Identifying Existing and Novel Compound Words in Reading Finnish: An Eye Movement Study Jukka Hyönä Minna Koski Alexander Pollatsek Department of Psychology Department of Psychology Department of Psychology University of Turku University of Turku University of Massachusetts Turku, Finland Turku, Finland Amherst, MA, USA hyona@utu.fi mmjkos@utu.fi pollatsek@psych.umass.edu latest stages of processing during the first-pass 1 Introduction reading, indexed by fixation time spent on the target word after fixating away from the first According to the dual-route race model of com- constituent but before exiting the word, only a pound word identification (Pollatsek, Hyönä, & main effect of novelty was observed. As regards Bertram, 2000), the holistic route and the mor- to the processing of long novel compound words, phological decomposition route operate in tan- the pattern of results was taken to suggest a two- dem. Bertram and Hyönä (2003) posited that stage process. During the first stage, lexical ac- word length modulates the interplay between the cess is achieved for the compound word constit- two access routes. When a compound word is uents. During the second stage, the meaning of sufficiently short so that all or most of its letters the novel compound word is composed out of the fall on the foveal region when fixating it during constituent meanings. The second stage is as- reading, the holistic route gets a head start and sumed to take longer when the frequency of the completes faster than the morphological route first constituent is low, because the prototypical and thus the word is more likely to be identified relationships that the low-frequency first constit- as a whole. On the other hand, when a compound uent would be engaged in compounding are not word is so long that a subset of letters is beyond firmly established. foveal reach, the identification is initiated by first In the present study, we further investigated recognizing the initial constituent followed by the processing of novel and lexicalized Finnish the recognition of the second constituent and that two-noun compound words. This time we ma- of the whole word. nipulated the frequency of the second constituent In their study examining the processing of (the compound head). It was done separately for novel compound words, Pollatsek et al. (2011) existing and novel compound words. Moreover, demonstrated that the decomposition route we also manipulated the length of the compound played even a more prominent role in processing words. If indeed word length strongly determines novel than lexicalized compound words. Pol- the interplay between the holistic and decompo- latsek et al. (2011) compared the processing of sition route in compound word identification, as novel and existing Finnish compound words by argued by Bertram and Hyönä (2003), the ma- manipulating the frequency of first constituent as nipulation of the second-constituent frequency an independent word, separately for long (aver- tapping into the decomposition process should age length of 13 letters) existing and novel com- result in different types of processing especially pound words. The length of the first constituent for short existing versus novel compound words. as well as the frequency of the second constituent Short existing compound words are more likely was matched across conditions. For first fixation be identified by the holistic route, whereas short duration, which indexes early effects in word novel compound words have to be processed via processing, an effect of first-constituent frequen- the morphological decomposition route. For long cy was observed that was similar in size for ex- compound words, on the other hand, the manipu- isting and novel compound words. For gaze du- lation of the second-constituent frequency should ration (i.e. the summed duration of fixations lead to less dramatic differences between exist- made on the word before exiting to the right or ing and novel compounds, as the decomposition left) first-constituent frequency was greater for novel than existing compound words. For the Copyright © by the paper’s authors. Copying permitted for private and academic purposes. In Vito Pirrelli, Claudia Marzi, Marcello Ferro (eds.): Word Structure and Word Usage. Proceedings of the NetWordS Final Conference, Pisa, March 30-April 1, 2015, published at http://ceur-ws.org 95 route is assumed to be in operation for both word constituent frequency effect was only observed types. for short novel compounds. Adult readers read sentences silently for com- Gaze duration: In gaze duration, summing up prehension while their eye movements were reg- all fixations made during the first-pass reading, istered. The target compound words were em- the main effect of word type, word length and bedded somewhere in the middle of the sentenc- second-constituent frequency were all signifi- es. The frequency of the second constituent as a cant. Gaze duration was significantly longer for separate word was manipulated for short (7-9 novel than existing words, longer for long than letters) lexicalized (e.g., savukala = smoked fish) short words, and longer for compounds contain- and novel (e.g. hymykisa = smile contest) com- ing a low-frequency than high-frequency second pounds as well as for long (12-16 letters) lexical- constituent. Similarly to second fixation dura- ized (e.g., hiekkapaperi = sand paper) and novel tion, gaze duration also revealed a reliable three- (e.g., skandaalivaali = scandal election) com- way interaction between the manipulated factors. pound words. Thus, the experimental design was In order to examine in more detail the interac- a 2 (low vs. high frequency second constituent) x tion, it was broken down into two separate 2x2 2 word type (existing vs. novel) x 2 word length ANOVAs, one for the short and another for the (short vs. long) within-participants design. Com- long compound words. prehensibility of the novel compound words was For the long compound words, there was a secured by a rating test conducted prior to the main effect of word type and second constituent experiment proper. Only novel compound words frequency, but no reliable interaction between whose meaning could be computed without them, suggesting that the second-constituent fre- providing any linguistic context were chosen for quency effect was of similar magnitude for exist- the study. The frequency of the first constituent ing (an effect size of 91 ms) and novel (an effect was matched across the conditions, as was the size of 111 ms) compound words. However, for frequency of the short and long existing com- short words, the Word Type x Second- pound words. Constituent Frequency proved significant. This interaction reflected the fact that the second- 2 Results constituent frequency effect was considerably greater for novel (an effect size of 155 ms) than Several eye fixation measures were used to tap for existing (an effect size of 42 ms) compound into the time course of compound word pro- words. cessing. The earliest effects were measured by Total fixation time: We also analyzed the first fixation duration. Early, but less immediate total fixation time spent reading the target words. effects were measured by second fixation dura- This measure indexes late effects; it sums up the tion and gaze duration. Still later effects were duration of all fixations made on the word during measured by total fixation time, which is the sum the first-pass and second-pass reading. In this of all fixations, both first-pass and second-pass, measure, the three-way interaction obtained for made on the target word. second fixation duration and gaze duration was First fixation duration: In the earliest stages no longer significant. However, the interaction of foveal word processing, indexed by first fixa- between word type and second-constituent fre- tion duration, no effects of novelty or second- quency was almost significant. This interaction constituent frequency were observed. reflects the fact that in total fixation time the ef- Second fixation duration: A bit later in the fect of second constituent frequency was greater processing timeline, main effects of novelty and for novel than existing compound words, regard- second-constituent frequency were obtained. less of word length. The size of the second- These effects were modified by interactions in- constituent frequency effect was 51 ms for the volving word length, including the three-way existing compounds and 151 ms for the novel interaction. This interaction was broken down by compounds. computing a separate 2x2 ANOVA for short and Summary of results: The following picture long compounds, respectively. These analyses emerges from the pattern of results presented revealed no effect of novelty or second- above. In the earliest stages of word processing, constituent frequency for long compounds, no signs of either novelty or second-constituent whereas for short compounds both main effects frequency were seen, which suggests that these and their interaction proved significant. The in- effects took some time to develop during com- teraction reflected the fact that the second- pound word identification. For long compounds, 96 these effects were still absent in second fixation also becomes active when identifying short lexi- duration but emerged in gaze duration. For short calized compounds. compounds, the effects were already visible in A possible theoretical framework that can ac- second fixation duration. Finally, the measure count for the obtained pattern of results is a dual indexing second-pass reading demonstrated a route cascade model assuming that identification greater second-constituent effect for novel than always starts out with the decomposition route existing compounds. All in all, the pattern of re- with the process quickly cascading into the holis- sults suggests that meaning composition takes tic access in the case of short compounds and place with more delay for long than short com- with some delay in the case of long compounds. pound words. When a compound word is short, its constituents are also short and may be accessed rapidly. On the other hand, when the word is long, not only 3 Conclusions the constituents are likely to be longer and may thus lengthen their access, but the morphological The present study provided further evidence for segmentation process may also need additional the view (Bertram and Hyönä, 2003, 2013), ac- time. Hence, the holistic route is activated with cording to which word length modifies the rela- some delay after the decomposition route is acti- tive role of the holistic versus the morphological vated. The suggested model may be further test- decomposition route in compound word identifi- ed by replicating the present study by manipulat- cation. The decomposition route is an integral ing the frequency of first constituent separately part in identifying long compound words, be- for long and short, novel and lexicalized com- cause holistic processing is not viable due to vis- pound words. ual acuity constraints. This became apparent in the effect of the second constituent frequency indexing access via morphological constituents being similar in magnitude for the novel and lex- icalized compound words. On the other hand, References when lexical access via the holistic route is a vi- Bertram, R., & Hyönä, J. (2003). The length of a able option, as is the case with short existing complex word modifies the role of morphological compound words that fit in the foveal area of the structure: Evidence from eye movements when eyes when the word is fixated, the novelty effect reading short and long Finnish compounds. Jour- nal of Memory and Language, 48, 615-634. emerged relatively early (during second fixation) and the second-constituent frequency effect was Bertram, R., & Hyönä, J. (2013). The role of hyphen considerably smaller for existing than novel at the constituent boundary in compound word compound words during first-pass reading. Final- identification: Facilitative for long, detrimental for ly, the second-pass reading measure demonstrat- short compound words. Experimental Psychology, 60, 157-163. ed a greater effect of constituent frequency for novel than lexicalized compounds. This may be Pollatsek, A., Bertram, R., & Hyönä, J. (2011). Pro- taken to suggest that meaning composition takes cessing novel and lexicalised Finnish compound longer when the frequency of the second constit- words. Journal of Cognitive Psychology, 23, 795- uent is low, since the typical relationships low- 810. frequency constituents are engaged in com- Pollatsek, A., Hyönä, J., & Bertram (2000). The role pounds are less firmly established. of morphological constituents in reading Finnish There are also two findings that are not com- compound words. Journal of Experimental Psy- pletely in line with the visual acuity principle chology: Human Perception and Performance, 26, proposed by Bertram and Hyönä (2003). One is 820-833. the absence of an early novelty effect for short compound words. If the holistic route is immedi- ately activated when making the first fixation on the word, there should have been a novelty effect in first fixation duration. Second, there was a 42 ms effect of second constituent frequency in gaze duration even for existing short compound words, suggesting that the decomposition route 97