Stem and fragment priming on verbal forms of Italian Alessandro Laudanna Giulia Bracco University of Salerno University of Salerno Via Giovanni Paolo II, 132 - Via Giovanni Paolo II, 132 - 84084 - Fisciano (SA) 84084 - Fisciano (SA) alaudanna@unisa.it gcbracco@unisa.it Abstract verbal families have orthographically similar English. In this paper we investigate the respective members, but these words often become ‘unique’ roles of orthographic and morphological structure in only at the end, since information about mood, the processing of Italian verbal forms by using the tense and number is carried by affixes in the final masked priming paradigm. part of the word. The N-count is not the best meas- According to the morphology-based view, in a priming ure to describe their relatedness. On the other condition the recognition of an inflected word should hand, morphological parsing accounts (e.g., be facilitated by the presentation of the stem. The co- hort model, instead, postulates that the orthographic Baayen, Dijkstra, and Schreuder, 1997; Burani material from the word’s onset to the uniqueness point and Caramazza, 1987; Colé, Beauvillain, and Se- should be sufficient for the activation of the morpho- gui, 1989; Taft, 1979) support the argument that logical family. the stem is the key of access to the lexicon. Mor- phological priming shows that, even though mor- Italiano. In questo lavoro indaghiamo il ruolo della phologically related pairs share some ortho- struttura ortografica e della struttura morfologica graphic material (verbs sharing the stem also nella elaborazione delle forme verbali dell’italiano, share letters in initial position, unless they are pre- usando il paradigma del priming mascherato. Secondo fixed), the role played by the morphological struc- i modelli basati sulla morfologia, in una condizione di ture is different from the one played by the ortho- priming il riconoscimento di una forma flessa do- vrebbe essere facilitato dalla preventiva presentazione graphic structure (Pastizzo and Feldman, 2002). della radice. I modelli coorte, invece, propongono che In the following experiments, we used the priming il materiale ortografico dall’inizio della parola fino al paradigm to preactivate both the morphological punto di unicità sia sufficiente per attivare la famiglia and the orthographic keys of access. Many studies morfologica. employing the priming paradigm used an ortho- graphically similar baseline, while others (e.g., 1 Introduction Feldman and Soltano, 1999; Marslen-Wilson, Ty- According to the cohort models of visual word ler, Waksler, and Older, 1994) employed ortho- recognition (Johnson and Pugh, 1994), all the graphically and morphologically dissimilar base- sources of information that contribute to the iden- lines, and others, finally, (Giraudo and Grainger, tification of a target word proceed from left to 2000; Grainger, Colé, and Segui, 1991) estimated right: a single word of the cohort becomes unique morphological facilitation in comparison with at the uniqueness point, when it remains the only both types of baselines. To our knowledge, no candidate corresponding to the orthographic con- study employed the stem priming in Italian, by us- figuration of the stimulus. At that point the recog- ing both an orthographic and a morphological nition takes place. Cohort models use the neigh- baseline. We use the term ‘stem’ to denote the re- borhood size and the frequency distribution of sidual part of the word when all inflectional af- neighbors as predictors of the competition be- fixes are removed and that can, or cannot be, com- tween candidates and of the consequent recogni- plex. In contrast, the root is not analysable (Bauer, tion latencies. The same models not always deal 1988). The stem does not end in a vowel, and if with the internal morphological structure of the presented in isolation, it appears as an incomplete word (but see Marslen-Wilson, 1987). In Italian, word, an orthographic fragment, even if it carries lexical information. By this token, this feature could also turn out to be a benefit: studies employ- to contact the lexicon. According to the morphol- ing the priming paradigm usually have to deal ogy-based view, in a priming condition the recog- with the objection that the base form is a word. nition of an inflected word should be strongly fa- The stem primes provide an abstract lexical infor- cilitated by the presentation of the stem. Accord- mation in a non-lexicalized form. In order to as- ing to the cohort model, instead, the orthographic certain if the initial string of letters up to the material from the word’s onset to the uniqueness uniqueness point is the orthographic access code, point should be sufficient for the activation of the in Experiment 1 we used Italian verbs with the morphological family. uniqueness point occurring before the stem (e.g., ‘ABBAN’ is a fragment present only in the mor- 2.1 Method phological family of ‘abbandonare’, to abandon. Stimuli We selected 16 inflected forms of verbs All the words starting with the fragment ‘AB- with a ‘unique’ initial fragment (e.g., ‘abban- BAN’ (e.g. ‘abbandonai’, I abandoned, ‘abban- donare’, to abandon), which served as targets in donato’, abandoned) belong to the same morpho- three different experimental conditions: A) logical family). Then, in Experiment 2, we started primed by the stem, (e.g., ‘ABBANDON’); B) from the consideration that, when the initial frag- primed by the initial fragment up to the unique- ment before the stem is shared by more than one ness point (e.g., ‘ABBAN’); C) preceded by an or- family, only a non homographic stem is the true thographically unrelated fragment which shared ‘uniqueness point’: we used verbal forms with the no letter with the prime (e.g., ‘COTRU’). Mean uniqueness point only at the stem boundary (e.g., values for length were 6.9 letters for Stems and the fragment ‘DISTRI’ is shared by two morpho- 5.3 for Unrelated Prime and Fragments; prime- logical families: ‘distribuire’, to distribute, and target orthographic overlap was 67% in Stem ‘districare’, to unravel, whose stems, ‘DISTRIB’ Condition and 55% in Fragment Condition. Tar- and ‘DISTRIC’, respectively, have no homo- get mean frequency was 13; root frequency was graph). The masked priming paradigm was used 462 and initial stem cohort frequency was 245. in order to avoid intuitions or response strategies Three hundred eighty-four items were included in in participants (Forster and Davis, 1984; Forster, the list as fillers. One hundred eighty-four were Davis, Schoknecht and Carter, 1987): this tech- words, (40 adjectives, 106 nouns, 38 inflected ver- nique avoids the overt detection of any relation bal forms). Those words, together with those in between prime and target. Moreover, it has been the experimental list, displayed a distribution sim- argued that lexical decision latencies associated ilar to the one of written Italian (see CoLFIS, Ber- with masked priming also reflect the organization tinetto et al., 2005). The filler words were of the lexicon in the mind, rather than representing matched with experimental targets for their mean the mechanisms directly involved during single length in letters and for their surface frequency. words access (Baayen, 2014). In Experiment 2 co- The list included two-hundred items as horts defined by the fragment and by the stem had pseudoword targets. The whole list was composed different frequency distribution, with fragments of 200 words and 200 pseudoword targets pre- matching the initial part of lower or higher fre- ceded in turn by 100 existing primes and 100 non quency morphological families. By this token, on existing primes. the one hand we expected to detect the morpho- Participants Fifty-four participants, all logical vs. orthographic nature of the key(s) of ac- students of the University of Salerno, and native cess to the lexicon; on the other hand, we indi- speakers of Italian, took part into the experiment. rectly tested the stem frequency effect. They served for a session lasting about 40 minutes. The whole experiment was arranged in 2 Experiment 1 three different sessions and each session con- Italian verbs such as ‘abbandonare’ (to abandon) tained all the targets in one of the three experi- or ‘scivolare’ (to slip) contain fragments (‘AB- mental condition (either preceded by the frag- BAN’ and ‘SCIVO’), that, although shorter than ment, or preceded by the stem, or preceded by the the respective stems ‘abbandon’ and ‘scivol’ , can unrelated fragment). Each participant was submit- be considered ‘morphological uniqueness points’, ted to a single experimental session, for a total of because they belong to just one morphological 18 ‘superparticipants’. Each superparticipant was family. Those stimuli are relevant to decide composed of 3 participants, and constituted one whether the stem is the necessary key of access to data point in the statistical analyses. lexical information, or the fragment is sufficient Equipment Response box, connected to with the target word than fragments. The ANOVA an IBM PC running the E-Prime 1.1 software on error data did not reveal any significant result (Version 1.1). (ANOVA by participants F(2,34)=.50; p>.6; Procedure Participants had to press the ANOVA by item F(2,30)=.66; p>.5). button corresponding to their dominant hand for The ANOVA on response latencies the decision ‘word’, and another one for the deci- showed a main effect of prime type only in the sion ‘non word’. When the participants reached analysis by participants (F(2,34)= 7.01; p<.002; the 70 % of correct responses in a practice session, ANOVA by item F(2,30)=1.80; p>.2). Post-hoc the experiment started. All the stimuli appeared in analyses based on the ANOVA by participants Courier New font, 18 point size in the centre of showed a significant difference between the con- the computer screen. The fixation was 51 ms, fol- ditions ‘Fragment’ vs. ‘Unrelated Fragment’ lowed by a 51 ms pause. Primes appeared for 51 (p<.002) and between the conditions ‘Stem’ vs. ms, followed by a 12 characters backward mask ‘Unrelated Fragment’ (p<.002), but not between ############ (150 ms). The targets remained on ‘Fragment’ vs. ‘Stem’ (p >.9). The results are in- the computer screen for a maximum of 1 second. consistent with predictions of morphologically- If the participants did not produce any answer based view: the orthographic uniqueness point is within 1 second, the feedback ‘Fuori tempo’ (Out sufficient to contact lexical information. of time) appeared on the screen. The reaction times (RT) were measured from target’s onset to 3 Experiment 2 subject’s response, and the lack of a response was In Experiment 2 we selected fragments (e.g., DIS- scored as an error. TRI) shared by verbal families with different fre- quencies (e.g. ‘districare’, to unravel, lower fre- 2.2 Results and Discussion quency, LF, and ‘distribuire’, to distribute, higher In Table 1 the mean reaction times and percentage frequency, HF). Both the accounts (orthographic of errors are shown. Table 2 shows the size of and morphological) suggest that the fragment is Stem and Fragment Priming effects in response not sufficient for the activation of the morpholog- latencies and percentage of errors. For ‘size of ical family, while the stem, which is also the priming effect’ we mean the difference between uniqueness point, should determine a stronger fa- mean Reaction Times (or number of errors) in cilitation. The aim of the Experiment 2 was also Stem Condition (or in Fragment Condition) and to address the stem frequency effect. mean Reaction Times (or number of errors) in Control Condition (Unrelated Fragment Condi- 3.1 Method tion). Stimuli We selected 32 inflected forms of verbs in 16 pairs with the same initial fragment (e.g., ‘dis- Stem Unrelated tribuito’, and ‘districato’). One half of the list was Condition Fragment Fragment composed of 16 targets belonging to higher fre- Reaction quency morphological families (HF, e.g. dis- 626 ms 650 ms 626 ms Times tribuire, distributed); the other half was composed Errors 12% 15% 13% of 16 targets belonging to lower frequency mor- phological families (LF, e.g., districato, unrav- Table 1: Mean correct lexical decision latencies eled). Target frequency was 2 for LF words, 15 and percentage of errors in each priming condi- for HF words, 50 for LF Stems and 216 for HF tion. Stems; initial stem cohort frequency was 216 for HF words and 50 for LF words, root frequency Stem Priming Efffect - 24 ms (-3%) was 676 for HF words and 60 for LF words; Fragment Priming Effect - 24 ms (-2%) prime- target orthographic overlap was 72% in Stem condition and 53% in Fragment condition. Table 2: Priming effects in response latencies. In The same three different experimental conditions parentheses the effect in percentage of errors. of Experiment 2 were arranged. Three hundred sixty-eight items were included in the list as fill- As shown in Table 1, the conditions ers. One-hundred sixty-eight were words, two- ‘Fragment’ and ‘Stem’ were faster than ‘Unre- hundred items as pseudoword targets. The whole lated Fragment’ (Control) condition and they did list was composed of 200 words and 200 not differ from each other despite stems were on pseudowords targets preceded in turn by 100 ex- average longer and with more letters in common isting primes and 100 non existing primes. Participants Fifty-four participants, all and items (F(2,60)=4.42; p<.01), but no interaction students of the University of Salerno, and native (ANOVA by participants F(2,32)=1.23; p>.3; speakers of Italian, took part into the Experiment. ANOVA by item F(2,60)=1.29; p>.2). The ANOVA Each participant was submitted to a single session on RT showed a main effect of frequency in anal- (like in Experiment 1), for a total of 18 superpar- yses on both participants (F(1,17)=25.80; p<.0001) ticipants. Each superparticipant was composed of and items (F(1,30)=4.41; p<.04), no effect of prime 3 participants. type (ANOVA by participants: F(2,34)=.07; p>.9, Equipment and procedure They were the ANOVA by item: F(2,60)=.18; p>.8), and no inter- same as in Experiment 1. action (ANOVA by participants F(2,34)=1.07; p>.3; ANOVA by item F(2,60)=.15; p>.8). On average, 3.2 Results and Discussion HF targets were recognized better than LF targets (621 ms Vs. 647 ms), with faster latencies and a In Tables 3 and 4 the mean reaction times and per- lower percentage of errors (13% Vs. 24%). The centage of errors are shown. Table 5 shows the lack of priming effect for the Stem condition as size of Stem and Fragment Priming effects in re- compared with the Unrelated Fragment condition sponse latencies and percentage of errors. is the most surprising result. Post-hoc correlations were performed using main lexical and ortho- LF graphic variables as predictors, and size of stem Unrelated and fragment priming effects for RT and errors as Condition Stem Fragment criteria. The correlations on results in Fragment Fragment Reaction Times 652 ms 644 ms 647 ms condition showed a significant length effect for Errors 22% 24% 28% the fragment prime on HF words (r=-58, p<.02). More interestingly, correlations in Stem condi- Table 3: LF verbal forms: mean correct lexical tions showed that the ratio between the surface decision latencies and percentage of errors in frequency and the frequency of the stem in initial each priming condition. position is inversely correlated with the size of stem priming (r= -.36, p<.04). The higher the ra- HF tio, the faster the latencies: the “relative fre- quency” of the form in its cohort determines the Unrelated direction of the effect.The correlation was reliable Condition Stem Fragment Fragment on LF words (r=-.60, p<.01), while it was not sig- Reaction Times 615 ms 627 ms 621 ms nificant on HF words (r=.31 p>.2). The effect did Errors 14% 9% 18% not occur in the Fragment condition, and this might suggest that the effect occurs at the point where the morphological family is selected: the Table 4: HF verbal forms: mean correct lexical more frequent the cohort, the stronger the inhibi- decision latencies and percentage of errors in tion for a verbal form that has a low surface fre- each priming condition. quency. No effect of cumulative root frequency occurred when the frequency count was obtained LF HF by including words embedding the stem in any po- + 8 ms - 12 ms sition (for instance prefixed words), and this al- Stem Priming Effect (-2%) (-2%) lows us to assume that the effect is orthographic Fragment Priming + 3 ms - 6 ms in nature. We conclude that not only the word sur- Effect (+4%) (+ 3%) face frequency, but also the “relative frequency” of the word with respect to its cohort is responsi- ble for recognition. Table 5: Priming effects in response latencies. In parentheses the effect in percentage of errors. 4 General Discussion The ANOVA on error data showed an ef- Results of Experiment 1 show that when ortho- fect of frequency in analyses on both participants graphic information about initial part of the word (F(1,16)=22.33; p<.0005) and items (F(1,30)=3,91; is exhaustive, it is as reliable as stem priming, and p<.05): LF frequency words elicited higher error these results are difficult to reconcile with the rates; we also found an effect of prime type in morphologically- based view which postulates analyses on both participants (F(2,32)=5.00; p<.01) that the stem is critical for lexical access. In addi- tion, the ‘relative frequency’ effect (Experiment 2), which arises in Stem Condition, suggests that, vious data on stem priming and the “relative fre- during recognition, when a low frequency word quency” effect in Italian (Bracco and Laudanna, shares the stem with higher frequency members, 2012), suggesting that the relations between it is disadvantaged. In order to gain lexical access, words in the paradigm need to be taken into ac- the word has to sustain a harder competition with count, even if we maintain that whole word repre- other members according to their frequency dis- sentations are the keys for lexical access. tribution in the morphological family. This effect In summary, the results presented in this has been largely described for orthographic neigh- paper about the processing of Italian verbal forms borhood (Grainger, O'Regan, Jacobs, and Segui, suggest that it is performed sequentially and it 1989), but, again, it is difficult to reconcile with proceeds from left-to-right. Morphological struc- the stem frequency effect (Burani, Salmaso, and ture does not play a deterministic role, and recog- Caramazza, 1984) which states the opposite, with nition is guided by the information carried by the cumulative frequency of morphologically related initial part of the word, whether it matches a mor- words facilitating the recognition of a low fre- pheme or not. quency word. These results are in line with previ- Priming induced by ‘unique’ fragments is ous data on Italian, which failed in replicating root as reliable as stem priming. In addition, stem frequency effects (Laudanna and Bracco, 2009). priming is not explainable in terms of a stem fre- Root morpheme frequency effects are crucial in quency effect. the general issue of whether words are accessed Furthermore, the observation of a ‘rela- through decomposition rather than as full forms, tive frequency’ rather than a ‘stem frequency’ ef- since it has been widely used as the strongest evi- fect, suggests that we are tapping into a phenom- dence in favor of the hypothesis of the root mor- enon concerning connections among whole pheme representation. words. Nevertheless, the morphological parsing account is not the unique explanation for root fre- quency effects: also full listing models (see Gi- References raudo and Grainger, 2001) can provide a general Alegre, M., and Gordon, P. (1999). Frequency effects outline in which this effect is explained, for in- and the representational status of regular inflec- stance, in terms of lexical connections. This view tions. Journal of Memory & Language, 40, 41-61. has been inspected also in linguistics: Bybee (1995) proposed that the activation level of a word Baayen, R.H. (2014) Experimental and psycholinguis- tic approaches to studying derivation. In: R. Lieber is the result of lexical connections and lexical and P. Stekauer (Eds), Handbook of derivational strength, with the first one determined by the fre- morphology. Oxford: Oxford University Press, 95- quency of the lexical item, and lexical connections 117. corresponding to the pattern of weight connec- tions associated with the links between related Baayen, H., Dijkstra, T., and Schreuder, R. (1997). Sin- items. For high frequency words the individual gulars and plurals in Dutch: Evidence for a paral- lexical strength is elevated; low frequency words lel dual-route model. Journal of Memory and Lan- have a weak lexical activation and need the sup- guage, 37, 94-117. port of the activation of lexical connections. This theory is consistent with the claim that whole Bauer, L. (1988). Introducing Linguistic Morphology. Edinburgh, Edinburgh University Press. word frequency effects arise over a precise thresh- old (Alegre and Gordon, 1999). If the measure for Bertinetto, P.M., Burani C., Laudanna, A., Marconi L., these connections is the stem frequency, the more Ratti n , D., Rolando, C. and Thornton A.M. frequent the stem, the faster the recognition (Bu- (2005). Corpus e Lessico di frequenza dell’Italiano rani et al., 1984, Traficante and Burani, 2003, Co- Scritto (CoLFIS) http://alphalinguistica.sns.it/CoL- lombo and Burani, 2002). Meunier and Segui FIS/CoLFIS_Presentazione.htm (1999) found that the relative frequency of family members affects the recognition of auditory stim- Bracco G., and Laudanna, A. (2012). “Meccanismi di uli: words with high-frequency suffixed candi- competizione tra forme verbali nell’accesso al les- dates derived from the same stem were recognized sico mentale”. In: Pier Marco BERTINETTO, Va- lentina BAMBINI and Irene RICCI e Collaboratori more slowly than words with morphological fam- (a cura di). Linguaggio e cervello / Semantica, Atti ily members of a lower frequency. The effects dis- del XLII Convegno della Società di Linguistica Ita- cussed in this paper are also consistent with pre- liana (Pisa, Scuola Normale Superiore, 25-27 set- tembre 2008). Roma: Bulzoni. Volume 2. Johnson, N. F. and Pugh, K. R. (1994). A cohort model Burani, C., and Caramazza, A. (1987). Representation of visual word recognition. Cognitive Psychology, and processing of derived words. Language and 26, 240-346. Cognitive Processes, 3, 217-227. Laudanna A., and Bracco, G. (2009). “Family Fre- Burani, C., Salmaso, D., and Caramazza, A. (1984). quency, Stem Frequency and Family Size in Visual Morphological structure and lexical access. Visi- Word Recognition of Italian Verbal Forms”. In: ble Language, 18, 348-358. Proceedings of the 6th International Morphological Processing Conference. Turku, June 14-17, 2009, Bybee, J. L. (1995). Regular morphology and the lexi- Turku: Morproc, p. 63-64. con. Language and Cognitive Processes, 10, 425- 455. Marlsen-Wilson, W.D., Tyler, L., Waksler, R., and Older, L. (1994). Morphology and meaning in the Colé, P., Beauvillain, C. and Segui, J. (1989). On the English mental lexicon. Psychological Review, representation and processing of prefixed and suf- 101, 3-33. fixed derived words: A differential frequency effect. Journal of Memory and Language, 28, 1-13. Marslen-Wilson, W.D. (l987). Functional parallelism in spoken word-recognition. In U. Frauenfelder and Colombo, L., and Burani, C. (2002). The influence of L.K. Tyler, Spoken word recognition. Cambridge. age of acquisition, root frequency and context MA, MIT Press, 71-102. availability in processing nouns and verbs. Brain and Language, 81, 398-411. Meunier, F. and Segui, J. (1999) Morphological prim- ing effect: the role of surface frequency. Brain and Feldman, L.B., and Soltano, E.G. (1999). Morphologi- language, 68, 54-60. cal priming: The role of prime duration, semantic transparency and affix position. Brain and Lan- Pastizzo, M. J. and Feldman, L. B. (2002) Discrepan- guage, 60, 33-39. cies between orthographic and unrelated baselines in masked priming undermine a decompositional Forster, K. I., and Davis, C. (1984). Repetition priming account of morphological facilitation. Journal of and frequency attenuation in lexical access. Journal Experimental Psychology: Learning, Memory and of Experimental Psychology: Learning, Memory, Cognition 28, 244–249. and Cognition, 10, 680–698. Taft, M. (1979). Recognition of affixed words and the Forster, K. I., Davis, C., Schoknecht, C., and Carter, R. word frequency effect. Memory and Cognition, 7, (1987). Masked priming with graphemically re- 263-272. lated forms: Repetition or partial activation?Quar- terly Journal of Experimental Psychology, 39, 211– Traficante, D., and Burani, C. (2003). Visual pro- 251. cessing of Italian verbs and adjectives: The role of the inflectional family size. In R. H. Baayen, R. Giraudo, H. and Grainger, J. (2001). Priming complex Schreuder, Morphological Structure in Language words: Evidence for supralexical representation of Processing, Berlin, Mouton de Gruyter, 45-64. morphology. Psychonomic Bulletin & Review, 8, 127-131. Giraudo, H., and Grainger, J.(2000). Effects of prime word frequency and cumulative root frequency in masked morphological priming. Language and Cognitive Processes 15, 421-444. Grainger, J., Colé, P., and Segui, J. (1991). Masked morphological priming in visual word recognition. Journal of Memory and Language, 30, 370-384. Grainger, J., O'Regan,K., Jacobs,A., and Segui,J. (1989). On the role of competing word units in vis- ual word recognition: The neighborhood frequency effect. Perception and Psychophysics, 45, 189-195.