Exploring Interpretation and Misinterpretation of Garden-Path ...

1 downloads 0 Views 762KB Size Report
Garden path sentences have long been used as test material by ... In Section 3 examples of Polish garden path sentences are presented ..... begin the next trial.
International Review of Pragmatics (2014) 1–30 brill.com/irp

Exploring Interpretation and Misinterpretation of Garden-Path Sentences in Polish Agnieszka Solska University of Silesia, Poland [email protected]

Arkadiusz Rojczyk University of Silesia, Poland [email protected]

Abstract The paper reports on an experiment designed to investigate the interpretation and misinterpretation of Polish sentences such as Gdy Jan pisał list spadł z biurka (e.g. “While John was writing the letter fell off a desk”), paralleling English garden path sentences such as While John hunted the deer ran into the woods. The locally ambiguous np-v-np region in sentences of this sort may result in the key noun phrase being incorrectly interpreted as the object in the fronted temporal clause. The experiment was conducted using E-Prime software and involved comparing the participants’ comprehension of garden path sentences in three different conditions with their comprehension of equivalent non-garden path sentences. This was achieved by examining the participants’ performance on comprehension questions probing the interpretation of the subordinate clauses, establishing the times they needed to answer the questions and checking the level of their confidence in their responses. The goals of the experiment were to establish whether the Polish sentences in question force the comprehender to form erroneous interpretations which persist in his mind and to determine to what extent the persistence of the initial incorrect interpretation is influenced by the morphosyntactic properties peculiar to Polish. The results of the experiment indicate that though the Polish sentences under investigation do not trigger a particularly strong garden path effect, the information carried by the case endings and the thematic roles nouns are likely to have in specific contexts affect the way locally ambiguous sentences are analysed.

© koninklijke brill nv, leiden, 2014 | doi: 10.1163/18773109-01802100

2

solska and rojczyk

Keywords garden path sentences – reanalysis – ambiguity – natural language processing

1

Introduction

Garden path sentences have long been used as test material by psycholinguists in their attempts to gain insight into natural language processing. These confusing sentences owe their name and their popularity with researchers to the fact that they entice the comprehender into constructing a parse and hence an interpretation which turns out to be incorrect and has to be reanalysed. For example, on being presented with sentence (1) the comprehender forms an initial interpretation in which the string hunted the deer is treated as a legitimate verb phrase, even though such an interpretation is syntactically unlicensed and will turn out to be untenable. The smooth processing of the input will break down as soon as the verb ran is encountered, which will act as an error signal prompting the comprehender to reconsider his initial assumption. (1)

While John hunted the deer ran into the woods. (i.e. The deer ran into the woods, while John hunted.)

The expected course of action, shown in Figure 1, would be for the comprehender to either instantly reject the sentence as grammatically incorrect or to attempt reanalysis and either fail in doing so or end up performing a successful repair. rejection anticipatory error ↗ → assumption signal ↘

rejection back tracking/ ↗ “forward tracking” ↘ reanalysis → repair

figure 1

The expected stages in the online interpretation of a garden path sentence

Studies involving monitoring eye movements of people reading such sentences, for instance those conducted by Frazier and Rayner (1982), indicate that indeed most comprehenders attempt to revise their initial interpretation. However, an interesting fact, first observed by Christianson et al. (2001), is that the reanaly-

10.1163/18773109-01802100 | International Review of Pragmatics (2014) 1–30

3

exploring garden-path sentences in polish

sis is not always complete, i.e. “the initial, ultimately incorrect, interpretation tends to persist in the minds of the comprehenders” (Christianson et al., 2001: 368–407). Thus comprehenders of (1), instead of readjusting the initial misapprehension that John hunted the deer, continue to believe that the man was indeed chasing the deer in the woods. The interpretive steps undertaken in the online recovery of the linguistic meaning of a garden path sentence are therefore more accurately represented in Figure 2, in which the persistence of the initial misapprehension is indicated as “partial repair”: rejection anticipatory error ↗ → assumption signal ↘

rejection back tracking/ ↗ ‘forward tracking’ ↘ reanalysis

figure 2

↗ ↘

partial repair complete repair

The actual stages in the online interpretation of a garden path sentence (Solska, 2008:24)

This paper reports on an experiment designed to investigate the interpretation and misinterpretation of Polish sentences, such as (2), paralleling the English example above. The experiment had two major objectives: (i) to establish whether the Polish sentences in question do indeed trigger the garden path effect, i.e. whether they too force the comprehender to form erroneous interpretations which persist in his mind and (ii) to determine to what extent the persistence of the initial incorrect interpretation is influenced by the morphosyntactic properties peculiar to Polish. (2) Gdy Jan pisał list spadł z biurka. while John write.3sg.pst letter.sg.nom fall.3sg.pst off desk “While John was writing the letter fell off the desk”. The structure of the paper is as follows. Section 2 focuses on the sources of the garden path effect. In Section 3 examples of Polish garden path sentences are presented and discussed. Section 4 reports on the experiment we have conducted. The general discussion of the findings emerging from the experiment is presented in Section 5 and is followed by concluding remarks in Section 6.

International Review of Pragmatics (2014) 1–30 | 10.1163/18773109-01802100

4 2

solska and rojczyk

The Sources of the Garden Path Effect

2.1 Local Ambiguities The interpreting problems encountered by the readers of garden path structures result from the presence of a local ambiguity in the linguistic string being processed, which acts as a cut-off point endorsing more than one interpretation, only one of which will turn out to be viable. As Pritchett put it (1992: 6), garden path sentences contain a place of indeterminacy “in which some locally tenable choices will ultimately lead to global ungrammaticality and resultant processing failure”. The ambiguities in question can be structural, categorial or lexical. For instance, in sentence (1) the verb hunted in the fronted temporal clause can be treated as a transitive verb followed by the object noun phrase the deer or as an intransitive verb followed by the noun phrase the deer, acting as the subject of the main clause. In (3), the word old can be interpreted as an adjective premodifying the head noun in a noun phrase or as the head of a noun phrase. Moreover, the word man is locally ambiguous between a nominal and a verbal reading. In (4), which is a classic example of a garden path sentence, the verb raced can be perceived as either a past tense form or as a passive participle and the noun phrase the horse can be treated as an entire noun phrase or as the head of a noun phrase postmodified by a non-finite relative clause. (3) The old man the boat. (i.e. The boat is manned by the old people.) (4) The horse raced past the barn fell. (i.e. The horse which was raced [by someone] past the barn fell [down].) In order to recover from the initial misanalysis of a locally ambiguous string the comprehender has to disambiguate ambiguous items and to revise his judgments concerning the subcategorization frame of the verbs used. These tasks rarely pose a problem if the linguistic material is presented in speech since pauses and changes in intonation tend to indicate the constituent structure of a potentially ambiguous material. In written discourse, punctuation can sometimes disambiguate a potentially confusing structure, as can be seen in example (5). (5) While John hunted, the deer ran into the woods. Indeed, all of the examples used by Christianson et al. (2001) in their experiments would be disambiguated by a simple expedient of inserting a comma

10.1163/18773109-01802100 | International Review of Pragmatics (2014) 1–30

exploring garden-path sentences in polish

5

after the fronted clause. However, the existence of sentences such as (3) and (4) above shows that a disambiguating comma is not always an option. 2.2 Accounts of the Garden Path Effect The reason why comprehenders almost invariably get lured into choosing the interpretation which will have to be revised is a matter of debate. The most widely accepted view, advocated by the proponents of the so-called garden path model of sentence comprehension (e.g. Frazier and Fodor, 1978; Frazier and Rayner, 1982; Ferreira and Henderson, 1991), is that the garden path effect is a result of two principles which guide sentence processing, namely the Principle of Late Closure and the Principle of Minimal Attachment. The former describes the tendency of a parser to attach new words or phrases to the phrase or clause currently being processed and the latter refers to a strategy of parsimony, which prompts the parser to initially attempt to construe sentences in terms of the simplest syntactic structure compatible with the input so far decoded. In sentence (1), the Principle of Late Closure would thus prevent the noun phrase the deer from being attached outside the fronted temporal clause as this would result in the early closure of the argument structure of the subordinate verb despite the fact that the input string contains a seemingly legitimate object. At the same time, the Principle of Minimal Attachment would cause the parser to interpret the second np in a np-v-np sequence John hunted the deer as the direct object, since not to do so would require the parser to construct a more complex structure. A more semantically-motivated view is taken by Pritchett (1988, 1992), who attributes the existence of the garden path effect to the so-called Theta Reanalysis Constraint, which disallows costly mental operations such as a semantic reanalysis of linguistic data involving the reinterpretation of a theta-marked constituent as outside of a current theta domain. On this proposal, the reason why the comprehender of (1) does not instantly interpret the deer correctly as the agent in the main clause is the fact that to move the np the deer from the theta-domain of the predicator hunted into the domain of the predicator ran and to change its theta role from patient to agent would require undue effort on the part of the processor. The problem with both of these approaches is that while capturing certain tendencies exhibited by language users on encountering ambiguities, they fail to account for all instances of garden path utterances. As Solska (2008) tried to demonstrate, in examples such as (4) the Principle of Late Closure should block rather than trigger the garden path effect, since it should prevent the comprehender from closing off the noun phrase immediately after the head noun horse. In that same sentence, the Principle of Minimal Attachment should

International Review of Pragmatics (2014) 1–30 | 10.1163/18773109-01802100

6

solska and rojczyk

prevent the comprehender from starting to build a vp node despite the fact that the incoming material could be attached to the current np node. Moreover, neither account can explain why not all sentences containing local ambiguities lead to a processing breakdown. For instance, sentences (6) and (7), modelled on examples (1) and (4), do not seem to trigger the garden path effect despite having exactly the same syntactic structures and the same set of theta domains as their garden path counterparts. (6) While the man hunted a meteorite crashed into the woods. (7) The horse raced at the Belmont died. Arguably, a more promising account of why the garden path effect arises in some but not all locally ambiguous structures can be found in the pragmatic framework of utterance comprehension developed by Sperber and Wilson (1986/1995). On this proposal, which advocates the radically inferential character of interpreting verbal inputs both at word and utterance level, untenable interpretations may sometimes be the first to access due to the combination of two factors: (i) the interpretive strategy followed by the language user during the online process of deriving meanings, which involves considering interpretations in their order of accessibility and (ii) the lexical, logical and encyclopaedic information provided by the concepts encoded by the words and phrases in the incoming material. The information conveyed by the lexical expressions creates a context (understood as the language user’s cognitive environment) in which the sentence is processed and which makes certain scenarios, hence certain interpretations, more prominent in the comprehender’s mind and therefore more likely to be accessed or constructed first. What makes (1) different from (6) is the fact that the encyclopaedic entry for deer, unlike the encyclopaedic entry for meteorite, contains information about being a potential quarry in a hunt. The difference between the conceptual content of (4) and (7) is that the phrase at the Belmont, unlike the phrase at the barn, provides information connected with a horse-racing competition at which horses do not merely race but are raced against each other. In constructing interpretations the comprehender draws inferences based on contextual assumptions available to him and follows the strategy of taking a path of least effort in computing cognitive effects, so he is bound to accept the first interpretation as the one intended by the addresser. Thus, depending on the conceptual content of a locally ambiguous input, the inferential processes which normally help comprehenders resolve interpreting difficulties may lead them astray and impede the correct understanding of a sentence.

10.1163/18773109-01802100 | International Review of Pragmatics (2014) 1–30

exploring garden-path sentences in polish

7

The fact that inferential processes play a role in utterance comprehension is not lost on psycholinguists (cf. Bransford and Franks, 1971; Bransford and Johnson, 1972, 1973). Indeed the way pragmatic inferences may affect the understanding of certain types of garden path sentences was investigated in one of the experiments conducted by Christianson et al. (2001). Nonetheless, they tend to perceive interpreting verbal inputs as primarily involving linguistic decoding, i.e. creating syntactic representations of verbal inputs and constructing final semantic interpretations based on those representations. The position they take on pragmatic inferences is that rather than affecting all aspects of verbal understanding they are merely processes which “can intrude upon both sentence recall and discourse comprehension” (Christianson et al., 2001: 371).

3

Garden Path Sentences in Polish

Interestingly, examples of garden path sentences presented and discussed in Polish literature tend to involve global rather than local ambiguities. For instance, Marta’s response (8b), może to olej, in the exchange presented below, taken from the Polish translation of Sperber and Wilson’s (1995) book Relevance: Communication and Cognition (Sperber and Wilson 2011: 255), is used as an example of a garden path utterance, even though it does not have one but two sustainable interpretations, namely (9a) and (9b), and remains ambiguous after the whole sentence has been read. Utterance (8b) is produced in response to Ania’s observation in (8a) that there is something leaking in the car (Coś mi w tym aucie cieknie), so the reader is likely to treat it as pointing to a possible source of the leak in the car, i.e. to form interpretation (9a). The remark becomes a garden path utterance only in the light of Ania’s response in (8c), which clearly demonstrates that, unlike the reader, she interpreted it as conveying the idea presented in (9b). In other words, she understood Marta’s words as a colloquial way of saying Perhaps ignore it, i.e. as giving advice as to what Ania should do. (8) a. Ania: Coś mi w tym aucie cieknie. Something me.dat in this car leak.3sg.prs “There is something leaking in the car”. b. Marta: Wiesz, może to olej. know.2sg.prs, [interpretation (9a) or interpretation (9b)] “You know, perhaps it’s oil” or “You know, perhaps ignore it”.

International Review of Pragmatics (2014) 1–30 | 10.1163/18773109-01802100

8

solska and rojczyk

c. Ania: Masz rację, oleję to, jakoś dojedziemy. have.2sg.prs right, ignore1sg.fut it, somehow arrive1pl.fut “You are right, I’ll ignore it, somehow we’ll get there”. sperber and wilson, 2011: 255–256

(9) a. może to [jest] olej perhaps it.sg.nom [be.3sg.prs] oil “Perhaps it is oil”. b. może to olej perhaps it.sg.acc ignore.sg.imper “Perhaps ignore it”. An example of a local ambiguity giving rise to a garden path effect can be found in (10), in which the string został porzucony is more likely to be interpreted as a passive auxiliary followed by a passive participle rather than a sequence of two words, each belonging to a different clause. In other words, it tends to be assigned the unlicensed structure (11a) instead of the correct structure (11b). Needless to say, the ambiguity would evaporate if a mandatory comma were inserted between the two key words, since, as shown in (12), this would clearly indicate that each of them is a constituent of a different clause: (10) Henryk został porzucony odjechał. (Nęcka et al., 2006: 622) (11) a. *[Henryk został porzucony] [odjechał] [Henry get.3sg.pst abandoned.part.nom] [drive off.3sg.pst] “*[Henry got abandoned] [drove off]”. b. [Henryk został] [porzucony odjechał] [Henry stay.3sg.pst] [abandoned.part.nom drive off.3sg.pst] “[Henry stayed] [the abandoned drove off]”. (12) Henryk został, porzucony odjechał. “Henry stayed, the abandoned drove off”. Polish equivalents of sentences used by Christianson et al. (2001) in their experiments are not mentioned in Polish literature discussing the garden path effect. The main reason could be purely “technical”: Polish punctuation conventions require that subordinate clauses be separated from the main clause by a comma, and commas would disambiguate not only all of the sentences in question but also many of their Polish counterparts. Another reason might be that even without the commas the Polish equivalents of many of Christianson et al.’s sentences are not locally ambiguous and thus cannot trigger the garden path effect. For instance, in some sentences in which a v-np string would correspond to a v-np string in Polish, the nominative case ending of the pivotal noun would

10.1163/18773109-01802100 | International Review of Pragmatics (2014) 1–30

exploring garden-path sentences in polish

9

make it an impossible candidate for the object of the verb in the subordinate clause. The Polish equivalents of examples (13a) and (14a) are unambiguous since to be considered the object of such verbs as pił (“drank”) in (13b) or żonglował (“juggled”) in (14b) the Polish equivalents of the pivotal English nouns water and balls would have to appear respectively in the accusative case (wodę) and in the instrumental case (piłeczkami). The Polish equivalent of the English verb phrase hunted the deer in example (1), repeated here as (15a), would not be a v-np string (*polował sarnę) but a v+pp string (polował na sarnę), which renders the Polish translation of (15a), given below as (15b), unproblematic. Finally, Polish lacks “semi-reflexive” verbs so the Polish equivalents of such verbs as dressed (“ubierała”) in (16a) would have to be followed by the reflexive particle się. (13) a. While the woman drank the water spilled on the floor. b. Gdy kobieta piła woda wylała się na while woman drink.3sg.pst water.nom spill.3sg.pst rfl on podłogę. floor (14) a. While the clown juggled the balls fell to the ground. b. Gdy klown żonglował piłeczki upadły na ziemię. while clown juggle.3sg.pst balls.pl.nom fall.3pl.pst on ground (15) a. While John hunted the deer ran into the woods. b. Gdy Janek polował sarna uciekła do lasu. while John hunt.3sg.pst deer.sg.nom run.3sg.pst into woods (16) a. While Anna dressed the baby played in the crib. b. Gdy Anna ubierała się dziecko bawiło się w while Anna dress.3sg.pst rfl child.sg.nom play.3sg.pst rfl in kojcu. crib

4

The Experiment

4.1 Constructing the Test Material Since one of our goals was to determine whether the phenomenon of partial analysis, reported by Christianson et al. (2001), can be observed in Polish, to prepare our test material we had to construct sentences in which the pivotal string would be locally ambiguous. This meant creating complex sentences with a fronted temporal clause whose verb would be optionally transitive and followed by a noun which could potentially be its object. The natural

International Review of Pragmatics (2014) 1–30 | 10.1163/18773109-01802100

10

solska and rojczyk

candidates for verbs of this sort were verbs such as pisać (“write”), rysować (“draw”), gotować (“cook”) or pić (“drink”). The obvious candidates for the locally ambiguous pivotal nouns were nouns with the same form in the nominative case (typically functioning as subjects) and in the accusative case (typically functioning as objects), i.e. nouns such as list (“letter”), mleko (“milk”) or warzywa (“vegetables”). Sentence (2), repeated here as (17), is one of the sentences meeting these criteria: (17) Gdy Jan pisał list spadł z biurka. while John write.3sg.pst letter.sg.nom fall.3sg.pst off desk “While John was writing the letter fell off the desk”. A less obvious choice was to use nouns which could be the potential object of both the fronted and the main clause. One way to achieve this was to use in the main clause the so-called impersonal -no/-to verbal forms, corresponding to the English passive constructions. In this case the pivotal noun could either have an identical form in the accusative and the nominative case, as is the case with the noun list (“letter.sg.nom”) in example (18), or it could simply appear in the accusative case, as is the case with the noun recenzję (“review.sg.acc”) in example (19): (18) Gdy Jan pisał list włożono do koperty. while John write.3sg.pst letter.sg.nom placed.impers into envelope “While John was writing the letter was placed in an envelope”. (19) Gdy Jan pisał recenzję włożono do koperty. while John write.3sg.pst review.sg.acc placed.impers into envelope “While John was writing the review was placed in an envelope”. Using sentences such as (17), (18) and (19) as the test material would allow us to achieve the second goal of our project, which was to check if the morphosyntactic properties of Polish nouns, specifically their case endings, and the semantic roles they tend to play in sentences affect the way sentences are understood. In the thematic grid of the -no/-to verbal forms noun phrases preceding them get assigned the role of a theme. The prediction would therefore be that items such as (18) and (19) would be less likely to be misanalysed than items such as (17). The garden path (gp) items to be used in the experiment thus appeared in three conditions, which in the tables and figures below are marked as gp1, gp2 and gp3. Their comprehension was to be tested by a question probing

10.1163/18773109-01802100 | International Review of Pragmatics (2014) 1–30

exploring garden-path sentences in polish table 1

11

The three conditions of the garden path sentences used in the experiment

The syntactic function of the pivotal noun in the main clause, Item type its theta role and its form Sample items and probing questions gp1

subject, agent, acc=nom

gp2

object, theme, acc=nom

gp3

object, theme, acc≠nom

Gdy Jan pisał list spadł z biurka. (“While John was writing the letter fell off the desk”) Czy Jan pisał list? (“Was John writing the letter?”) Gdy Jan pisał list włożono do koperty. (“While John was writing the letter was placed in an envelope”) Czy Jan pisał list? (“Was John writing the letter?”) Gdy Jan pisał recenzję włożono do koperty. (“While John was writing the review was placed in an envelope”) Czy Jan pisał recenzję? (“Was John writing the review?”)

the interpretation of the subordinate clause, such as Czy Jan pisał list? (i.e. “Was John writing a letter?”), and requiring a “no” answer. Sample garden path items in the three conditions and the probing questions are presented in Table 1. The garden path sentences were to be contrasted with non-garden path (ngp) control items in which the pivotal noun phrase could not be interpreted as the object of the key verb. The control items were thus unambiguous counterparts of gpss with the verb in the fronted clause followed by an explicit object np. Control items too appeared in three conditions corresponding to the three conditions of garden path material and were followed by a “yes/no” question probing the comprehension of the subordinate clause. Sample non-garden path items in the three conditions and the related probing questions are presented in Table 2. Since both experimental items and control items required a “no” response, the filler items to be used in the experiment had to require a ‘yes’ response. To create such fillers we used three types of sentences:

International Review of Pragmatics (2014) 1–30 | 10.1163/18773109-01802100

12 table 2

solska and rojczyk The three conditions of the non-garden path sentences used in the experiment

The syntactic function of the pivotal noun in the main clause, Item type its theta role and its form Sample items and probing questions ngp1

subject, agent, acc=nom

ngp2

object, theme, acc=nom

ngp3

object, theme, acc≠nom

Gdy Jan pisał podanie list spadł z biurka. (“While John was writing an application the letter fell off the desk”) Czy Jan pisał list? (“Was John writing the letter?”) Gdy Jan pisał podanie list włożono do koperty. (“While John was writing an application the letter was placed in an envelope”) Czy Jan pisał list? (“Was John writing the letter?”) Gdy Jan pisał podanie recenzję włożono do koperty. (“While John was writing an application the review was placed in an envelope”) Czy Jan pisał recenzję? (“Was John writing the review?”)

(i)

sentences which were identical in form with the control sentences, followed by the test question probing the comprehension of the subordinate clause (fsy) or followed by the test question probing the comprehension of the main clause (fmy), (ii) sentences created by reversing the clause order in the garden-path sentences, followed by the test question probing the comprehension of the subordinate clause (rsy) or probing the comprehension of the main clause (rmy), (iii) additional non-ambiguous filler sentences similar in structure to other filler sentences, yet containing no pivotal element (afy).

10.1163/18773109-01802100 | International Review of Pragmatics (2014) 1–30

exploring garden-path sentences in polish

13

To prevent the participants from automatically giving a positive response to all questions following the filler items we also constructed a small number of additional fillers whose probing questions required a “no” response. They included: (i) reversed clause order items followed by the test question probing the comprehension of subordinate clause (rsn), and (ii) items without a pivotal noun (afn). Sample filler items in their diverse conditions are presented in Table 3. table 3

Diverse types of filler sentences used in the experiment

Item type Sample items and probing questions fsy

fmy

rsy

rmy

afn

rsn

afy

Gdy Jan pisał recenzję list spadł z biurka. (“While John was writing the review the letter fell off the desk”) Czy Jan pisał recenzję? (“Was John writing the review?”) Gdy Jan pisał recenzję list spadł z biurka. (“While John was writing the review the letter fell off the desk”) Czy list spadł z biurka? (“Did the letter fall off the desk?”) List spadł z biurka gdy Jan pisał. (“The letter fell off the desk while John was writing”) Czy Jan pisał? (“Was John writing?”) List spadł z biurka gdy Jan pisał. (“The letter fell off the desk while John was writing”) Czy Jan pisał list? (“Was John writing the letter?”) Dziadek zasnął zanim babcia zgasiła światło. (i. e. “Grandfather fell asleep before grandmother switched off the light”) Czy dziadek zgasił światło? (i. e. “Did grandfather switch off the light?”) List spadł z biurka gdy Jan pisał. Czy Jan pisał list? (“Was John writing the letter?”) Zanim nauczyciel przyszedł studenci opuścili salę. (i. e. “Before the teacher arrived the students left the room.”) Czy studenci opuścili salę? (i. e. “Did the students leave the room?”)

International Review of Pragmatics (2014) 1–30 | 10.1163/18773109-01802100

14

solska and rojczyk

The general method we adopted was to present native speakers of Polish with a randomized set of 48 different sentences, including 12 garden path items, 12 non-garden path control items and 24 filler items, each followed by a question probing the comprehension of the sentence. An incorrect answer to the probing question could serve as evidence of a communication breakdown. Since a correct answer would mean that either the participant had recovered from the initial misanalysis or that misanalysis never took place, we decided to test the persistence of the erroneous interpretation by applying the same method as Christianson et al. (2001) in their experiments, namely by asking the participants to rate their confidence in the “yes/no” response. A set of 50 different sentence items was created and the form of each item was manipulated so as to create the different conditions which would be used as garden path, non-garden path and filler items in the experiment proper as well as in the practice session, which preceded the experiment. A sample set of a sentence item in 9 conditions is presented in table 4 below. The complete list of sentences used in the experiment is to be found in the Appendix. table 4

Item type

Diverse types of sentences used in the experiment

Sample items and probing questions

gp

Gdy Jan pisał list spadł z biurka. (“While John was writing the letter fell off the desk”) Czy Jan pisał list? (“Was John writing the letter?”) gp Gdy Jan pisał list włożono do koperty. (“While John was writing the letter was placed in an envelope”) Czy Jan pisał list? (“Was John writing the letter?”) gp Gdy Jan pisał recenzję włożono do koperty. (“While John was writing the review was placed in an envelope”) Czy Jan pisał recenzję? (“Was John writing the review?”) ngp or Filler Gdy Jan pisał podanie list spadł z biurka. (“While John was writing an application the letter fell off the desk”) Czy Jan pisał podanie? (“Was John writing an application?”)

10.1163/18773109-01802100 | International Review of Pragmatics (2014) 1–30

exploring garden-path sentences in polish

Item type

15

Sample items and probing questions

ngp or Filler Gdy Jan pisał podanie list włożono do koperty. (“While John was writing an application the letter was placed in an envelope”) Czy Jan pisał podanie? (“Was John writing an application?”) ngp or Filler Gdy Jan pisał podanie recenzję przeczytano. (“While John was writing an application the review was read”) Czy Jan pisał podanie? (“Was John writing an application?”) Filler List spadł z biurka gdy Jan pisał. (“The letter fell of the desk while John was writing”) Czy Jan pisał list? (“Was John writing the letter?”) Filler List włożono do koperty gdy Jan pisał. (“The letter was placed in an envelope while John was writing”) Czy Jan pisał list? (“Was John writing the letter?”) Filler Recenzję włożono do koperty gdy Jan pisał. (“The review was placed in an envelope while John was writing”) Czy Jan pisał recenzję? (“Was John writing the review?”)

4.2 Participants A total of 21 participants took part in the study (17 females and 4 males). They were all undergraduate students at the University of Silesia, Poland, ranging in age from 19 to 21 years (m = 19.8, sd = 0.09). All participants were native speakers of Polish, born and raised in Poland. They were all naïve with respect to the object of the study. 4.3 Apparatus The stimuli were displayed on a hp 17-inch monitor with the resolution set at 1280×768. The experiment was run in E-Prime software scripted to register the accuracy of the responses, the reaction times (rts) of the participants and the confidence rating of their responses. Participants used the Serial Response Box (Psychology Software Tools) to collect their “yes/no” responses and reaction times (rts). Confidence ratings were obtained from mouse clicks on a displayed scale. International Review of Pragmatics (2014) 1–30 | 10.1163/18773109-01802100

16

solska and rojczyk

4.4 Procedure The experiment took place in the Acoustic Laboratory at the Institute of English, University of Silesia. Upon arriving for the experimental session, participants were given an oral description of the experiment along with a set of instructions. They were informed that they would see sentences presented on a computer screen, would have to answer a “yes/no” question about each sentence and then rate their confidence on a scale. Participants were seated at a computer monitor with the response box placed in front of them. Each participant was instructed that the left button on the response box was labeled as a “yes” response and the right button as a “no” response. They were encouraged to keep their index fingers touching buttons while reading each prompt sentence to avoid distorted reaction times due to inaccurate manual control. The experiment started with a welcome screen briefly summarizing the instructions. After the participant pressed a button they were informed by the next screen that they would go through a button practice session. It contained eight prompt sentences and eight related questions together with a confidence rating. The practice sentences were not included in the experimental session. Only accuracy for “yes/no” responses was registered. The participants proceeded to the experimental session after achieving 80 % accuracy, which had been assumed to reflect sufficient understanding of the experimental procedure. After completing the practice session, the screen was displayed informing the participants that the experimental session was to begin. Another screen encouraged the participants to set their index fingers on the “yes/no” response buttons. Subsequently, a blank screen appeared for 1000 ms and the prompt was displayed. The participants read the prompt sentence without any time limit and when ready pressed one of the response buttons. No rts were collected at this stage. Another screen appeared with a fixation point “+” in the centre for 1000 ms. Finally, the “yes/no” question appeared and remained on the screen until the participant pressed one button to respond “Yes” or another button to respond “No”. rts were collected as the time interval between the stimulus onset and the registered response. After the response, the words Czy jesteś pewien swojej odpowiedzi? (i.e. “Are you sure your response was correct?”) appeared on the screen. The participants used a mouse click to select one of the four boxes labeled “Zdecydowanie tak” (“very sure”), “Raczej tak” (“quite sure”), “Nie bardzo” (“not really sure”) and “Ani trochę ” (“not at all sure”). After the confidence response, the finger setting message appeared on the screen to begin the next trial.

10.1163/18773109-01802100 | International Review of Pragmatics (2014) 1–30

exploring garden-path sentences in polish

figure 3

17

Accuracy for ngp sentences and gp sentences

The participants responded to 48 prompt sentences, displayed one at a time. Each session took approximately 20 minutes. 4.5 Analysis and Results Before statistics for rts and confidence ratings were calculated, the accuracy was determined as a proportion of correct responses to the total number of responses for all participants. As shown in Figure 3, the gp sentences were correctly evaluated 92% of the time compared to 91 % for non-gp sentences. Within the gp group, the accuracy was 85% for gp1, 95 % for gp2 and 96 % for gp3. As shown in Figure 4, one-way repeated measures anova for the three types of gp sentences revealed that accuracy was significantly affected by the gp type [f(2, 40)=6.43, p