cutting bock 1997

8/3/2019 Cutting Bock 1997

1/15

1 Memory & Cognition1997. 25 (I).57-71

That's the way the cookie bounces:Syntactic and semantic components ofexperimentally elicited idiom blendsJ. COOPER CUTTING and KATHRYN BOC KUniversity of Illinois at Urbana-Champaign, Cham paign, Illinois

Idioms are sometimes viewed as unitized phrases with interpretations th at ar e independent of theliteral meanings of their individual words. In three experiments, the nature of idiom representationwas explored with a spe ech-error elicitation task. In the task, speakers briefly viewed paired idioms.After a short delay they were p robed to prod uce one of the two idioms, and their production latenciesand blend erro rs were assessed . The first experiment showed greater interference between idioms withthe sam e syntactic structure, demonstrating that idiom representations contain syntactic information.The second experiment indicated that the literal meaning of an idiom is active during production. Thesesyntactic and literal-semantic effects on idiom errors argu e against a representation of idioms as no n-compon ential lexicalized phras es. In the final exper iment, no differences wer e found between de-composab le and nondecom posable idioms, suggesting that the lexical representation of these twotypes of idioms is the sam e.

Recently, a friend commented that "the road to Chi-cago is straight as a pancake." This w as a blend of two id-ioms, straight as an arrow andflat a s apan cak e. Althoughslips of the tongue are often studied in order to provideinsight into language-production processes, phrase andidiom blends have received little attention (Fay, 1982;Garrett, 1 980). The purpose of the experiments presentedhere was to examine such e rrors for clue s to the nature ofidiom representation.Idioms are often regarded as phrases with interpreta-tions that are not directly related to the literal meaningsof their individual words (Dik, 989; Fraser, 1970; Katz& Postal, 1963; Weinreich, 1969; Wood, 1986).Challenges

I to this view have come from psycholinguistic researchon how idioms are understood. This research has focusedmainly on two related issues: (1) how idioms are mentally! represented, and (2) whether their literal meaning plays1 any role in their use.The standard pragmatic m odel of idiom comp rehen-i sion (Bobrow & Bell, 1973; Katz & Postal, 1963) pro-/ poses that idioms are stored as whole chunks, roughlyI1 The experime nts reported here are part of the first author's master'sthesis presented to the Department of Psychology at the University ofIllinois at Urbana-Champaign. The research was supported in part bygrants from the National Science Foundation (BN S 90-0961 1, SBR 94-1 1627) and the National Institutes of H ealth (R01 HD 21011). Portionsof the work were presented at the 66th Annual Meeting of the Mid-western Psychological Association in Chicago, in 1994. The authorsthank Gary Dell, Gregory Murphy, Rose Zacks, and Thomas Carr, fortheir comments and suggestions, and Brian Kleiner, for his assistancein administering the experimental tests and in scoring transcriptions.Requests for reprints and o ther correspondence may be sent to J.C. Cut-ting, Dep artment of Psychology, University of Illinois, 603 E. DanielSt., Champaign, 1L 61 820 (e -mail: [email protected]).

equivalent to a single word, in an idiom lexicon that isseparate from the general lexicon. The model further pro-poses that the literal me aning o f a phrase is first derivedand then check ed against its context. Ifth is meaning failsto fit the context, then the idiom lexicon is consulted forretrieval of the figurative meaning of the idiom. This modelrests on the assum ption that literal meaning is always de-termined and that this determ ination precedes figurativeinterpretation,Challenging the literal-first proposal, Swinney and C ut-ler (1 979; see a lso Gib bs, 1980, 19 86) showed that id-ioms were classified a s acceptable English phrases fasterthan matched co ntrol strings. They proposed that an idiomwas represented as a single lexical unit within the generallexicon, rather than in a sp ecial lexicon (see also Fraser.1970; Katz, 19 73). Retrieval and storage of an idiom pro-ceeds in the sam e way as any other word, with access to lit-eral and figurative meanings occurring simultaneously.A strong interpretation of Swinney and Cutler's (1979 )unitization hy pothesis sug gests that the idiomatic phraseis just a large word-like unit. The internal componen ts haveno identity as individual lexical items, blurring or losinginformation about gram matical class, syntactic structure.and sem antics. This interpretation is que stionable becauseidiom representations must retain som e information aboutthe internal com pone nts (Katz, 1973): They need phono-logical information in order to app ropriately assign stressand syntactic information about the components, be-cause m any idioms are syntactically flexible. For example.lay down the law can be syn tactically manipulated w ith-out losing its idiom atic meaning, as in "after the law hadbeen laid down by Joh n, there was no more trouble."A weaker interpretation of the unitization hypothesisretains structural information in the representation. Fraser

1 57 Copyright 1997 Psychonom ic Society, Inc.


2/15

58 CUTTING AND BOCK

(1970 ) proposed that a phrasal idiom was a single lexi-cal item that contained multiple complex symb ols (syn-tactic features and phonemic representations) but a sin-gle semantic marker. For example, the idiom "hit the sack"might be abstractly represented as ([+V] hit; [+DET ] the;[+N] sack; {go to bed}), but without semantic informa-tion associated to the individual parts of the idioms.Instead, there is a single semantic interpretation for theidiom as a whole. This solves the problem of syntacticflexibility, but raises ano ther. Idiom s vary in the extent towhich they can undergo lexical manipulation witho ut dis-rupting their figurative meaning (Gibbs, Nayak, B olton,& Keppel, 1989). Wasow, Sag, and Nunberg (1983)showed that the individual parts o f some idioms can be in-ternally modified by the insertion of adjectives or quan-tifiers as in "Mary really touched a couple of nerves."Likewise, one can em phasize the parts of idioms throughtopicalization, as in "The strings that John was able to pullseem ed to be the right ones for getting the job." T hus, as-pects of an idiom's overall meaning can be altered bymodifying its individual components, implying that themeanings of these individual parts contribute me aning tothe whole.To explain these properties of idioms, Nunberg (1978)proposed the idiom decomposition hypothesis. This hy-pothesis suggests that idioms are not single lexical units,either lexically or semantically, but partially analyzablephrases. Their productive behavior is determined by theassumptions that a speaker makes about how the mean-ings of the parts contribute to the figurative meaning ofthe whole (see also Cacciari & Glucksberg, 1991;Gibbs& Nayak, 1989; van der Linden, 1989; Wasow et al., 1983;Wood, 1986). The theory argues that idioms exist on acontinuum o f decomposability. T he idiom pop the ques-tion is considered decomposable because its componentparts refer, in so me way, to their id ioma tic referents. Poprefers to "a sudden noise" (that of speaking) and the ques-tion refers to "a proposal of marriage." Such idioms arepredic ted to be syntactically flexible. In contrast, for thenondecomposable idiom kick the bucket , it is hard toimagine what the relationship between the componentsand their idiomatic meaning is. This idiom should be syn-tactically inflexible ("frozen"; see Fraser, 1970). Sup port-ing this hypothesis, Gibbs and Nayak (1989) found thedecomposab ility of idioms to be highly correlated withtheir syntactic flexibility.Although m ost psycholinguistic research on idiom s hasaddressed how they are understood (Cacciari & Glucks-berg, 199 1; Gibbs & Nayak, 1989), the issues of repre-sentation and literal meaning use also arise with respectto the production of idioms. A production hypothesis de-rived from either the standard pragmatic model or theunitization hypotheses might suggest that idioms arerepresented and accessed as whole units. Fu rthermore, astrict division between an idiom's figurative mean ing andthe literal meaning of its individual parts would predictthat, in speaking, one should be able to access an idiomwithout interference from the literal meanings o f its com -ponent w ords. This is because speake rs have in mind the

idea that is to be conveyed prior to producing an idiom,and that idea presumably ma ps onto the figurative mean-ing. For example, a speaker who intend s to convey theidea that somebod y died suddenly can say "he kicked thebucket" without accessing the literal interpretation "heknocked the pail over with his foot."The standard pragmatic hypothesis for production suf-fers from the same problem that its comprehension pre-decessors did. If idioms were single units, lexical and syn-tactic variations could not be produced . As an alternative,Cacciari and Glucksberg (199 1) proposed that idiom s wereretrieved as phras es in which the internal com pone nts re-tained individual syntactic and seman tic annotations. Theyfurthe r argue d that both the literal and figurative m eaningsplayed a role in using idioms generatively in discourse.The literal meaning of an idiom's componen ts mediatesthe production of id iomatic variants, both lexical and syn-tactic (McG lone, Glucksberg, & Cacciari, 1994). For ex-ample, "once the ice was shattered, the party becamemuch m ore enjoyable," retains the basic idiomatic meaningof break the ice. but with a nuance from the literal mean-ing of shattered.The present experiments were designed to explorethese issues in idiom representation by examining sp eecherrors that accidentally combine two different idioms, oridiom blends. Blends in general are viewed as the resultof two competing speech plans which interfere with oneanother' (Baars, 1980a; Fay, 1982; Fromkin, 1 971; Har-ley, 1984; Levelt, 19 89; Motley, Baars, & Camden, 1983a,1983b). The term blend thus refers to the replacement ofan element related to one speech plan by an elem ent re-lated to another speech plan, resulting in an utterance thatdoes not exactly match either one.The most familiar blends involve individual words:(1) The competition is a little stougher [stifferltougher](Fromkin, 1973), (2 ) Irvine is quite clear [close/near](Fromkin, 1973), and (3) . . and would like to enlicit[enlist/elicit] your support (Garrett, 1980). Examp les 1and 2 illustrate the strong semantic relationships typicalof word blends (Fromkin, 197 1 Garrett, 1980; Shattuck-Hufnagel, 1979) and Example 3 illustrates a phonologi-cal relationship (Dell & Reich, 1981). The errors usuallycombine words from the same syntactic class (Butter-worth, 1982; Harley, 1984; Levelt, 1989) and respect syl-lable boundaries (MacKay, 1972).If idioms are sing le lexical items, as the strong versionof the unitization hypothesis suggests, then one wouldexpect them to demonstrate only characteristics similarto those just described for word blends. Idioms o f simi-lar figurative meaning should b e most likely to blend to-gether, but these erro rs should be independent of the in-ternal syntax of the idiom. On the other hand, the weakunitization and decom posability hypotheses suggest thatthe representation of an idiom includes syntactic infor-mation. In this case, one w ould expect that sy ntactic infor-mation should influence the production of errors suchthat idioms with the same syntactic structure should in-terfere with one another more than those with differentstructures, resulting in m ore idiom blends. Fu rthermore,


3/15

THE PRODUCTION OF IDIOM BLENDS 59

substituted words should tend to be of the sa me gram mat-ical class as the idiomatic compon ents that they replac e.If idioms are partially analyzable phrases, then blendsbetween them should resemble phrase blends more thanword blends. Although relatively little is known aboutphrase blends, their occurrence is well attested. Butter-worth ( 1982) attributed them to interference between twoal ternative but complete formulat ions of an intendedthought. Th e result may be a syntactic blend (e.g., "1 couldnever did fin d it" is a blend of "I could never find it" an d"I never did find it"; Stemberger, 1 982 ) or a blend of alter-native plans with related meanings but widely differentsurface structures (e.g., "I miss being out of touch withacademia" is a blend of "I miss being in touch with aca-demia" and "I am out of touch with academia").Phrase blends are a diverse lot. A small corpus com piledby Fay (1 982 ) is made up o f 15% substitution blends (c om-bining w ords from compe ting phrases; e.g., "It's spentme a year . " from "It's taken me a year" and "I spent ayear"), 21% splice blends (tackin g part of one phraseonto a competitor; e.g., "She loo ks it that type" from "Shelooks it" and "She looks that type"), and 59% blends thatexhibited both of these characteristics ("I've never hadsuch worse hands" from "I've never had such bad hands"and "I've never had worse hands"). The remaining 5%were complex errors.There is anecdo tal evidence that idioms, like phrases ,blend. In Stemberger's (1 985) corpu s of speec h error s, id-ioms appear in blends in which an entire word from oneidiom replaced a word in another (69 instances) or com-ponent words them selves blended (9 instances). Notably,the word replacements som etimes involved words thatwere similar in literal meaning. Stem berger observed thatthe predominance o f word replacements su ggests that theblends occu rred mostly at a level where words were rep -resented as wholes rather than individual speech sounds.Beca use idiom blends occur too rarely in spontaneousspeech to reveal much about how idioms are representedand processed in production, the present research em-ployed a control led error-el ici tat ion procedure. Ourmethod, like other m ethods for inducing unintentionalspeech er rors, involved presenting spe akers with two al-ternative production targets. The speak ers are required toproduce only on e while under time pressure (Baars, 1980a,1980b). These techniques have been applied most ex-tensively to elicit sound ex chang es, but Motley (1 985)sketched a related technique for eliciting phrase errors.Two sentences were presented to speakers for 10 sec,tagged by a sing le letter: T: Do I have to put on m y seatbelt? and M: Do I have to put my seat belt on? After a5-sec delay, one of the letters (T or M ) was presented a sa cue to produce the corresponding sentence. The taskreportedly elicited blends like "Do I have to put on myseat belt on?" (also see Lapointe & Dell, 1989).In the three experiments below, speake rs were brieflyshown two idioms. After a short pause, they were cued toproduce one or the other as quickly as they we re able. Inthe first experiment, the syntactic and sem antic properties

of the competing idioms were m anipulated to see whetheridioms with the sam e syntactic frame were more or lesslikely to interfere with o ne another, and whether idiomswith the sam e figurative meaning were more or less likelyto interfere with on e another. The second experime nt usedpairs of phrases, on e an idiom and the oth er a phrase witha mean ing similar to or d ifferent from the idiom's literalmeaning. If literal mean ing plays a role in idiom produc -tion, literally similar phrases should blend with idiomsmore often than literally dissimilar phrases. The final ex-periment exam ined whether paired decomposable idiomswere any more likely than paired nondecom posable id-ioms to interfere with one another. If decom posability isimportant for the production of idioms, decomposableidioms should be easier to modify and thus more sus-ceptible to production errors.The goal o f the three experiments was to explore someof the representational factors that constrain idiom er-rors, particularly their meanin gs (both literal and figura-tive) and their syn tax. The focu s was on idiom blends, withthe assumption tha t constraints on idiom errors might re-flect fundamental features o f the idiom representation.In general, if idioms are lexicalized phrases without in-ternal syntactic and sem antic structure, the structure andliteral meaning of competing idioms should not affectproduction. However, if an idiom's representation is ar-ticulated in terms of both structure and meaning, idiomblends should respect the structures and meanings of thecompetitors.

EXPERIMENT 1In this experiment, speakers were induced to create

idiom errors with a competing-plans methodology. Thespeakers silently read two idioms from a com puter screen,and shortly thereafter they were cued to produce one orthe other idiom. P aired idioms had the same or differentidiomatic meaning and the sa me or different syntacticform. The question was whether these two factors con-strain idiomatic speech errors. The decomposability hy-pothesis an d both unitization hypotheses predict that idiompairs with similar figurative meanings should producemore inte rference than those w ith different meanings, re-sulting in longer nam ing latencies and more blends. Thehypotheses differ in their predictions about the effect ofsyntactic structu re. If idioms are completely lexicalizedphrases, a s the strong unitization view w ould have it, thesyntax of an idiom shou ld not play a significant role in theproduction o f blends. On the other hand, if idiomatic rep-resentations encode information about internal syntax(as decom posability and weak unitization sugges t), thenidiom pairs with identical syntactic forms should pro-duce more interference (more blends and longer produc-tion latencies) than those with different syntactic forms.MethodParticipants.Ninety-six Michigan State University students par-ticipated in the main experiment for extra credit in introductory psy-


4/15

60 CUTTING AND BOCK

chology classes or for payment. Five were replaced because theyreported that they w ere unable to read both of the idioms in the timeallowed. An additional 80 students served in idiom norming stud-ies. None of them took part in the main experiment. All of the stu-dents were native speakers of English.Ma ter ial s. The idioms were assembled from a list of 129 Amer-ican idioms from Mak kai (1987) and 129 additional literal phrases.These materials were no rmed prior to the experiment in order to as-sess their idiomatic properties. In the first norming study, 40 stu-dents were asked to rate the phrases on a 5-point idiomaticity scale(1 = not an idiom, 5 = very idiomatic) .The idioms selected for theexperiment had a median score of 4.2 and a range from 2.9 to 5 .0.In comparison, the literal phrases yielded a median score of 2.0 anda range from 1.1 to 4.0 (only six had mean ratings higher than 3 .0).In the second n orming study, another 40 participants received alist of paired idioms and were asked to rate on a 5-point scale whetherthe two phrases had the same meaning (5 ) or different meanings (1).The median m eaning-similarity ratings for the items used in the ex-periment are presented in Table 1.Thirty-six experimental items were formed, each with three ver-sions. Each version con tained a pair of idioms. One idiom, thestan-dard, appeare d in all pairs. In one experimenta l condition, the stan-dard idiom and its pairmate had similar figurative meanings and thesame syntactic structures (e.g., if the standard idiom had the s truc-ture verb-determiner-noun, so did the paired idiom). In a secondcondition, the standard idiom and its pairmate had the same syn-tactic form but different figurative meanings. In the final cond ition,the standard idiom and its pairmate had different syntactic forms(e.g., if the standard idiomwas verb-determiner-noun, then the pairedidiom might have the structure preposition-determiner-noun) anddifferent figurative meanings. We will call these respective condi-tions same meaning-same syntax, different meaning-same syntax,and different meaning-different syntax . Table 1 gives an exam ple ofan item in each of the three condition s and the median idiomaticityrating for each type of idiom. In addition to the experimental idioms,72 other idiom pairs were used as practice and filler items. All ofthese pairs differed in syntactic form and figu rative meaning. All ofthe idioms in a pair had the same number of words. The median

number of words in the idioms across all of the experimental itemswas three. The Appendix lists the complete idiom set.Pro ced ure . Speakers were tested individually, and each sessionwas recorded on aud io tape. The stimu li were presented visually onthe screen of a Macintosh computer with an EM M-2048a Univer-sal Microcon troller (Metaresearch Inc.) for millisecond timing. Re-sponse s were spoken into a microp hone that was attached to a voicekey (Gerbrands Corporation, Model GI 34 IT), which in turn signaledthe computer that a response had been made.Figure 1 depicts an exam ple of a typical trial. The idioms in eachpair were presented o ne above the other, left justified, in the centerof a compu ter screen. A single experimental trial proceeded a s fol-lows. The speak er viewed a pair for a period o f time that varied ac-cording to the length of the idioms. The duration was determined byadding 40 msec for each letter in a content word and 20 msec for

Table 1Median Idiomaticity and Meanin g Similarity Ratingsfo r Exoer imen t 1Mater ia lsMeaningIdiom SimilarityCondition Example Idiom aticity to Standard

Standard idiom shoot the breeze 4.3Same meaning-same syntax chew the fat 4.1 4.3Different meaning-same syntax raise the roof 4.2 1.7Different meaning-differen t syntax nip and tuck 4.2 1. 9

1 I 150 msec Speaker's responsemeet your maker blank screen

2000 msecprobe

"kick the bucket"blank screen1000msec

blank screen2000 msec

Figure 1.A representative experimental trial.

each letter in a function word to a base duration of 250 msec. Thisprocedure w as followed in order to allow speakers to read each idiomjust o nce. The idio m pair was followed by a blank screen for 2 sec.Then a probe number (a 1 or 2) appeared in the center of the screen.The speaker's task was to recall aloud as quickly as possible one orthe other idiom. If the probe number was 1, the speaker was to re-call the top idiom; if the probe number was 2, the speaker recalledthe bottom idiom. The probe number remained on the computerscreen until a response had be en made. At the voice-key trigger, thecomputer recorded the production latency of the response. If thespeake r did not respond within 2 sec, the probe number disappearedand was replaced with a beep and a message telling the speaker torespond faster. One second after the voice-key trigger, a questionappeared on the center of the screen asking, "Did you respond cor-rectly'?'' Th e speaker's task w as to press either a"Y" if they believedthey had answered with the correct idiom and had said it correctly,or "N" for "No, I think that I made a mistake."2 T he next trial began2 sec after this response.The filler trials had the same stru cture as the experimental trials,but the duration s of the blank sc reens differed. The blank sc reen forhalf of the fillers had a 1-sec duration; for the other half, it had a3-sec duration. This was done to discourage the speakers from antic-ipating the onset of the probe number. An initial block of trials servedas practice for the speakers, although they were not explicitly toldthat it was practice. There w ere 16 filler items in this block. Theseitems were probed after delays representative of the items in theother blocks, five of the delays were 1 sec, five were3 sec, and sixwere 2 sec.Following the practice block, the remainin g 56 filler pairs and 36experimental pairs were presented in blocks of 23 items each. Theorder of items was random , with the constraint that at least one filleroccur between any two experimental items. A single random orderwas used for all speakers, and all speakers received the same filleritems.The spea kers were given a brief explanatio n of the procedure fora trial, along with examples that consisted of two complete trials dis-played on a sequence of index cards. They were also instructed torespond as quickly as possible, without laughing, coughing, or mak-ing any unnecessary noises that would set off the voice key. Theywere told that they would be recorded on audio tape.Design . Every speaker received o ne version of each experimen-tal item, an equal number (1 2) in each of the three conditions. Withineach conditio n, the order of idioms in each idiom pair and the probesfor each pair were completely counterbalanced across speakers,yielding 12 distinct presentation sets. Every set was presented to 8speakers so that 24 speakers received each version of every item(summing over the counterbalanced variations).Scoring. The audio tapes were transcribed and the responsesscored. Two raters independently scored each transcrip tion. The re-


5/15

sponses were classified into six m utually exclusive categories: (1) cor-rect responses, (2) idiom blends, (3) intra-idiom er rors,(4) intertrialerrors, (5) complex errors, and (6) miscellaneous responses. Cor-rect responses occurred when speakers correctly produced theprobed idiom . Idiom blends were errors that combined elements fromthe two idioms in a single trial pair. Intra-idiom erro rs resulted frominterference within a single idiom, while intertrial errors resultedfrom identifiable interference from any of the three previous trials.Complex errors consisted of intra-idiom or intertrial errors that in-volved more than one type of error (e.g., word addition and a sub-stitution), errors that involved word blending or sound exchanges(e.g., shoot an d throw produced show; an d turn the tide an d tip thescales produced turn the tides), and responses with an alternativeidiom that shared a content word with the intended idiom (e.g.,zipyour lip presented but zip it up produced, when "it up" was not apart of the paired idiom or any of the preceding idioms). M iscella-neous responses were scored when the speaker produced the wro ngidiom of a pair, an idiom from a previous trial, an incom plete idiomcontaining the word "something" (e.g., "kick the something"), o r noresponse. Table 2 presents examples of each error type.The three simple error types ( 2 4 above) were further subcate-gorized into substitutions, additions, deletions, or movements. Sub-stitutions and additions could be identified in all three error cate-gories, but deletion and movement errors were relevant only in theintra-idiom category. Sub stitution s were errors in which a word wasinserted in place of another word in the produced idiom (e.g.,taketo the woods produced as head to the woods).Additions were errorsin which additional words were inserted without replacing any words(e.g., blow your top, produced as blow your top off ). Deletions in-volved the omission of a word (e.g .,hold a candle to, produced ashold a candle ).Movements were errors in which the word orderwaschanged (e.g.,put down your foot, produced asput your foot do wn).The application of these scoring criteria resulted in 81.3% cor-rects, 1.7% idiom blends, 4.9% intra-idiom errors, 1.3% intertrialerrors, 2.7% complex errors, and 8 .0% miscellaneous errors. Therewas 98.4% agreement between the two raters over all of the re-sponses. The disagreements were discussed, re-evaluated, and thenrescored according to the same set of criteria.Separate one-wa y analyses of variance (ANOVAs) were performedon the data in each category. The error terms from these analyseswere used to calculate 95% confidence intervals in order to performTukey tests. The first comparison contrasted the same-meaning-same-syntax condition with the different-meaning-same-syntaxcondition in order to assess the contribution of the figurative mean-ing. The second comparison contrasted the different-meaning-same-syntax with the different-meaning-different-syntax condition toassess the contribution o f the syntactic structure. These co ntrasts were

Table 2Examples of Error Categories for the Idiom Pair"Kick the Bucket" and "Meet Your Maker"Error Category Example ErrorsIdiom blendAddition kick the bucket makerSubstitution kick the makerIntra-idiomAddition kick the bucket overDeletion kick bucketMovement bucket the kickSubs titution kick the rockIntertrialAddition kick the bucket arou nd*Substitution kick the millsto ne*Complex kick the big rockMiscellaneou s kick the some thine

*The idiom mills tone aroundyour neck occurred in one of the prev iousthree trials

THE PRODUCTION O F IDIOM BLENDS 61

performed over speakers and items. All effects at or beyond the .05level were treated as significant, and those between .05 and .I 0 asmarginally significant in this and both subsequent expe riments.Results

Production latencies.Times under 100 msec were in-terpreted as false voice-key triggers and excluded fromthe analyses (0 .3% of the total). A further 3.8% of thetrials were m issing du e to failure to meet the 2-sec dead -line. The mean production latencies for the remainingcorrect-response trials were 8 18 msec for idiom pairs inthe same-meaning-sam e-syntax condition, 785 msec foridioms in the different-meaning-same-syn tax cond ition,and 795 msec for idioms in the different-meaning-different-syntax condition. These times are show n in Fig-ure 2. A one-way ANOVA yielded an effect of cond itionthat was significan t by speakers [Fl(2, 95 ) = 5.54, MSe =4,7981 but not by items [F2(2,35) = 2.70, MS, = 3,6951.Tukey tests were carried out on the con dition differenceswith confidence intervals of 23 msec by speakers and34 msec by items. For idiom pairs with the sam e syntax,those with the sam e meaning were slower (by 33 msec)than the pairs w ith different meanings, significantly so byspeakers and marginally by items. For idioms with dif-ferent mea nings , the 1 0-msec difference between the pro-duction latencies for idiom pairs with same versus dif-ferent synta x was not sign ificant.

Error analyses.Table 3 shows the number of responsesin each scoring category. The pattern of idiom blends isshown in Figure 2. Recall that these errors are a conse-quence of the su bstitution or addition of a word from oneidiom for a word in the other idiom of the pair. Becauseonly on e addition occurred, we will focus on the substitu-tions. M ost erro rs appeared in the same-meaning-same-syntax con dition (29), followed by the different-meaning-same-syntax condition (2 0) and the different-meaning-different-syntax condition (9). ANOVAs dem onstrated asignificant effect of condition [F1(2 ,95)= 6.10, MS. =0.17; Fy(2,35) = 3.34, MS, = 0.841. Tukey tests wereperformed on the condition differences with 95% c onfi-dence interva ls of 1 3.4 for speakers and 18.4 for items. Thedifference of 20 errors between the same-m eaning-same-syntax condit ion and the different-meaning-different-syntax condition was significant by both speakers anditems. The 1 1-error difference attributable to syntacticstructure (contrasting the different-meaning-same-syntaxand different-meaning-different-syntax cond itions) wasmarginal by speakers (90% confidence interval of 11.0 byspeak ers and 15.8 by item s). There was no reliable effectof meanin g in the comp arison of the same-meaning-same-syntax and different-meaning-same-syntax cond itions.An ANOVA for the intra-idiom errors, collapsed acrossall of the su bcatego ries, yielded a marginal effect of con-dit ion [F,(2,95) = 2.84, MS. = 0.49; F,(2,35) = 2.49,Me= 1.491, but the Tukey tests revealed no effect of syn -tax or meaning. For the intra-idiom substitution errors,the ANOVA yielded a signif icant effect of condit ion[ F , ( 2 .95)= 4.94, MS, = 0.41; F2(2,35)= 4.06, MS, =1.341. The co nfid enc e intervals for the Tukey tests were


6/15

62 CUTTING A N D BOCK

Figure 2. Correct response latencies and idiom blends in Ex-periment 1. The three conditions are same meaning-same syn-tax (SWS S) , different meaning-same syntax (DMISS), and dif-ferent meaning-different syntax (DMIDS). The bars representthe blend errors, and the individual box points represent thenaming times. The error bars reflect confidence intervals for theTukey tests.

20.2 by speakers and 23.4 by items. The numb er of sub-stitution errors in the different-meaning-different-syntaxcondition (56) was significantly larger than the numberin the different-meaning-same-syntax condition (30). Thetest of the meaning difference was significant by speakersand marginal by items, with more errors whe n the pairedidioms had the same meaning (52 vs. 30). There wererelatively few err ors in the intra-idiom add ition, deletion,and movement subcategories, so we will not report analy-ses for these.For the correct responses, the ANOVA yielded no sig-nificant differences between the conditions [F,(2,95) =2.1 l . .W e = 1.63; F2(2 ,35) = 1.21, MS, = 6.751. Tukeytests showed no effect of meaning or syntax (confidenceintervals were 41 .3 by speakers an d 52.6 by items). Therewere no differences among the errors in the intertrial ormiscellaneous categories (F s < 1) .Within the complex error category, most of the errorsoccurred in the different-meaning-different-syntax condi-tion ( 42) , followed by the different-meaning-same-syntaxcondition (34) and the fewest in the same-meaning-same-syntax condition (18). The ANOVA yielded a significanteffect ofcondition [F,(2,95 ) = 4.17, MS, = 0.37; F2(2,35 )= 3.19, MS. = 1.301, but the Tukey tests disclosed noeffect attributable to m eaning or syntax (confidence in-tervals of 20.2 by speakers and 23.0 by items).

Qualitative analysis of the errors. The qual i tat ivepatterns of error s were similar across all three experiments,so we will only report them in detail for this experiment.For the sake o f clarity, we will restrict the discu ssion tocategories in which at least five errors occurred. Thesewere the idiom blends, intra-idiom, and complex errors.A breakdown of the blending errors showed that 93%(5 4 o f 58) of the subst i tuted words were of the samegrammatical class as the word that they replaced. Ofthes e,20 involved prepositions, 19 involved the substitutionof determiners, and 15 involved nouns or verbs. In thedifferent-syntax condition, 6 of 9 of the substitution er-

rors involved words o f the same grammatical class and 4of these 6 were words fro m different positions within theidioms. Thi s helps to discount a positional interpretationof the gramma tical class effect in the same-syn tax condi-tion. The rema ining 3 erro rs in this condition involved sub-stitutions of words that were not of the same grammaticalclass, and all of these involved words in the same posi-tion. Along with constraints observed on spontaneousblends, this suggests that gramm atical class constrainsidiomatic blending to som e degree, with position perhapsplaying a role in the experimen tal task as well.Like the idiom blends, most of the intra-idiom substi-tution errors involved words of the sam e gramma tical class(88.4 % collapsed across the three conditions). U nlike theinteridiom substitution erro rs, most of the intra-idiom er-rors involved nouns and verbs (89.1% collapsed across thethree conditions). O f these, 24.4% involved inflectionalchanges (e.g., wrap to wrapped), 56.9% involved sub-stitution of similar sounding words (not including the in-flectional changes), 8.9% involved substitution of wordsrelated in l i teral meaning (3 o f these were also in thesimilar-sound category), and 12.2% involved the substi-tution o f some apparently unrelated word. The movementerro rs generally involved moving a preposition or a par-ticle to the end of the phrase (7 of 9). Most o f the deletionerrors involved leaving out a function word (1 1 of 14).The complex errors ( n = 94) consisted of multiple intra-idiom substitution errors (52) or combinations o f substi-tutions with additions (14), deletions (lo) , or movementerrors (2). These errors were distributed fairly evenlyacross the three conditions. T he remaining err ors in thisclass stem med from the prod uction of an alternative idiomthat shared part of the target idiom ( 8 errors, suc h as thereplacement o f zip it up w ith zip you r lip) or blends be-tween words from each idiom (one in each condition).

Table 3Number of Correct Responses and Distribution of Errorsin Exneriment 1Condition

Scoring Category SMISS DMISS DMIDSCorrect 95 1 942 916Miscellaneous 80 96 101Complex 18 34 42

Idiom BlendsAddition 0 [ol 0 [ol 1 [OlSubstitution 29 [ I 01 20 [21 9 [41

Intra-Idiom Err orsAddition 3 [ I ] 4 [31 0 [01Deletion 0 [o] 6 [21 8 [I ]Movement 5 [o] 3 [o l 1 [ I ]Substitution 52 1441 30 [28 ] 56 [5 I]

Intertrial Errors

Note-SMISS = same meaning-same syntax , DMJSS = differen tmeaning-same syntax, DMI DS = different meaning-different syntaxValues in square brackets ar e the number o f errors involving contentwords.


7/15

DiscussionTh e results of Experiment 1 suggest that the produc-tion of an idiom atic blend is somewha t sensitive to the in-ternal syntactic structure and to the figurative meaningof the idioms involved. Idioms from pairs with similar fig-urative meanings were more slowly produced than id-ioms with different figurative m eanings, and idioms withthe same syntactic structure were more likely to blendtogether than those with different syntactic structu res.To examine whether the figurative meaning effect couldbe traced to literal similarity, the idiom pairs from thesame-meaning-same-syntax condition were divided intotwo group s, (1) those that w ere intuitively diffe rent in lit-eral meaning ( n = 27) , an d ( 2 ) hose that were similar( n = 9). The different-literal idioms took 22 msec longerin the same-meaning-same-syntax condition than in thedifferent-meaning-same-syntax condition and accoun tedfor 56 % of the idiom blends. The same-literal idiom s took42 msec longer. So, although it appears that literal simi-larity enhanced the meaning effect (something that Ex-periment 2 examines in greater detail), figurative simi-larity played a part too.The pattern of idiom blends (e.g., kick the maker ) re-vealed an effect of the idioms' syntax. Idiom pairs withthe same form tended to blend mo re than pairs with dif-ferent forms. This reflects a general feature of speec h er-rors: interacting elem ents tend to be similar to one another.Stemberger (1985) reported that 99.7% of the noncon-textual errors (phonological or semantic substitutions) inhis error corpus obeyed the grammatical category con-straint. In the idiom blends, 9 3% involved the substitu-lions of words from the same grammatical class. Th e rateexpected by chance was 29%.

Turning to the intra-idiom errors (e.g., kick the rock),we see a similar grammatical category constraint, with88% of the substitutions involving words of the sameugramm atical category . In other ways, however, the errordistribution differed from that for blends. Among the intra-idiom errors, there were more substitutions for idiom pairswith similar meanings and fewer for idiom pairs with thesam e structure, while for blends, syntactic similarity re-sulted in more er rors. This differen ce may reflect the na-ture of the two error types. When the paired idioms haddifferent structures, there was less support from the syn-tax d ue to the absence of s t ructural repet it ion (Bock,1990). This may allow for the intrusion o f lexical items thatwere semantically or phonologically similar. But whenthe paired idioms had the sa me structure, structural rep-etition supported and m aintained the components of thetwo idioms in integrated structure s.Taken together, the in fluence of syntactic form and theinvolvement of grammatical class argue against the strongunitization hypothesis. Idio ms are not produc ed as "frozenphrases" in which the compon ents lose all of their indi-vidual word-like properties. Instead, they are syntacticallyanalyzed.

THE PRODUCTION O F IDIOM BLENDS 63

EXPERIMENT 2A subsidiary finding in Experiment 1 suggested thatthe literal meanings of idioms might enhance their par-ticipation in errors. The seco nd experiment was designedexpressly to see whether the literal meanings of idiomswere active during production.Th e literal and fig urative meanings of idioms play dis-tinct roles within the standard prag matic hypothe sis, theunitization hypothesis, and the direct acce ss hypothesis.All of these m aintain that the literal meaning is used insome way during comprehension. However, the role ofliteral meaning may be quite different in production. Alistener begins with the task o f trying to resolve an inc om-ing stream o f words, while a speaker begins with the con-ceptual content. The comprehender must decide whetherthe literal or figurative interp retation of the phrase is ap-propriate. Th e speaker, knowing what is to be conveyed,should access the idiom through its idiomatic meaning

without mediation from the literal meaning. For example,a speake r intending to convey that [How ard died] shouldbe able to activate "Howard k icked the bucket" withoutfirst accessing a literal meaning alon g the lines of "How-ard punted the pail." However, Cacciari and Gluck sberg(1991) have argued that the literal meaning of an idiomis critical for use in d iscourse. This implies that the literalmeaning of an idiom must be accessible at some pointduring production.The present e xperiment examined whether literal mean-ing affects idiom production. If idiom production is in-depend ent of literal mean ing, then there should be no in-fluence of an idiom's literal mean ing on the productionof idiom errors. O n the other hand, if the literal meaningof an idiom does play a role in its use, one would expectto see perturbations from literal-meaning similarity inthe production of errors.Four pairings of phrases were used: (1) A standardidiom and a literal phrase with the same literal meaningas the standard (hold your tongue; grab your l ip), ( 2 ) hestandard idiom and a phrase with a different literal mean-in g (hold your tongue; s ign your nam e), ( 3 ) he standardidiom and an idiom with the sam e figurative meaningbut a different literal meaning (hold your tongue; buttonyo u r l i p ) , and ( 4 ) the s tandard idiom and an idiomwith a different figurative meaning and a different literalmeaning (hold your tongue; l ip your l id) . All of themhad the same syntactic structure. The critical compar-isons are tested in the first two conditions, those with lit-eral phrases. If the literal mean ing of an idiom is avail-able during production, a phrase with a similar literalmeaning should part icipate in more blends than thepaired phrase with a different literal meaning. The lasttwo conditions correspond to the same-meaning-same-syntax and different-meaning-same-syntax con dition s inExperiment 1, and provide a baseline for the contribu-tion of figurative meaning.


8/15

64 CUTTING A N D BOCK

MethodParticipants. Ninety-six different undergrad uates from the samesource as the previous experiment participated in this experiment;18 speakers were replaced for the same reasons as in Experiment 1.All were native speakers of English.Materials. Thirty-two same-meaning-same-syntax idiom pairswere selected from the stim uli used in Experiment 1. For the stan-dard idiom in each pair, a paraphrase of its literal mean ing was cre-ated by substituting two (in 30 instances) or three (in 2 instances)of the words in the idiom with wo rds of similar meanin g. Each stan-dard idiom was also paired with an unrelated literal phrase. All ofthe related and unrelated literal phrases had the same syntacticstructure as their paired standard idioms. The complete set of ex-perimental items is presented in the Appendix.In addition to the experimental idioms. there were 80 practiceand filler items. Of these, 40 w ere pairs of idioms with differentfigurative meanings and 40 were pairs consisting of one idiom andone literal phrase with different literal meanings. The members ofall of these pairs differed in syntactic form (but had the same num-ber of words).Each list contained 32 experimental and 80 filler items. The 112pairs were broken into five blocks, with the first block, consistingof 16 filler items, serving as a practice block. Each o f the remain-ing four blocks consisted o f 8 experimental and 16 fillers randomlydistributed with the constraint that there was always at least onefiller item between any two experimental items.Design. Crossing meaning type (figurative or literal) and simi-larity of meaning (similar or different meaning) yielded four con-ditions. In the figurative-meaning conditions, both item s of the pairconsisted of idioms with either similar or different figurative mean-ings. In the literal-meaning con ditions, the pair consisted of oneidiom and one literal phrase, with either similar or different literalmeanings. Note that the type of meaning that is sim ilar in thefigurative-meaning condition is the figurative meaning and in theliteral-meaning condition is the literal meaning. An example of anexperimental pair in each condition is given in Table 4 . As in Ex-periment 1, probe number and pair order were completely crossedwith these conditions. This resulted in 16 counterbalan ced lists, eachcontaining eight items in each of the four conditions, with two itemsof each type in each of the four probelorder conditions. Ac ross lists,every item appeared once in each of the conditions.Procedure. The procedure was the same as in Experiment 1 .Scoring. The scoring criteria were identical to those in the firstexperiment. The scoring yielded 8 3.4% corrects, 1.8% interidiomerrors. 5.2% blend errors, 1.9% intertrial errors. 2.0% complex er-rors, and 5.8% miscellaneous errors. There was 97.9% agreem entof scores betwee n the two raters over all of the response s.Separate two-way (meaning type X similari ty of meaning)ANOVAs were performed on each scoring category by speakersand items.ResultsProduction latencies. The production latencies forcorrect responses in each condition are shown in Fig-ure 2 . There was an effect of meaning type such that- ..phrases in the figurative co nditions (con sisting of idiom

Table 4Examp les of the Idiom Pairs Used in theConditions of Experiment 2Meaning Type

Meaning Similarity Figurative LiteralSimilar meanin gs hold your tongue hold your tonguebutton your lip grab your lipDifferent meanin gs hold your tongue hold your tongueflip your lid sign your name

pairs) were produced faster than phrases from the literalcondition s (consisting of idiom-literal-phrase pairs),765 and 784 m sec, respectively. This difference was sig-nificant by both speakers and i tems [Fl(l ,95) = 4.89,MSe = 7,091; F2(1,31 ) = 4.33, MSe = 2,560). There wasno difference between phrases in similar-meaning versusdifferent-meaning pairs (775 vs. 774 msec, all Fs < 1.0).The interaction between m eaning type and meaning sim-ilarity was also not significant [F ,(l ,9 5) = 2.27, MSe =6,651; F2( l ,31 ) = 0.03, MSe = 4,4511.Error analyses. Table 5 shows the number of responsesin each scoring category. The pattern of idiom blends isshown in Figure 3. Ther e was an overall effect of mean-ing similarity for blend erro rs. Phrase pairs with the sam emeaning (either literal or figurative) produced signifi-cantly more blend errors than did pairs with differentmeanings [Fl(1,95)= 5.72,MSe = 0.13; F2(1,3 1) = 6.37,M S = 0.361. There was no main effect of meaning type,and nor did the meaning type X meaning similarity inter-action reach statistical reliability [F,( 1,95 ) = 1.67, MSe =0.13; F2(1,3 5) = 1.44,MS, = 0.441.There were twice as m any complex errors in the figura-tive conditions (those with paired idioms) a s in the literalconditions (those with paired idioms and literal phrases).40 to 20. This result yielded a significant main effectof meaning type by speakers [F,( l ,95 ) = 5.67, MS. =0,181 and marginal by i tems [F 2(l ,31 ) = 3.97, MS. =0.791.The analyses of corre ct responses, intra-idiom, inter-trial, and miscellaneous errors detected no significant dif-ferences attributable to meaning type or meaning simi-larity; nor w ere there significan t interactions between thetwo factors.DiscussionThe most important result from this study was thatliteral-meaning similarity between an idiom and a phraseproduced as many errors as did figurative-meaning sim-ilarity. Underscoring this is the fact that mor e than halfof the erro rs occurred on content words. Since contentwords carry the burden o f the meaning, this finding im-plies that the errors reflect a tendency for sim ilar mean-ings (either literal or figura tive) to interact. Finally, thetendency for blending errors to obey the gram matica l classconstraint held in both the figurative and literal condi-tions, sugge sting that this cons traint is blind to whetherthe blending phrases are idioms. All of these findingsoffer evidence that literal meaning is active during theproduction of idioms, consistent with Cacciari and Glucks-berg's ( 1991) contention that the literal meaning of anidiom is involved in their productive use in disc ourse.One objection to thes e results is that the speake rs mayhave ignored the figurative meanings of the idioms andfocused on the literal meaning s. However, there are rea-sons to believe otherwise. The figurative phrases wereproduced faster than the literal phrases, paralleling manyof the finding s in the com prehension l i terature. Andthere were more blends in the same-figurative-different-literal meaning condition than in the different-figurative-


9/15

Table 5Numb er of Correct Responses and Distribution of Errors inExperiment 2Meaning Condition

Literal FigurativeScoring Catego ry Similar Different Similar DifferentCorrect 635 663 633 63 1Miscellaneous 44 38 50 45Complex 1 1 9 17 23

Idiom BlendsAdditio n 0 [Ol 0 [OI 0 [ol 0 [olSubstitution 18 [I51 5 [I ] 18[8] 14[2 ]

Intra-Idiom ErrorsAddition 2 [ol 2 [ I ] 0 [OI 9 [71Deletion 5 [ol 2 [ol 2 [OI 5 [ I ]Movement 0 [ol 1 [ol 1 [ol 0 [olSubstitution 40 [33] 30 [I91 34 [25] 28 [20]

Intertrial ErrorsAddition 1 [I ] 1 [ol 2 [ol 0 [olSubstitutio n 12 [4] 17 [3] 1 1 [2] 13 [4]

Note-Values in square bracke ts are the number of errors involving con-tent words. Literal refers to the con ditions consis ting of an idiom and aliteral phrase. Figurative refers to the co nditions consisting oftwo idioms.

different-literal meaning condition, replicating Experi-ment 1. This is most evident in the content word errors.The implication is that the speakers processed the figu -rative as well as the literal meanings.EXPERI MENT 3

Experiment 3 was designed with a different dimensionof idiomatic meaning in mind. Gibbs and Nayak (198 9)demonstrated that speakers' intuitions about the decom-posability of an idiom are correlated with the idiom'ssyntactic flexibility. In decom posable idioms, the lexicalparts of the idiom map o nto distinguishable meaning com -ponents. Decomposable idioms may thus be less rigidlyencoded in the lexicon than nondecornposable idioms.For the purposes of the experiment proposed here, thisleads to the prediction that decom posable idioms may bemore susceptible than nondecomposable idioms to theproduction of idiom blends.As in Experiment 1, same-meaning-same-syntax idiompairs were created. The pairs consisted o f either decom-posable or nondecomposable idioms, as determined fromdecomposability ratings. T he syntactic structure and fig-urative meaning s of the idiom pairs were the same, leavingany differences in the error pattern attributable to de-composability alone. If the lexical representation of de-comp osable idioms is less rigidly sp ecified and more sus-ceptible to chang e (Gibbs & Nayak, 1989), they shouldbe more predisposed to blending than nondecom posableidioms, if the latter tend to be lexically frozen.


Materials. The idiom pairs were created from a set of idiomsthat were rated for decomposability . The rating procedure wasadapted fro m Gibbs a nd Nayak (1989 ). Twenty-six judges receiveda list of 141 idioms and a paraphrase of their idiomatic meaning(e.g. ,"Chew the fat: To talk casually"). The judges were asked to rateon a 5-point scale how analyzable they thought the idiom was. Theinstructions explained the nature of a decomposable relationshipbetween an idiom's words and its figurative meaning in terms of anexamp le (pop the question) and contrasted this with a nondecom-posable idiom (givethe bounce). The ratings were made on a 5-pointscale, on which 1 was nondecomposable and 5 was decomposable.Included in the rated idioms were the 72 idioms from the same -meaning-same-syntax condition of Experiment 1. From these 36pairs, 10 pairs in which both idioms were relatively decomposableand 10 pairs in which bo th idioms were relatively nondecomp osablewere selected as the experimental items for the third experiment.These items and their decom posability ratings are presented in theAppen dix. The median rating of the idioms used in the decompos-able pairs was 3.5 (range 3.1 to 4.2); for nondecomposable pairs, itwas 2.3 (range 1.7 o 2.9). The mean differences between the decom-posability ratings for the paired idioms were 0.04 for the d ecompos-able pairs and 0.03 for nondecomposable pairs. The mean same1different figurative mean ing ratings (see Experiment 1) were 4.5for the decomposable pairs and 4.0 for the nondecomposable pairs.In addition to the 20 experimental items, 40 filler items were cre-ated . Like the experim ental items, these items were idiom pairs withthe sam e meaning a nd sa me syntax. Ten items were same-meaning-same-syntax pairs from Experiment 1. in which one idiom had beenrated as decomposable and the other as nondecomposable. The re-maining 30 pairs were made up of idioms with different meanings.The 6 0 items were group ed into three blocks of trials. The firstblock, which consisted of 10 fillers, served as practice. The remain-ing two blocks each had 25 idiom pairs (10 experimental items and15 fillers) which were ordered with the constraint that there was al-ways at least one filler item between any two experimental items.Design. The idio m pairings yielded two conditions, decompos-able versus nondecomposable idioms. As in Experiments 1 and 2 ,probe number and idiom-pair order were completely crossed withthese conditions. This created fo ur counterbalanced lists, each con-taining 10 different pairs from each exp erimental condition.Procedure. The procedure was identical to that of the previousexperiments.Scoring. The sam e scoring criteria were applied. This resulted in81 l % corrects, 3.6% idiom b lends, 5.8% intra-idiom errors, 0.7%

Figurative LiteralMethod Meaning TypeParticipants. The participants in this experiment were 80 dif- Figure 3. Correct-response latencies and idiom blends in Ex-ferent undergraduates drawn from the same source as Experiments periment 2. Th e bars represent the blend errors and the individ-I an d 2. Sixteen were replaced due to problems w ith the voice key or ual box points represent the naming times. Th e error bars reflectwith reading the idioms. All were native speake rs of English. confide nce intervals for the Tukey tests.


10/15

66 CUTTING AND BOCK

intertrial errors, 1.8% com plex errors, and 7.0% m iscellaneous er -rors. There w as 99.2% agreemen t between the two raters o\e r all ofthe responses.One-way repeated measures ANOVAs were computed for theerror and production-latency data, with speakers as the random fa c-to r (F,), nd one-way between-groups ANOVAs were computedwith i tems as the random factor (F, ) .Results

Production latencies. The production latencies forcorrect responses are shown in Figure 4. The nam ingtimes were slightly faster for decomposable idiom pairs(791 msec) than for nond ecomposable pairs (804), butthe difference was not statistically reliable [F,(l,79) =1.15,MS, = 5,959; F2(1,1 8) = 1.43,MSe = 1,6371. Usingth e MS, from the ANOVA as an estimate of the popula-tion variance, we estimated that this experiment couldhave detected a 34-msec difference with a power of .80.This difference is comparable to the significant differ-ence found in Experiment 1 .Error analyses.Table 6 shows the number of responsesin each scoring category. The ANOVA detected no dif-ference between the co nditions in numb ers of correct re-sponses or errors, except in the miscellaneous error cat-egory. Nondecomposable idiom pairs produced moremiscellaneous errors than did decomposable pairs, sig-nificantly so by speakers but not by item s [68 vs. 45;F , ( l , 7 9 ) = 7.41, MS, = 0.49; F2(1,18) = 1.05, MS, =27.41, The pattern o f idiom blends is shown in Figure 4.Using the MSe from the ANOVA as an estimate of thepopulation variance, we estimated that the experimentcould have d etected a differenc e of 17 blend errors witha power of .80. This difference is comparable to the sig-nificant 20-error difference found in Experiment I .

DiscussionThe p redicted differences in the error rates between thedecomposable and nondecornposable pair ings did notmaterialize. This s ugge sts that the lexical representationsof decomposable and nondecornposable idioms are thesame when they enter into the production process. and

Figure 4. Correct-response latencies and idiom blends in Ex-periment 3. The bars represent the blend errors and the individ-ual box points represent the naming times. The error bars reflectconfidence intervals for the Tukey tests.

Table 6Number of Correct Resnonses and Distribution of Errors

in Experiment 3Decomposability Classification

Scoring Category Nondecomposable DecomposableCorrect 64 1 657Miscellaneous 68 44Complex 17 13

Idiom BlendsAddition 0 [ol 0 [olSubstitution 27 [71 30 [41

Intra-Idiom ErrorsAddition 0 [ol 2 [o lDeletion 2 111 3 Lo]Movement 1 [I ] 0 Lo]Substitution 40 [40] 45 [41]

Intertrial ErrorsAddition 0 I01 1 I01. Substitution 6 [ I ] 5 [o ]Note-Values in squa re brackets are the numb er of errors involvingcontent words.

supports a weak version of the unitization hyp othesis overthe decomp osability hypothesis.To ensure that the null result could not be attribute d tonorm ing disparities, we compared ratings for overlappingitems in the published literature (Gibbs, 1987; Gibbs &Nayak, 1989 ; Gibbs, Nayak, Bolton, & Keppel, 1989;Gibbs, Nayak, & Cutting, 1989). Of the items in the pres-ent experiment, there were six used by Gibbs and his col-leagues. All were in agreement. Three of the items wererated as nondecomposable in Gibbs's work and by ourjudge s and three items were rated as decomposable in bothsets of materials. So , althoug h the number o f overlappingitems is not large, the convergence is reassuring.The single difference obtained in this experiment wasthat nondecom posable idiom pairs produced more miscel-laneous errors. If nondecomposable idioms are harder tounderstand and remember than decomposable idioms,these errors would stem from mem ory failure. This is thesame result that Gibbs and Gonzales (1985) reported.They presented participants with short stories that endedin either a sy ntactically frozen idiom, a syntactically flex-ible idiom, or a literal phrase. The participan ts' task wasto recall the fin al sentence of each story correspondingto a cu e. The flexible idioms were recalled more often thanthe frozen idioms (46% vs. 32%), implying that nonde-composable idiom s are more difficult to remember. Theseresults offer further evidence that the decomposable id-ioms in the present experiment differed in the relevantproperties.

GENERAL DISCUSSIONThe experiments presented here offer a new kind ofevidence for the view that idioms are composit ional .Across all three experiments, idiom blend errors con-sistently involved structurally and semantically similarcom pone nts. Like literal phrases, idioms fall apart along


11/15


linguistically sensible lines. Apparently, idioms are notlexicalized chunks, comparable to large single words,but phrases with internal syntactic and sem antic compo-nents. Among these com ponents, there are some that pro-vide access to an idiom's literal meaning during lan-guage production.The first experiment suggested that the production ofidioms was sensitive to their internal syntactic prope rties.Idioms with the same syntact ic s tructures were morelikely than idioms with different structures to interfere withone another, resulting in a greater numbe r of blends in theidioms with the same structures. Additionally, substitu-tion errors tended to obey a gram matical category con-straint. These results argue against the unitization hy-pothesis (Swinney & Cutler, 1979 ), suggesting instead thatidiomatic representations include syntactic information.In the results for production latencies, Exp eriment 1 fur-ther suggested that idioms with similar figurative mean-ings tend to interfere with each other. The second exper-iment went beyond this to ex amine the influence of anidiom's litera l mea ning on its blending behavior. T he re-sults indicated that the literal meanings o f idioms m ightbe active during production, consistent with the claimsof Cacciari an d Glucksberg (199 1). Phrases w ith sim ilarmeanings produced m ore blending errors than did phraseswith dissimilar m eanings, regardless of whether the mean-ing similarity was literal or figurative.If we assume that idioms are stored phrases with con-stituent parts, someth ing must be said about the nature ofthe individual parts. The final experiment explored therole of an idiom's decomposab ility during production. De-composability refers to the assumptions that speakersmake about how the meanings o f the parts of an idiomcontribute to the idiom's mea ning as a whole, and ratingsof decomposability are correlate d with judgments of syn-

Syntax Lexicon

tactic flexibility. However, in Experiment 3 decompos-able idioms blended together at the same rate as nonde-composable idioms, suggesting that the componen ts ofdecomposable and nondecomposable idioms are accessedsimilarly during production.This result contrasts with the outcome of studies byGibbs, Nayak, and Cutting (1989). They presented par-ticipants with strings of words an d asked whether theywere meaningful English ph rases. Half of the meaning-ful i tems were idioms and the other half were l i teralphrases. Responses to nondecom posable idioms wereslower than those to decompo sable idioms. Gibbs et al.proposed that during co mprehension, parts of the idio-matic meaning are distributed over the components ofthe idiom. For example, in interpreting the decomposableidiom spill the beans, a comprehender recognizes thateach componen t can be assigned an independent mean-ing, and understands how it contributes to the whole. How -ever, for a nondecomposable idiom like kick the bucket,

the comprehender attempts to assign independent mean-ings to the parts but cannot, leading to a slower response .For language production, the story may be different be-cause the producer starts with the ideas to be com muni-cated . That is, the production system begins with a thoughtand then con structs an utterance to convey it. A languag eproduction model like that developed in Dell (1986; asimilar model might be constructed from the model de-velope d in Levelt, 1 989) offers an explicit framewo rk forexplain ing how this proceeds. Figure 5provides a sketch.In the m odel, there is a distinction between syntax andthe lexicon. The syntactic part of the model consists of aset of rules that create a structural frame, The terminalnodes o f these frames a re grammatically categorized slots.The lexicon consists of a network of nodes for linguisticunits such a s concepts, words, morphemes, and phone mes.

Figure 5. A model of the lexicon, includ ing representations for the id iom spo p thequestion an d kick the bucket.


12/15

68 CUTTING AND BOCK

The conceptual nodes a re connected to the word nodes,the word node s to the morphemes, and so on. The crosstalk between the syntactic frames and the lexicon occ urswithin each level. For example, at the lexical level, in-sertion rules select activated lexical items from the lexi-con to fill the slots in the syntactic frame. Th e lexical itemassigned to a slot must be of the grammatical type that aslot specifies.Let us sup pose that the lexicon has nodes representingidioms as wholes. These nodes are situated between theconceptual a nd lexical layers of nodes, so we will referto them as lexical-concept nodes (fol lowing Roelofs ,1992). These lexical-concept n odes are associated withthe syntactic portion o f the system, and can retrieve phrasalframes that specify the ordering of gramm atical s lotsrather than single lexical categories. That is, the lexicalrepresentation of an idiom (e.g ., kick the bucket) is asso-ciated with a phrasal node ( a verb phrase) in the syntacticpart of the system, not to a single grammatical category(e.g., verb). In terms of the mode ls of idiom representa-tion presented here, this corresponds to the differencebetween the weak and strong versions of the unitizationhypothesis.In the lexicon, an idiom's lexical-conceptual node isalso associated with lexical nodes (lemmas) correspon d-ing to its componen t parts. That is, the representation ofan idiom like kick the bucket is asso ciated with individuallemm as (kick, the, and bucket). A lemm a is a representa-tion of a word that is associated with its gramm atical class(Roelofs, 1992). On this view, the decomposability of a nidiom is related to the way in which lexical-conceptualrepresentations are associated to conceptual representa-tions. If an idiom is conceptually decomposable, there willbe multiple links between the lexical-conceptual a nd con-ceptual levels, while a nondecom posable idiom will havea single association betwe en the two levels. For examp le,in Figure 4, the lexical-conceptual node for kick thebucket is linked to the single conceptual no de died, whilethe node fo rp op the question is linked to the conceptua lnodes for both suddenly and to prop ose.This model p redicts that in production, nondecompos-able idioms will be as svntacticallv flexible as decomo os-

conceptual level of two idioms with similar figurativemeaning s (e.g., kick the bucket and meet your m aker) iscomparable to the relationship between the conceptualand lexical-conceptual levels for two words with similarl i teral meaning s (e.g. , bucket and pa i l ) . Thus, the in-crease in error production that occurred when an idiomwas paired with a literally similar phrase may arise as aconseque nce of sh ared conceptual representations.The m odel also shows how the increase in blending er-rors that occurred when idioms had the same syntacticform may result from sharing the same syntactic repre-sentation. In the m odel, the gram matical class restrictionis a conseque nce of the insertion rules that are responsi-ble for the interaction between the lexical and syntacticparts of th e system. The influence of literal meaning arisesfrom com petition between related lexical nodes that areactivated by the idiomatic representation.Before we co nclude, there are several concerns to beaddressed a bout the results of the present experiments.Th e first is the possibility that the errors that occurred werenot production e rrors but reading erro rs, slips of the eyerather than slips of the tongue. In an attempt to distinguishthese two potential source s of error, speakers were askedafter each trial whether they had responded correctly, onthe assum ption that if speakers were aware that they hadmad e an error it was safe to conclude that the error wasin speech rather than in reading. Th e pattern of these "pureproduction" erro rs was similar to the overall distribution,but their numbers were too small to permit formal analy-sis. Table 7 presents these pure production errors for allthree experiments. Although it is possible that the bal-ance of the errors a rose in reading, a more plausible in-terpretation is that error detection w as difficult. In one ofvery few studie s to examine de tection rates, Levelt (1983)reported that only 46% of color-word errors were re-paired in a task that required c olor naming. Levelt con-cluded that this was due to failure s in the detection of er-rors by the speakers.Another concern relates to our focus on idiom repre-sentation and production. Because the task had an inher-

able idioms. On ce the lexical-conceptual representation Table 7Comparison of All the Idiom Blends and theof an idiom has been selected (through activation from pure production Idiom in All of thethe conceptual nodes), syntactic production processes are All Pureindifferent to an idiom's decomposability, implying that Idiom Productionspeakers should be as likely to vary the f orm of a non- Experimental Condition Blends Blendsdecomposable idiom as a decomposable one. Although Experiment 1comprehension studies have demonstrated that nonde- same syntax 29 9compo sable idioms are more difficult to comprehend a Different meaning/same syntax 20 11s t ructura l t rans format ion than decom posable id ioms Differe nt meaningldifferent svntax 9 4(Gibbs& Nayak, 1989), the question ha sn ot yet been ad-dressed in production.In terms of the mode l sketched above, the increase inerror production w hen idioms have the same figurativemeaning can be seen as a consequence of similar con-ceptual representations, creating more competition thandissimilar conc eptual representations. Note that the rela-tionship between the conceptual level and the lexical-

Experiment 2Simil ar literal meanin g 18Different literal meani ng 5Similar figurative meani ng 18Different figurati ve meanin g 14

Experiment 3Nondecomposable 27Decomposable 30


13/15

THE P R ODUC TION O F I DI OM B LENDS 69

ent comprehension component, the results could reflectcomprehension more than production processes. Thereare three reasons to discount this possibility. First, therewas a close parallel between the overall error pattern andthe "pure produ ction" errors for which the speakers iden-tified their slips as production errors. The second argu-ment against the com prehension explanation is the lackof any effects of decomposab ility in Experiment 3. Thisis at odds with results in the idiom -comprehension liter-ature (Gibbs & Gonzales, 1985; Gibbs & Nayak, 1989).The last argument stems from a finding by Lombard! andPotter (1992), which showed that short- term memorytends to be mean ing-based even after a very short delay.Recall is based on the m eaning o f the utterance, expressedthrough normal production p rocesses. Their results sug-gest that the utterances produced after the 2-sec delay inthe present experim ents were likely to have involved nor-mal production processes.~ a k e nogether, these points suggest that the com pre-hension contamination of the experimental measures w asminimal. O f course, these effects cannot be comp letelyruled out and some aspects of our results may be morereadily explained in terms of co mprehension processes.One is the literal mean ing effect in the second experiment.Cacciari and Tabossi (1988) showed that, once a phrase isidentified as an idiom, literal processing is halted and theliteral interpretation is lost. his challenges the interpret a-tion that literal meaning h ad its impact during production.A final conce rn is the relative weakness o f the effects.The task that we used produced very sm all differences inerrors and production latencies. his is an unfortunatebut unsurprising consequ ence of the highly practiced na-ture of normal production and the resulting rarity of speecherrors. Still, as a vehicle for a preliminary exploration ofthe representation o f idioms by speakers, the me thod hasdecided advan tages over the observation of naturally oc-curr ing errors.In conclusion, these experiments suggest that idiomsare not produced a s "frozen phrases," devoid of informa-tion about their internal syntax and semantics. As othershave argued from studies of comprehension (C acciari &Glucksberg, 1991; Gibbs & Gonzales, 1985; Gibbs &Nayak, 1989), idiom representations are linked to infor-mation about the grammatical class of their constituents,about their overall syntactic structures, and about literalmeaning. W ithin the model sketched above, idioms areidiom s because of how they are associated to conceptualrepresentations, a s well as how they are represented in thelexicon. Idiom s may be special in their relationships tononl inguis tic concepts, but they are not special in the waythey are produced in normal lang uage use.

REFERENCESB A A R S , . J . (1980a). The competing plans hypothesis: An heuristicviewpoint on the causes oferrors in speech. In H. W. Dechert & D. A.Raupach (Eds.), Temporal variables in speech (pp. 39-49). New York:Mouton.B A A R S .. J. (1980 b). On eliciting predictable speech errors in the lab-oratory. In V A. Fromkin (Ed.), Errors in linguistic performance:

Slips of the tong ue, ear, pen and han d (pp. 307-3 18). New York: Aca-demic Press.BORROW,. A,, & B E L L , . M. (197 3). On catching on to idiomatic ex-pressions. Memory & Cognition, 1, 343-346.BOCK, . K. (1990). Structure in languag e: Creating form in talk. Amer-icon Psychologist, 45, 1221-1236.BUTTERWORTH,. L. (1982 ). Speech errors: Old data in search o f newtheories. In A. Cutler (Ed.), Slips of the tongue and language pro-duction (pp. 73-108). Amsterdam: Mouton.CACCIARI ,., & G L U C K S B E R G ,. (1991). Understanding idiomatic ex-pressions: The contribution of word meanings. In G. B. Simpson(Ed.), Understanding word and sentence (pp. 2 17-240). Amsterdam:North-Holland.CACCIARI ,., & TABOSSI,. (1988). The comprehension of idioms. Jour-nal of Memory & Language, 27, 668-683.DELL,G. S. (1986). A spreading activation theory of retrieval in lan-guage production. Psychological Review, 93,283-321.DELL,G . S . ,& RE ICH , . A. (1981). Stages in sentence production: Ananalysis of speech error data. Journal of Verbal Learning & VerbalBehavior, 2 0 , 6 1 1-629.DIK,S. C. (1989). Idioms in a computational functional grammar. InM. Everaert & E. Van der Linden (Eds.), Proceedings of the firstTilburg workshop on idioms (pp. 41-55). Tilburg: ITK Proceeding s.FAY,D. (1982). Sub stitutions and splices: A study of sentence blends.In A. Cutler (Ed.) , Slips of the tongue and language production(pp. 163- 195). Amsterdam: Mouton.F R A S E R ,. (1970). Idioms w ithin a transformational grammar. Foun-dations of Language, 6 , 22-42.F R O M K I N ,. A. (1971). The non-anom alous nature of anomalous utter-ances. Language, 47.27-52.FROMKIN,.A. (1 973). Speech errors as linguistic evidence. The Hague:Mouton.GA RRE TT , . F. (1980). Levels of processing in sentence production. InB. L. Butterworth (Ed.), Language production (Vol. 1, pp. 177-220).New York: Academic Press.GIBBS, .W. (1984). Literal meaning an d psychological theory. Cogni-tive Science, 8,275-304.GIBBS, . W. (1986). Skating on thin ice: Literal meaning and under-standing idio ms in conversation. Discourse Processes, 9, 17-30.GIBBS, . W. (1987). Linguistic factors in children's understanding ofidioms. Journal of Child Language, 14 , 569-586.

GIBBS, .W., & G O N Z A L E S ,. P. (1985). Syntactic frozenness in pro-cessing and remembering idioms. Cognition, 20,243-259.GIBES, . W., & N A Y A K ,. P. (1989). Psycholinguistic studies on thesyntactic behavior of idioms. Cognitive Psychology, 21, 100-138.GIBBS, . W. , NAYAK,. P., BOLTON,. L., & KEPPEL,M. E. (1989).Speakers' assumptions about the lexical flexibility of idioms. Mem-or y & Cognition, 17, 58-68.GIBES, .W,, NAYAK,. P., & C U T T I N G ,. C. (1989). How to kick thebucket and not decompose: Analyzability and idiom processing. Jour-nato/Mernory & Language, 28, 576-593.H A R L E Y ,. (1984). A critique of top-down independent levels modelsof speech production: E vidence from non-plan-internal speech errors.Cognitive Science, 8, 191 2 19,KATZ, . J. (1973). Compositionality, idiomaticity, and lexical substitu-tion. In S. R. Anderson & P. Kiparsky (Eds.), A festschrift for MorrisHalle (pp. 357-3 76). New York: Holt, Rinehart & Winston.KATZ, . J. , & POSTAL,. ( 1963). Sema ntic interpretation of idiom s andsentences containing them (Quarterly Progress Report No. 70, pp. 275-282). Cambridge, MA: MIT Research Laboratory of Electronics.LAPOINTE,. G. , & D E L L ,G. S. (1989). A synthesis of some recentwork in sentence production. In G. N. arlson & M. K. Tanenhaus(Eds.), Linguistic structure in language processing (pp. 107-156).Dordrecht: Kluwer Academic Publishers.LEVELT, . J . M . (1983). Monitoring and self-repair in speech. Cogni-t ion, 14 , 41-104.LEVELT, . J . M. (1989). Speaking: From intention to articulation. Cam-bridge, MA: MIT Press.L O M B A R D I ,., & P O T T E R , . C. (1992 ). The regeneration of syntax inshort term memory. Journal ofMemo ry & Language, 31,713-733.MACKAY,. G. (19 72). The structure of words and syllables: Evidencefrom errors in speech. Cognitive Psychology, 3, 210-227.


14/15

70 C UTTI NG AND BOCK

M A K K A I ,. ( 1987).A dictionary of American idioms. New York: Bar-ron's.MCGLO NE, . S . , GLUCKSB ERG,., & CA CCI A RI ,. (1994). Semanticproductivity and idiom comprehension. Discourse Processes, 17,167-190.MOTLEY, . T. (1985, Septem ber). Slips of the tongue. Scientific Amer-ican, 253, 116-127.MOTLEY, . T., BAA RS.. J. ,& C A M D E N ,. (1983a). Experimental ver-bal slip studies: A review and editing model of language encoding.Communication Monographs, 50, 79-101.MOTLEY, . T.,BA A RS,. J . , & C A M D E N ,. (1983b). Formulation hy-potheses revisited: A reply to Stemberger.Journal of Psycholinguis-tic Research, 12, 561-566.N U N B E R G ,. (1978). The pragmatics of reference. Bloomington: Indi-ana University Linguistics Club.ROELOFS,. (1992). A spreading activation theory of lemma retrievalin speaking. Cognition,42 , 107-142.SH A TTU CK - H U FN A G EL,. (1 979). Speech errors as evidence fora serialordering mechanism in sentence production. In W. E. Cooper &E. Walker (Eds.), Sentenceprocessing: Psycholinguistic studiespre-sented to Merrill G arrett (pp. 295-324). Hillsdale, NJ: Erlbaum.STEM BERG ER,. P. (1982). Syntactic errors in speech. Journal of Psy-cholinguistic Research, 11, 3 13-333.S T E M B E R G E R ,. P. (1985). An interactive activation model of lang uageproduction. In A. Ellis (Ed.), Progress in the psychology of language(Vol. 1, pp. 143-186). London: Erlbaum.S W I N N E Y ,. A , , & C U T L E R ,. (1979). The access and processing ofidiomatic expressions. Journal o f Verbal Learning & Verbal Beha v-ior, 1 8, 523-534.VA NDE R L I N D E N ,. (1 989). Idioms and flexible categorical grammar.In M. Everaert & E. Van der Linden (Eds.), Proceedings of the firstTilburg workshop on idiom s (pp. 127-143).Tilburg: ITK Proceedings.WASOW,., SAG , . A, , & N U N B E R G ,. (1983). Idioms: An interim re-port. In S. Hattori & K. Inone (Eds.), Proceedings of the XIII Inter-national Conference o f Linguists (pp. 102-115). Tokyo: CIPL.W EI N REI CH ,. (1969). Proble ms in the analysis of idiom s. In J. Puhvel(Ed.), Substance and structure of language (pp. 23-81). Berkeley:University o f California Press.WOOD,M . (1986).A definition of idiom. Bloomington: Indiana Univer-sity Linguistics Club.

NOTES1. In theories of language production, a speech plan is not a con-scious, strategic plot of what to say, but a tacit, abstract represen tationof the linguistic components of an utterance.2. This was done to separate "pure production" errors from potentialreading errors. The assumption was that if speakers were able to reportmaking the error, then they had not made a reading error. Because thepattern in both sets of errors looked very much like the pattern in the fullset, and because the numbers of errors were small, the results are re-ported in terms of the total error set in each of the experim ents.3. Restricting the analysis to the items that differed in literal mea ning,there were 46 errors in the same-meaning-same-syntax condition com-pared to 24 in the different-meaning-same-syntax condition.

APPENDIXIdiom Pairings from Experiment 1Key: s tandard id iom (same meaning-same syntax; d ifferentmean ing - sam e syn tax ; d i f f e ren t mean ing -d i ff e ren t syn tax )

1 shoo t the b reeze ( chew the f a t ; r a i se the roo f ; n ip and tuck )2 hold your tongue (bu t ton your l ip ; f lip your l id ; neck andn e c k )3 eat your words (swallow your pr ide; miss the mark; g o l ikec l o c k w o r k )4 hi t the cei l ing (blow a fuse; hold your horses; take by s torm )5 in your t racks (on the spot ; up the creek; have dibs on)

6 card up y our s leeve (ac e in the hole; com e to the point ; s t ickyour nose in to )7 kick the bucket (meet your maker; h i t the bul lseye; screambloody murde r )8 kick in the pants (s lap in the face; miss by a mile ; catch your

d e a t h o f )9 l ine your pockets ( feather your nest ; s teal the show ; in thecards)10 ru le the roos t (wea r the pan ts ; s t ru t you r s tu f f; head ove rheels)

11 upse t the app leca rt ( rock the boa t ; p romise the mo on ; fu l lo f you r se l f )12 end o f you r rope (back to the wa l l ; f ly o ff the hand le ; invour mind 's eve),1 3 w r a p a r o u n d y o u r f i n g e r ( l ea d b y t h e n o s e ; f o a m a t t h emou th ; thank your lucky s t a r s )14 t ip the sca le s ( tu rn the t ide ; p re s s you r luck ; hea r t o f go ld )15 knoc k fo r a loop ( th row fo r a lo s s ; h i t be tween the eyes ;move heaven and ea r th )16 head fo r the h i l l s ( t ake to the woods ; g r ind to a ha l t ; makethe fur f ly)17 egg on your f ace ( foo t in you r mou th ; ge t o f f you r t a i l; keepyour nose c l ean )18 rocks in you r head (ba t s in the be l f ry ; ge t on the s t i ck; onp ins and need le s )19 on the ropes (ove r a ba r re l ; i n a fog ; nu r se a g rudge )20 mi l l s tone a round your neck (monkey on your back ; ou t o fyour head; s tar t the bal l ro l l ing)21 th row in the towe l (g ive up the sh ip ; shoo t o ff you r mou th ;o f f the bea ten t r ack )22 z ip you r l ip (ho ld your peace ; keep your head ; fo r the b i rd s )23 have a cow ( lo se your coo l ; f ace the mus ic ; o n the f ly )24 bed o f ro ses (bowl o f che r r i e s; cup o f t ea ; on c loud n ine )

25 ove r you r head (beyond your dep th ; on the bal l ; s e t downroots)26 ba t an eye ( r a i se an eyeb row; shake a l eg ; of f the cu f f )27 make the g rad e ( cu t the mus ta rd ; l ay an egg ; wa lk on a i r )28 go wi th the f low (bend wi th the wind ; come to you r senses ;l eave no s tone un tu rned )

29 hi t the sauce (wet your whis t le ; lend a hand; feet of c lay)30 pain in the neck ( thorn in the s ide; feet on the ground; pushthe pan ic bu t ton )3 1 see the l ight (get the pic ture; draw a blank; le t i t r ip)32 we igh one 's w ords (wa tch your mou th ; change your tune ; inthe p ink )

33 take the bul l by the horns ( take the bi t in your mouth; takeyour l i f e in to you r hands ; s t r aw tha t b reaks the came l ' sback )34 lay dow n the law (put down your foot ; take for a r ide; turnthe t ab le s on )35 t ighten your bel t (b i te the bul le t ; l ick your chops; make i t

snappy)36 l ay it on (ham i t up ; g ive i t up ; g rease your pa lm)Idiom Pairings from Experiment 2

Key: s tandard id iom (sam e f igurat ive; same l i teral ; d ifferentf igurat ive; d if ferent l i tera l)1 shoot the breeze (c hew the fa t; f i re the wind; ra ise the roof;ea t the food )2 swallow your pr ide (eat your words; chew your ego; missthe mark; throw th e bal l )3 hold your to ngue (but ton your l ip ; grab your l ip; f l ip yourl id ; s ign your name)


15/15

THE PRODUC TION OF IDIOM BLENDS 7 1

4 hit the ceil ing (blow a fuse; smack the roof; hold yourhorses; read the book)5 kick the bucket (meet your maker; punt the pail; hit thebullseye; answer the phone)6 kick in the pants (slap in the face ; hit in the trousers ; missby a mile; pick up a penny)7 rule the roost (wear the pants; command the nest; pull yourpunches; deposit your check)

8 upset the applecart (rock the bo at; overturn the fruitstand;promise the moon; dial the number)9 end of your rope (back to the wall; last of your string; fly offthe handle; put on your shoes)10 wrap around your finger (lead by the nose; twist aroundyour thumb; foam at the mo uth; go to the bank)11 tip the scales (turn the tide; tilt the balance; hit the road ;mail the letter)12 head for the hills (take to the woods; go to the mountains;grind to a halt; turn on the television)13 egg on your face (foot in your mouth; yolk on your head;get off your tail; go to the party)14 rocks in your head (bats in the belfry; stones in your skull;get on the stick; get in your car)15 monkey on your back (millstone around your neck; chimpon your shoulders; needle in the haystack; coats on thehooks)16 throw in the towel (give up the ship; toss in the rag; shootoff your mouth; live in the house)17 zip your lip (hold your peace; fasten your mouth; keep yourhead; pet your dog)18 have a cow (lose your cool; own a bull; face the music; buya bike)19 bed of roses (bowl of cherries; garden of flowers; cup oftea; glass of milk)20 make the grade (cut the mustard; get the mark; lay an egg;ask the farmer)21 go with the flow (bend with the wind; travel with the cur-rent; come to your sen ses; clean out the garage)22 raise an eyebrow (bat an eye; lift your foreh ead; shake a leg;see an opera)23 over your head (beyond your depth; above your face; on theball; in the kitchen)24 ace in the hole (card up your sleeve; jack in the pit; com e tothe point; stir in the cheese)25 lay down the law (put down your foot; set down the rules;take for a ride; slide down the pole)26 take the bull by the horns (take the bit in your mouth; takethe deer by the antlers; take your life into your hands; takethe child to the hospital)

27 hit the sau ce (wet your whistle; strike the juice; lend a hand;accept the gift)28 pain in the neck (thorn in the side; ache in the throat; feeton the grou nd; fly in the airplane)29 see the light (get the picture; watch the glow; draw a blank;drink the cola)30 watch your mouth (w eigh your words; see your lips; greaseyour palm; write your paper)3 1 on the spot (in y our tracks; atop the dot; up the creek; abovethe door)32 line your pockets (feather your nest; pad your pouch; stealthe show; tie your laces)

Idiom Pairings from Experiment3Key: Idiom 1 (decomp osability rating); Idiom 2 (decompo s-ability rating)

Nondecomposable Idiom Pairs1 shoot the breeze (2.3); chew the fat (2.8)2 line your pockets (2.3); feather your nest (1.7)3 upset the applecart (2.5); rock the boat (2.5)4 knock fo r a loop (2.6); throw for a loss (2.1)5 rocks in your head (2.4); bats in the belfry (2.4)6 on the ropes (2.3); over a barrel (2.1)7 bed of rose s (2.3); bowl of cherries (2.2)8 tighten your belt (2.2); bite the bullet (2.6)9 lay it on (2.2); ham it up (2.1)10 raise the roof (2.0); flip your lid (2.9)

Decomposable Idiom PairsI 1 hold your tongu e (4.2); button your lip (4.1)12 end of your rope (3.2); back to the wall (3.2)13 wrap around you r finger (3.2); lead by the nose (3.2)14 tip the scales (3.4); turn the tide (3.2)15 throw in the towel (3.1); give up the ship (4.2)16 zip your lip (4.0); hold your peace (4.0)17 over your head (3.5); beyond your depth (3.6)18 go with the flow (3.8); bend with the wind (3. 2)19 pain in the neck (3.9); thorn in the side (3.5)20 see the light (3.5); get the picture (3.2)

(Manuscript received January 25, 1995;revision acce pted for publication September 26, 1995.)

cutting bock 1997

Documents