PROCESSING THEMATIC FEATURES: FROM DERIVATION TO COMPOUNDING Christina Manouilidou1, Valantis Fyndanis2, Eva Kehayia3, Angela Ralli4 1

Concordia University, Montreal, QC, Canada 2

Aristotle University of Thessaloniki, Greece

3

McGill University, Montreal, QC, Canada 4

University of Patras, Greece

[email protected] [email protected] [email protected] [email protected]

Abstract. We report on a lexical decision task with constituent priming probing the word recognition of deverbal compounds with the suffix -της/τής [tis/tís] (henceforth, -της [tis]), e.g. βιβλιοσυλλέκτης [vivliosiléktis] ‘book-collector’. In these compounds, the non-head (βιβλίο [vivlío] ‘book’) can be interpreted as an argument of the head (συλλέκτης [siléktis] ‘collector’). Previous research on the lexical access of deverbal nouns with the suffix -της [tis] (e.g. συλλέκτης [siléktis] ‘collector’) has shown that the thematic features (TFs) of these nouns are not accessed. By using deverbal compounds with the suffix -της [tis], we investigated whether the presence of an explicit argument in the position of the first constituent will trigger the processing of TFs of the second constituent. Results showed that even when a potential argument is present, the TFs of the deverbal nominals with the suffix -της [tis] are not activated. This was attributed to the non-eventive nature of these nominals which may block the surfacing of their TFs.

1. Introduction Compounds can be defined as grammatical combinations of lexical items or lexemes to form new words (Dressler, 2006), such as truck + driver > truckdriver. Given its universal character as a fairly productive process of word formation, compounding has been an important part of morphological investigations. The main reason behind this, is the fact that it provides us with the unique opportunity for assessing the combinatorial mechanisms that languages regularly employ as

a means of word formation. Although derivation by affixation1 also provides a similar opportunity as it involves (application of) combinatorial rules as well, the properties of these two processes of word formation (compounding and affixation) are quite distinct. While derivation by affixation involves only one lexical item (either free-standing or bound, depending on the language), which is usually combined with a bound morpheme (suffix or prefix), usually with grammatical function, in compounding, both constituents are lexical items and are usually free morphemes. In this sense, compounds tend to be easily decomposable into their constituents, as words are easily segmentable and identifiable at the sentence level. On the other hand, in terms of storage and representation, compounds constitute one single lexical item which should be easily retrievable from the mental lexicon. This is not the case of sentences, since they are not stored as such and consequently not retrieved from the lexicon. These properties make compounds appear as being at the boundary of words and sentences and this is what makes their study highly informative for the linguistic and psycholinguistic theory. Moreover, compounding quite frequently incorporates derivation by affixation, thus providing us with the richest possible combination of processes of word formation. This is the case of compounds which include a derived word as one of their constituents, such as truck + driv-er. Thus, compounding provides us with a window to combinatorial properties of lexemes at a lexical (and not syntactic) level, reflecting some “primitive” principles of word formation. As mentioned above, apart from the purely linguistic perspective, compounding is fairly informative when studying it from a psycholinguistic perspective. Investigations on the mental lexicon aim at capturing both the common and “deviant” (subject to parametrization) properties of lexical representation and processing across languages. Primary among these investigations is the issue pertaining to what is listed in the mental lexicon, in terms of minimal units and the information encoded, and how mental representations are accessed and linked to each other. Given that compounding reflects the fundamental characteristics of morphology in language, in the sense defined in the previous paragraph, by studying the processing of compounds we tap into the fundamentals of the human innovative ability for morphological representation and processing. Seen in this light, compound words offer us a unique opportunity to understand the interaction between storage and computation in the mind, the manner in which morphological and semantic factors impact the nature of storage and the manner in which the computational processes serve the requirements of on-line language comprehension and production. Over the last two decades the psycholinguistic study of compound processing and representation has seen great progress. Jarema (2006) offers a detailed description of the current 1

There is also the derivation by conversion, although this is morphologically rather poor.

state of compound research from a psycholinguistic and neurolinguistic perspective. There have been many attempts to thoroughly describe the principles of compound lexical access and representation. There are two contrasting approaches to this issue. The first one is based on the “Maximization of Computation” principle2 and suggests that the processor looks to maximize computational efficiency by increasing the storage capacity. According to this approach, compounds are listed and retrieved as wholes, minimizing this way computational effort. The second approach is based on the “Maximization of Storage” efficiency (in Libben’s, 2006 terminology) which calls for a decompositional representation that minimizes storage resources. However, as Libben (2006) points out none of these two “extreme” views has received unequivocal experimental support. In contrast, compound processing is characterized by the principle of Maximization of Opportunity which states that morphological processing is better captured as a system that seeks to maximize the opportunities for activation at all levels of form and meaning. The processor seeks to compute and store as much as possible. Thus, Libben (2006) proposes that the activation of a compound word such as strawberry involves the activation of both constituents (straw + berry) as well as the whole word. Where such massive activation creates conflict between the meanings of the compound constituents and that of the whole word3, a set of post-activation ‘cleanup’ procedures resolves this conflict by deactivating spurious representations (Libben, 2006: vii). Libben’s discussion of compound processing deals with the activation of morphological and semantic properties of the constituents and of the whole compound. In the present study, we seek to investigate to what extent the processor computes grammatical information incorporated in the compound. A theoretical framework such as the Maximization of Opportunity would predict activation at all possible levels. Although Libben (2006) does not mention activation and processing of the grammatical information related to the constituents of a compound, within a system of maximal activation it is possible to assume so. Adopting the general idea of the Maximization of Opportunity framework, in the present study we explore the question whether compounding might help us tap onto the processing of features that we were not able to see in the processing of derived words. The type of compounds under investigation demonstrate Theta Role Saturation (TRS) according to DiSciullo and Ralli (1999), or Argument Linking according to Lieber (1983). These 2

As phrased by Libben (2006: 5) In the case of strawberry, as in any other case of opaque compounds, Maximization of Opportunity predicts that both straw and berry would be activated as well as the whole word strawberry. Given that the meaning of strawberry is not immediately derived from the meaning of its constituents, constituent activation can only create conflict and inhibit word recognition. It is in this time point when the clean-up processes interfere and deactivate misleading representations (straw and berry in the case of strawberry). 3

are compounds of the type βιβλιοσυλλέκτης [vivliosiléktis] ‘book-collector’, where the non-head constituent (βιβλίο [vivlío] ‘book’) appears to be an argument of the head constituent (συλλέγ-της > συλλέκ-της [siléktis] ‘collect-or’). In the present study, we are aiming at detecting evidence for the activation of Thematic Features (TFs) of the second constituent (συλλέκτης [siléktis] ‘collector’) which enable the first constituent to be interpreted as an argument of the second constituent. In a previous study (Manouilidou, 2005) we tested whether the TFs of deverbal nouns affect their lexical access. We used a variety of deverbal nouns, such as εξέτα-ση [ekséta-si] ‘examination’, πλύ-σιμο [plí-simo] ‘washing’, κατακτη-τής [katakti-tís] ‘conqueror’. The experiment compared the lexical access of deverbal nominals (e.g. καθαρισ-τής [kaθaristís] ‘cleaner’) to denominal nominals (e.g. σφουγγαρ-άς [sfougarás] ‘sponge-diver’) and non-derived ones (e.g. μανάβης [manávis] ‘grocer’). The reasoning behind the experimental design was that deverbal nominals would yield longer RTs due to the processing of their TFs in the same way as verbs demonstrate additional processing cost compared to nouns. The results revealed within category variation. In other words, processing of TFs was detected only for the deverbal nouns with the suffixes -σιμο [simo] and -ση [si]. The deverbal nouns with the suffix -της [tis] behaved in the same way as their denominal counterparts, suggesting that the processing of their TFs was not detected during lexical access. This discrepancy was attributed to the possibility that TFs are visible during lexical access only for those nominals with increased event/verbal properties. Notice that unlike the nominals with the suffix -ση [si] and -σιμο [simo], the nominals with the suffix -της [tis] do not belong to this category (Alexiadou, 2001)4. 4

What determines the event/verbal properties of a nominal is the number of modifications, which typically occur with verbs, that this nominal accepts. For instance, those formed with –της (tis) are the least ‘verblike’, since apart from referring to animate entities they do not permit manner adverbials (example 1a), cannot bear aspectual modification (example 1b) and do not take by-phrases in a sentential context (example 1c). Conversely, these nominals create subject nominalizations (e.g. χορεύω [χorévo] ‘(I) dance’ > χορευτής [χoreftís] ‘dancer’) Thus, we could say that –της (tis) nominals demonstrate a diminished verbal character. (1)

a. Ο καθαριστής του κτηρίου [O kaθaristís tu ktiríu the cleaner theGEN buildingGEN ‘the cleaner of the building carefully’

*προσεκτικά. prosektiká]. carefully

b. Ο καθαριστής του κτηρίου [O kaθaristís tu ktiríu the cleaner theGEN buildingGEN ‘the cleaner of the building for a month’

*επί ένα μήνα epí éna mína] for a month

c. Ο καθαριστής του κτηρίου *από το Γιάννη [O kaθaristís tu ktiríu apó to Jáni] the cleaner theGEN buildingGEN by the Jiannis ‘the cleaner of the building by Jiannis’

The above finding was discussed in Manouilidou (2005) in terms of the mental representation of these nominals and their lexical access. Although the class of deverbal nominals appears to be homogeneous in terms of their mental representation (every deverbal nominal carries thematic information), their lexical access appears varied. It was suggested that if TFs are processed only for those deverbal nominals with increased verbal properties, then this further implies that TFs are not an immediately accessible characteristic for every deverbal nominal, but rather only for those with an increased eventful character. In the light of the similar RTs obtained for deverbal nominals with diminished verbal character and their denominal controls, Manouilidou (2005) surmised that the processing of TFs never materializes for those deverbal nominals with a diminished verbal character. In other words, their recognition takes place before the thematic component of their representation can be accessed. If this is the case, then we have an indication that not all features of a word are immediately accessible, with some features possibly never being accessed. It further suggests that the accessing of lexical features occurs in stages5, with some of them being accessed in earlier stages of processing and others later on, if at all. Since the previous study failed to detect processing of TFs of deverbal words with the suffix -της [tis] suggesting that TFs are not a salient feature of their mental representation, and since the meaning of compounds with TRS as well as the syntactic restrictions applying to them6 call for their non-head constituent to be interpreted as an argument of the head constituent, in the present study our principle goal is to investigate whether the presence of an explicit argument in the position of the first constituent will trigger the activation and processing of the TFs of the second constituent. In other words, we are asking whether the activation and processing of the non In contrast, the nominals with the suffix –ση [si] and –μα [ma]/-σίμο [símo] create process nominals (αναπτύσσω [anaptíso] ‘(I) develop’ > ανάπτυξη [anáptiksi] ‘development’, φιλώ [filó] ‘(I) kiss’ > φίλημα [fílima] ‘(the) kiss’, κλέβω [klévo] ‘(I) steal’ > κλέψιμο [klépsimo] ‘stealing’), and permit by-phrase and aspectual modification[0] (examples 2 and 3, respectively). (2)

Η ανατίναξη της γέφυρας από το στρατό [I anatínaksi tis géfιras apó to strató] the blowing theGEN bridgeGEN by the army ‘the blowing up of the bridge by the army’

(3)

Το συχνό πλύσιμο των πιάτων από το Γιάννη [To sixnó plísimo ton pçáton apó to Jáni] the frequent washing theGEN dischesGEN by the Jiannis ‘the frequent washing of the dishes by Jiannis’ (Alexiadou, 2001)

5

Staged lexical access is well documented in the literature (e.g. Rastle et al., 2000; Tsapkini et al., 2002) For instance, combining these compounds with a “syntactic” complement or a by-phrase (agent) often results in ungrammatical utterances: e.g. *βιβλιοσυλλέκτης βιβλίων [vivliosiléktis vivlíon] ‘book-collector of books’, *ηλιοκαμένος από τον ήλιο [iλokaménos apó ton íλo] ‘suntanned by the sun’. 6

salient TFs of the second constituent (-κριτής [-kritís] ‘reviewer’) depend on the presence of an explicit argument within the compound, which in Greek compounds is realized by the first constituent (βιβλίο- [vivlío] ‘book’). 2. Experiment 2.1 Method Experiment, procedure and design To address the above questions, one on-line lexical-decision task within a constituent priming paradigm was administered to Greek-speaking individuals. In particular, in this experiment the compound acted as prime for the deverbal second constituent. This way, we had a direct comparison between the lexical decision times of the present priming experiment and the previous lexical decision task (Manouilidou, 2005). The experiment was administered on a Macintosh computer using the PsyScope (Version 1.2.5) program. The stimuli were presented in the center of the computer screen in bold 36-point lowercase SuperGreek. The prime appeared for a period of 150 ms, followed by a pause of 100ms, which was in turn followed by the target (unlimited time)7. At the outset of the experiment participants were told that they would see single words presented in the centre of the screen and were instructed to press the “yes key” (a key that by convention was defined as representing the “yes-answer”) if they had ever seen the word before and to press the “no key” (a key representing the “no-answer”) if they had never seen the word before. Instructions were followed by a training block of 10 trials and by 5 pretest items which were omitted from the analysis. The independent variables were prime and target type (for stimuli types see Materials) and the dependent variables were reaction time (RT) and accuracy of response. Participants Twenty-seven native speakers of Modern Greek (16 female and 11 male), 18-47 years old (average age: 31.5 years), with 12-23 years of education (average education: 12 years) participated in the experiment. Of these participants, 27 were right-handed and one was left-handed. Materials The stimuli comprised i) 111 experimental items (37 compounds that were used as primes and 74 targets, half of which were the second constituents of the compounds and half their controls) which 7

Hence, the Stimulus Onset Asynchrony (SOA) was set to 250 ms.

formed 74 prime-target pairs, ii) 184 fillers (102 nouns and 82 verbs) that formed 92 prime-target pairs (51 noun-noun and 41 verb-verb) and 128 nonwords (NW) which, preceded by 128 real words (non compounds), formed 128 prime-target (NW) pairs. Hence, overall the stimuli comprised 294 prime-target pairs. The NW were created by changing the first phoneme/grapheme of existing words. All NW were phonotactically legal forms. Finally, both members of the “filler” pairs matched their experimental counterparts for length. The set of compounds included i) 10 noun-noun (NN-[tis]) compounds8 with TRS, such as βιβλιοσυλλέκτης [vivliosiléktis] ‘book-collector’, ii) 9 adjective-noun (AN-[tis]) compounds without TRS such as καλοδουλευτής [kaloδuleftís] ‘strenuous-worker’, iii) 9 NN compounds with non-derived second constituent, which featured stress change (SC), such as αχλαδόκαμπος [aχlaδόkabos] ‘pear-plain’, and iv) 9 NN compounds with non-derived second constituent and without featuring stress change (non stress change -NSC), such as φρουτοχυμός [frutoçimόs] ‘fruit juice’. Εxamples of the experimental trials across conditions can be seen in Table 1. All compounds were controlled for grammatical gender, morphological and semantic transparency. Furthermore, compounds’ second constituents and their controls were matched for frequency of occurrence9, gender, morphological complexity (derived vs. non derived forms), and animacy. Finally, attempt was made for their matching for number of syllables and phonotactic properties. Table 1: Examples of experimental trials across conditions

Prime βιβλιοσυλλέκτης [vivliosiléktis] “book-collector”

Target συλλέκτης [siléktis] ‘collector’ αγοραστής [aγοrastís] ‘buyer’ (control)

καλοδουλευτής [kaloδuleftís] “strenuous-worker”

δουλευτής [δuleftís] ‘worker’ γλύπτης [γlíptis] ‘sculptor’ (control)

αχλαδόκαμπος [aχlaδόkabos] “pear-plain”

κάμπος [kábos] ‘plain’ βάλτος [váltos] ‘swamp’ (control)

φρουτοχυμός [frutoçimόs] “fruit juice”

χυμός [çimόs] ‘juice’ πηλός [pilόs] ‘clay’ (control)

8

The NN-[tis] compounds comprise items stressed on the ultimate syllable, such as καπνοκαλλιεργητής [kapnokalierjitís]‘tobacco cultivator’ and also items stressed on the penultimate syllable, such as κατσικοκλέφτης [katsikokléftis] ‘goat thief’. The stress difference between the two types of compounds was deemed not to affect their lexical access and for this reason no attempt was made to divide the stimulus set in two according to the stress pattern of the compounds. 9 Frequency measurements were obtained from the Institute of Language and Speech Processing corpus (Hatzigeorgiou et al, 2000).

2.2 Hypotheses/predictions Out of context lexical access of deverbal nominals with the suffix –της [tis], e.g. συλλέκτης [siléktis] ‘collector’ did not demonstrate processing of their TFs. This was attributed to the diminished eventive/verb-like character of these nominals and their non salient thematic component. The reasoning behind the present study was to trigger the processing of TFs of deverbal nouns with the suffix –της [tis] by creating a favorable environment, that is an environment were a potential argument of the –της [tis] nominal is present, e.g. βιβλιοσυλλέκτης [vivliosiléktis] ‘book-collector’. It was deemed appropriate to use compounding, given its nature and its resemblance to sentential environment. We assumed that since the first constituent should be interpreted as an argument of the second constituent, this would mean that the processor has to access the TFs of the nominal. Thus, we predicted processing of TFs for the second constituent which would be reflected in longer RTs for the lexical decision of the –της [tis] nominals when primed by a compound with TRS. This is the case of NN-[tis] compounds, such as λογοθεραπευτής [loγoθerapeftís] ‘speech pathologist’ > θεραπευτής [θerapeftís] ‘therapist’. In the case of AN-[tis] without TRS, such as μικροεπενδυτής [mikroepenδitís] ‘minor investor’, there is no potential argument and, therefore, there is no good reason to expect that the not salient TFs of the second constituent would be activated. Thus, in terms of raw RTs, we predict longer RTs for the recognition of the –της [tis] nominals when primed by a compound with TRS followed by RTs for the –της [tis] nominals when primed by a AN-[tis] compound, without TRS, followed by RTs for the second constituent of the NN compounds with a non-derived second constituent, either with of without stress change. This is demonstrated in Table 2. When it comes to priming effects there are two parameters that we should take into account. The first one was developed in the previous paragraph and it has to do with the nonsaliency of TFs in non eventive nominals. The second parameter is related to the Stimulus Onset Asyncrony, which was set to 250ms (see section 2.1). Rastle et al. (2000) report two sets of visual priming experiments in which the morphological, semantic, and orthographic relationships between primes and targets varied in three stimulus onset asynchrony (SOA) conditions (43 ms, 72 ms, and 230 ms). Results showed distinct priming effects for each set of primes (orthographic, morphological, semantic) in different SOAs. It was shown that morphological structure plays a significant role in the early visual recognition, whereas semantic priming was observed only in long SOAs, suggesting that semantic information emerges at later stages in lexical access. Thus, a SOA of 250 ms, as it is in our case, although it would be inadequate to activate the hidden TFs and the process of TRS, it would be enough to highlight the semantic relationship of the first and the second constituent (argument and deverbal nominal) thus paving the way for a full activation of the

target and its features, TFs included10. Thus, we predict stronger priming for NN-[tis] as opposed to AN-[tis], and NN-SC and NN-NSC compounds. A firm activation of the target has to do with activation of more information by the prime. The more information is activated with the prime, the easier the recognition of the target. Table 2: RTs predictions

1. βιβλιοσυλλέκτης [vivliosiléktis] ‘book collector’

συλλέκτης [siléktis] ‘collector’

2. μεγαλοεπενδυτής [meγaloepenδitís] ‘big investor’

επενδυτής [epenδitís] ‘investor’

3. χωματόδρομος [xomatóδromos] ‘dirt road’

δρόμος [δrómos] ‘road’

φρουτοχυμός [frutoximós] ‘fruit juice’

χυμός [çimós] ‘juice’

2.3 Results We performed separate analyses for raw RTs on word recognition and priming effects. Erroneous responses and RTs above and below two standard deviations above or below the mean were omitted from statistical analyses. No item and no participant were excluded from the analysis. Mean RTs calculated across subject means as well as mean differences (priming effects) are presented in Table 3. The raw RTs and the mean differences were submitted to planned comparisons, using two-tailed t tests. Paired samples tests revealed significant priming effects for the second constituent when primed by NN-[tis] compound (t = 2.72, df = 26, p = 0.011), by NN-SC compound (t = 2.49, df = 26, p = 0.019), by NN-NSC compound (t = 3.37, df = 26, p = 0.002), and marginally significant effects when primed by AN-[tis] compound (t = 1.83, df = 26, p = 0.07). In terms of raw RTs in the second constituent, two-tailed t tests showed that lexical decisions on the second constituent of NN-[tis] and AN-[tis] yielded longer RTs than lexical decisions on the second constituent of NNSC and NN-NSC compounds; these differences reached statistical significance (t = 2.05, df = 26, p = 0.023)11. However, there was no difference between the raw RTs on the second constituent of NN-[tis] and AN-[tis] (t = 2.05, df = 26, p = 0.98).

10

This is in accordance with the Maximization of Opportunity hypothesis. The difference between RTs on the second constituent of the compounds with –της [tis] and those without could also be attributed to a frequency difference, since the mean frequency for second constituent of the NN compounds was higher than the mean frequency of the derived words with –της [tis] that served as second constituent for the AN-[tis] and NN-[tis] compounds (0.003905‰ vs. 0.031237‰). 11

Table 3: Priming Results Priming

Target RT in ms

NN-[tis] 687.9 AN-[tis] 689.8 661.3 NN-SC 649.3 NN-NSC ªPriming significant at p<.05 b

Control RT in ms

Mean Difference (Priming)

734.7 720.6 702.2 702.5

46.8ª 30.8b 40.9ª 53.1ª

Priming significant at p<.07

3. Discussion The aim of the present study was to investigate whether the presence of an explicit argument in the position of first constituent would trigger the activation and processing of TFs of the second constituent. In particular, we predicted that when a derived word with the suffix -της [tis] is primed by a compound with TRS, this would activate its TFs yielding longer RTs. Thus, we predicted a difference in raw RTs and in mean differences between the –της [tis] nominals when primed by a NN-[tis] nominal and when primed by a AN-[tis] compound. The results showed that, in terms of priming, there was no difference between targets primed by compounds with TRS and without TRS. Similar priming effects were observed for all types of compounds, when acted as primes for their second constituents12. Also, in terms of raw RTs in the lexical access of the second constituent, the presence of an explicit argument in the form of compound with TRS did not seem to trigger activation of TFs since we did not observe increased processing load of the second constituent compared to lexical decisions on –της [tis] nouns when primed by a compound without TRS (e.g. AN-[tis], such as καλοδουλευτής [kaloduleftís] ‘strenuous worker’). In other words, even in a favorable environment (when a potential argument is present), the TFs of deverbal nominals with the suffix –της [tis] are not activated. In contrast, the fact that the –της [tis] second constituent yielded similar RTs when primed by a compound with TRS (NN-[tis]) and by a compound without TRS (AN-[tis]), allows us to think that it is the presence of the derivational suffix and the complex structure of the second

12

An issue that needs to be addressed is the slightly weaker priming obtained when the second constituent was primed by AN-[tis] compounds. Comparing the priming effects obtained for the second constituents when primed by two categories of compounds (with and without TRS), we could claim that the presence of TRS increased priming effects, confirming our hypothesis for activation of TFs. However, the difference is not statistically strong (p = 0.07). Moreover, the lack of difference between the priming effects of the second constituent when primed by a compound with TRS (βιβλιοσυλλέκτης [vivliosiléktis] ‘book collector’ > συλλέκτης [siléktis] ‘collector’) and when primed by a NN compound without TRS (μαντρότοιχος [mandrótixos] ‘fence wall’ > τοίχος [tíxos] ‘wall’), does not allow us to make any strong claims about stronger priming effects due to TRS.

constituent that increased the processing load13, as demonstrated in the difference between raw RTs on the second constituent of a –της [tis] compound (e.g. συλλέκτης [siléktis] ‘collector’, δουλευτής [duleftís] ‘worker’) and the second constituent of a NN compound (e.g. μαντρότοιχος [mandrótixos] ‘fence wall’ > τοίχος [tíxos] ‘wall’). Taking into account the results of the previous on-line experiment on the lexical access of deverbal nominals with –της [tis] (Manouilidou, 2005) and the results of the present study, it seems that the TFs of the -tis nominals are not activated during their lexical access. The reason for this lack of activation could be attributed to the specific nature of the suffix -της [tis] and its derivatives. As we saw, the specific suffix creates nominalizations which have reference to an animate entity and not to an event, as opposed to other nominalizations with the suffixes -ση [si] (ανατίναξη [anatínaksi] ‘explosion’) and –σίμο [simo] (πλύσιμο [plísimo] ‘washing’) tested by Manouilidou (2005). Thus, it is not the case that the presence of an explicit argument activates the TFs of the second constituent as we originally predicted. In contrast, it seems that it is the nature of the whole compound that calls for an eventive or non eventive interpretation and the saliency or non-saliency of their TFs, which in turn would trigger or not processing of TFs. Compounds and derived words with the suffix -της [tis] behave as non-deverbal derived words (in the sense that they do not exhibit an eventive character) and their processing load stems only from their complex structure. The results of the present study have implications for the current theory of compound processing put forward by Libben (2006), Maximization of Opportunity. According to Libben’s Maximization of Opportunity principle, we would expect opportunities of activation at all possible levels, the level of TRS included. However, this is not the case for the NN-[tis] compounds examined in the present study. Thus, it seems that the principle of Maximization of Opportunity might not apply to all the possible features of compounds, especially to non-salient features such as TFs and that what is described as TRs has no processing correlate with all the suffixes that create deverbal nominals or compounds. The priming effects show that constituents are activated for every type of compound, not allowing us to differentiate priming effects of TRS. Consequently, one might claim either that Maximization of Opportunity “cannot” reach the level of non salient features or that the activation and processing of TFs is conditioned by the nature of the specific nominal14. Nominals that refer to entities and not to events seem to lose their deverbal character, 13

That complex morphological structure increases processing load is widely reported in the literature (e.g Schreuder & Baayen, 1995). 14 Alternatively, one could say that the present results are in accordance with the Maximization of Opportunity hypothesis, in the sense that we have activation of TFs of the second constituent, both when primed by a compound with TRS (e.g. βιβλιοκριτής [vivliokritís] ‘book reviewer’ > κριτής [kritís]

thus the non-activation of their TFs. They cease to trigger thematic operations, and they act as plain derived words. Thus, it appears that the TFs of nominals in -της [tis] are inactive, and they do not get activated even when primed by a formation which requires the presence of an explicit argument. 4. Conclusion The goal of the present study was to tap into the processing of TFs of deverbal nouns with the suffix -της [tis] by priming their word recognition by a deverbal compound which requires TRS. The experimental results failed to show processing of TFs, since there was no difference between RTs on -της [tis] nominals primed by a compound with TRS and primed by compounds without TRS. In contrast, what seemed to make a difference is the complex morphological structure. The results of the present study together with previous results on the lexical access of -της [tis] nouns allows us to make the claim that the specific nominals have lost their deverbal character, in the sense that their TFs have become inactive. What seems to take precedence in the process of their lexical access is their general semantic properties and more specifically the fact that they refer to an entity and not to an event. The validity of the present conclusions could be reinforced by future research examining the lexical access of the deverbal nominals with the suffixes with the suffixes -ση [-si] and -μα/σίμο [-ma/-simo] when primed by the relevant compounds, e.g. χρονομέτρηση [xronométrisi] ‘timing’15 > μέτρηση [métrisi] ‘counting’, χειροφίλημα [çirofilima] ‘hand kissing] > φίλημα [filima] ‘kissing’. If indeed the processing of TFs of a deverbal nominal depends on the eventive/non-eventive character of the nominal and not on the presence of an explicit argument or the processing of TRS, then we would expect processing of the TFs of the deverbal second constituent both when primed by a compound with TRS χρονομέτρηση [xronométrisi] ‘timing’16 > μέτρηση [métrisi] ‘counting’ and when primed by a compound without TRS (αγριοκοίταγμα [aγriocítaγma] ‘frowning’17 > κοίταγμα [cítaγma] ‘look’).

‘reviewer’) and when primed by a compound without TRS (καλοδουλευτής [kaloduleftís] ‘strenuous worker’ > δουλευτής [duleftís] ‘worker’). In this case, one could claim that it is the environment of a compound that activates the TFs of the second constituent, independently of the presence or not of an explicit argument. However, in the present study, we have no evidence for exclusively attributing the difference between the -tis nominals (κριτής [kritís] ‘reviewer’) and the non-derived ones (σιταρόσπορος [sitarosporos] ‘wheat seed’ > σπόρος [sporos] ‘seed’) to the activation of TFs and not merely to the complex morphological structure of the –της [-tis] nominals. Thus, this interpretation would require comparisons that the present stimulus set is not appropriate for and for this reason we cannot adopted it. 15 Literally, ‘time counting’. 16 Literally, ‘time counting’. 17 Literally, ‘wild look’.

References Alexiadou, A. (2001). Functional Structure in Nominals: nominalization and ergativity. Linguistik Aktuell. Amsterdam/Philadelphia: John Benjamins Publishing Company. Dressler, W. (2006). Compound Types. In G. Libben and G. Jarema (eds.), The Representation and Processing of Compound Words. New York, NY: Oxford University Press. Hatzigeorgiou, N., Gavriilidou, M., Piperidis, S., Carayannis, G., Papakostopoulou, A., Spiliotopoulou, A., Vacalopoulou, A., Labropolou, P., Mantzari, E., Papageorgiou, H. & Demiros, I. (2000). Design and implementation of the online ILSP corpus. In Proceedings of the Second International Conference of Language Resources and Evaluation (LREC), Vol. 3. Athens, Greece. 1737-1740. Jarema, G. (2006). Compound representation and processing: a cross-language perspective. In G. Libben and G. Jarema (eds.), The Representation and Processing of Compound Words. New York, NY: Oxford University Press. Libben, G. (2006). Why study compound processing? An overview of the issues. In G. Libben and G. Jarema (eds.), The Representation and Processing of Compound Words. New York, NY: Oxford University Press. Manouilidou, C. (2005). 'Computation of thematic features in deverbal nouns: probing the effect of the verbal root'. In C. Gurski (ed.) Proceedings of the 2005 Canadian Linguistics Association Annual Conference, 12 pages. Available at http://ling.uwo.ca/publications/CLA-ACL/CLA-ACL2005.htm Rastel, K., Davis, M., Marslen-Wilson, W. & Tyler, L. K. (2000). Morphological and semantic effects in visual word recognition: A time-course study. Language and Cognitive Processes 15(4/5): 507-537. Schreuder, R. & Baayen, H. (1995). Modeling Morphological Processing. In L.B. Feldman (ed.) Morphological aspects of language processing. Hillsdale, NJ: Lawrence Erlbaum Associates, pp. 131-153. Tsapkini, K., Jarema, G. & Kehayia, E. (2002). Regularity Revisited: Evidence from Lexical Access of Verbs and Nouns in Greek. Brain and Language 81: 103-119.