpreting non-conventionalized meaning and costs much more time to discriminate other possible understanding， and finally to confirm the processing result. Participants might first treat it as implicated meaning， which usually needs further and evidence-testing inference（Grice， 1975）.

Conclusion

This experiment analyzes five standard eyetracking measures to capture theprocessingof familiar metonymy in a sentential context. The main finding is that Spatial Part & Whole metonymy is more prototypical than other three patterns of metonymy， and that the effect of metonymy pattern on the processing is stable and observable. It also indicates that not all the pieces of contextual information play an equal role in metonymic processing： Some ones are more decisive than others. Different patterns of metonymy experience different processing processes and differ significantly in these processes under a sentential-context condition， and results in prototype effects. Contextualinformation is especially critical for processing non-prototypical metonymy， but is redundant for processing prototypical metonymy.

References

Brysbaert， M.， & Mitchell， D. C. （1996）. Modifier attachment in sentence parsing： Evidence from Dutch. Quarterly Journal of Experimental Psychology， 49A， 664-695.

Clifton， C.， Staub， A.， & Rayner， K. （2007）. Eye movements in reading words and sentences. In R. P. G. Van Gompel， M. H. Fischer， W. S. Murray， & R. L. Hill （eds.）， Eye-movements： A window on mind and brain （pp. 241-372）.Oxford ： Elsevier Science.

Fass， D. （1997）. Processing metonymy and metaphor. London： Ablex Publishing Corporation.

Filik， L.， & Moxey， M. （2010）. The on-line processing of written irony. Cognition， 116， 421-436.

Frisson， S.， & Pickering， M. （1999）. The processing of metonymy： Evidence from eye movements. Journal of Experimental Psychology： Learning， Memory and Cognition， 25， 1347-1365.

Frisson， S.， & Pickering， M. （2001）. Obtaining a figurative interpretation of a word： Support for Underspecification. Metaphor and Symbol， 16（3-4）， 149-171.

Frisson， S.， & Pickering， M. （2007）. The processing of familiar and unfamiliar senses of a word： Why reading Dickens is easy but reading Needham can be hard. Language and Cognitive Processes， 22（4）， 595-613.

Frisson， S.， Rayner， K.， & Pickering， M. J. （2005）. Effects of contextual predictability and transitional probability on eye movements. Journal of Experimental Psychology： Learning， Memory， and Cognition， 31， 862-877.

Geeraerts， D. （2006）. Methodology in cognitive linguistics. In G. Kristiansen et al. （eds.）， Cognitive linguistics： Current applications and future perspectives （pp. 21-50）. Berlin/New York： Mouton de Gruyter.

Gerrig， R. J. （1989）. The time course of sense creation. Memory & Cognition， 17， 194-207.

Grice， P. （1975）. Logic and conversation. In P. Cole & J. Morgan （Eds.）， Syntax and semantics （Vol. 3， pp. 41-58）. New York： Academic Press.

Grice， P. （1989）. Studies in the way of words. Cambridge， MA： Harvard University Press.

Joue， G.， Boven， L.， Willmes， K.， Evola， V.， Demenescu， L. R.， Hassemer， J.， Mittelberg， I.， Mathiak， K.， Schneider， F.， & Habel， U.（2018）. Handling or being the concept： An fMRI study on metonymy representations in coverbal gestures. Neuropsychologia 109， 232-244.

Mitchell， D. C.， Shen， X.， Green， M. J.， & Hodgson， T. L. （2008）. Accounting for regressive eye-movements in models of sentence processing： A reappraisal of the Selective Reanalysis hypothesis. Journal of Memory and Language， 59， 266-293.

Peirsman， Y.， & Geeraerts， D. （2006）. Metonymy as a prototypical category. Cognitive Linguistics， 17（3）， 269-316.

Pickering， M. J.， & Traxler， M. J. （1998）. Plausibility and recovery from garden paths： An eye-tracking study. Journal of Experimental Psychology： Learning， Memory， and Cognition， 24， 940-961.

Pynte， J.， Besson， M.， Robichon， R.， & Poli， J. （1996）. Thetime-course of metaphor comprehension： An event-related potential study. Brain and Language， 55， 293-316.

Rapp， et al. （2011）. Neural correlates of metonymy resolution. Brain & Language， 119， 196-205.

Rayner， K.， & Well， A. D. （1996）. Effects of contextual constraint on eye movements in reading： A further examination. Psychonomic Bulletin & Review， 4， 504-509.

Reichle， E. D.， Rayner， K.， & Pollatsek， A. （2003）. The E-Z reader model of eye-movement control in reading： comprehensions to other models. Behavioral and Brain Sciences， 26， 445-526.

Searle， J. （1979）. Metaphor. In A. Ortony （Ed.）， Metaphor and thought （pp. 92-123）. Cambridge， England： Cambridge University Press.

Wilson， D.， & Carston， R. （2006）. Metaphor， relevance and the “emergent property” issue. Mind & Language， 21（3）， 404-433.

Wilson， D.， & Sperber， D. （2004）. Relevance theory. In G. Ward and L. Horn （eds.）， Handbook of pragmatics （pp. 607-632）. Oxford： Blackwell.