Directionality in the psych alternation: a quantitative cross-linguistic study

4.2.1 Areality and the structure of psych constructions

Figure 4 depicts the geographic distribution of the sample^[16] languages according to their preferred alternation type in the psych domain. As Figures 2 and 3 show, the most homogeneous group is made up of the detransitivizing languages, all of which occur in the European macro-area and belong to the Indo-European phylum (cf. Figure 4). Note however that the European macro-area is not exclusively detransitivizing (Basque, Finnish, Hungarian, Irish, Turkish) and neither is the Indo-European phylum (Irish, Persian, Marathi, Hindi). There appears to be an exceptional cluster within Indo-European languages whose geographic center is the European mainland (cf. Cysouw 2011). In contrast, neither the transitivizing set nor the underspecified one can be related to any specific macro-area (transitivizing set: four out of five macro-areas, i.e., Asia, Africa, Oceania, America; underspecified set: two macro-areas, i.e., Asia, Europe). If the mixed-transitivizing and mixed-underspecified languages are included, both types occur in five resp. four macro-areas. This typological dichotomy was also observed by Nichols et al. (2004) for their general set of verbs. Thus, the psych domain of the detransitivizing set recruits the same structural means as non-psych verbs. Our results lend further support to their observation that the presence of reduction in animate verbs (recall that our entire dataset consists of verbs belonging to Nichols et al.’s 2004 animate verbs) is negatively correlated with the availability of a causative operation. The scarcity of augmented and undirected pairs in the detransitivizing set thus plausibly falls out from this assumption, since both typically involve a causative derivation.

Figure 4:

Areal distribution of alternation directionality type in the languages of the sample.¹⁶

Strikingly, all four ‘outlier’ languages of the Indo-European phylum present with underspecified psych inventories instead. As the illustrative examples in (15) from Irish and (25)–(26) from Persian and Marathi show, the specific mechanism everywhere is auxiliary/light verb change.

(26)

Marathi

a.

plain
T-yā	māṇs-ā-lā	kiḍ-yā-ñ-c-ī	kiḷas	ā-l-ī.
dem-obl	man-obl-dat	maggot-obl-pl-gen-f	disgust	come-pst-f
‘The man got disgusted with the maggots.’

b.

induced
Kiḍ-yā-n-nī	t-yā	māṇs-ā-lā	kiḷas	āṇ-l-ī.
maggot-obl-pl-erg	dem-obl	man-obl-dat	disgust	bring-pst-f
‘The maggots disgusted the man.’

The periphrastic structure found in Persian is common and productive outside the psych domain as well (cf. Megerdoomian 2001). While Irish and Marathi also make pervasive use of light verb constructions elsewhere (Bloch-Trojnar 2010; Dhongde and Wali 2009), the expression of psych meanings is at least one highly salient function of a small number of domains the specific structures attested in our dataset appear in (cf. Adger and Ramchand 2006). The same holds for the other type of complex psych constructions, the so-called psycho-collocations (Matisoff 1986), introduced in Section 2, which are commonly found in Yucatec Maya (16) and South Efate (27) from our sample.

In our sample,^[17] languages with a (larger) proportion of complex psych constructions, i.e. constructions where either the mental state or a body part are coded as nominal constituents which alternate between subject and object function (see Section 2), are found in all five macro-areas (see Figure 5), which attests to the general psych-specific nature of these structures, rather than being an areal trait.

Figure 5:

Areal distribution of complex psych constructions in the languages of the sample.¹⁷

The overall morphological means used to effect the alternation are varied, but reveal some areal patterns. The strongest observation is the universal and global dominance of causative morphology in augmented pairs, which occurs in all macro-areas and conforms to Nichols et al. (2004). In the Americas, the induced (stimulus-oriented) alternant can take a factitive marker instead.^[18] Light verbs of various pertinent lexical origins also occur in four out of five macro-areas. The uppercase forms shown in Table 3 represent cross-linguistic translations. Where structures have been illustrated in examples in this work, references are given.

Reduced pairs most commonly occur in the European macro-area. As Table 4 shows, all morphological means which the languages in our sample use to derive the plain (experiencer-oriented) alternant are found here, with the exception of light verbs. The most common means are reflexivization and participial constructions.

Undirected pairs are the most varied group. While languages tend to generalize the use of only a few pairings of morphological means, almost all show some variability in individual items, i.e. most languages have a few lexically specified ‘outlier’ pairs whose combination of morphological means is not (or no longer) productive. The overview therefore only includes tuples that occur at least twice, either across multiple languages or within one language. For example, Icelandic evinces a few unique combinations of morphologically derived plain forms and complex induced forms with lexically specified verb-noun combinations (e.g. reiða-st anger-mid ‘get angry’ ∼ reita til reiði enrage to anger ‘anger’). Tuples in Table 5 are given in the order plain ∼ induced. Light verb constructions are the most common means for undirected pairs. Since our sample shows a large amount of variation in the possible combinations of light verbs, only the six most common means are shown for illustrative purposes. The second most frequent type is made up of pairs with no formal differentiation between the alternants. Combinations of morphological means heavily rely on the causative (or again, the factitive in the Americas) for the induced alternant, but show quite a bit of variation in the formation of the plain alternant. Interestingly, while the morphology of the induced alternant aligns neatly with purely augmented pairs (see Table 3), the most common means for reduced pairs (reflexives and participial constructions) almost never occur in undirected pairs in our sample, to the exception of a few items in Mapudungun where a factitive was found to alternate with a reflexivizer. We will return to this observation in Section 4.2.3.

4.2.2 Morphological alignment

Following the procedure in Nichols et al. (2004), we distinguish between the two major alignment types accusative and ergative, taking into account morphological case alignment in nouns and pronouns following Comrie (2013) and alignment in verbal agreement following Siewierska (2013), as depicted in Table 6.^[19] In accordance with Nichols et al. (2004: 167f) those languages where accusative alignment is dominant, i.e. either present in at least two of the three categories or the only non-neutral value in Table 6, are classified as accusative. We also follow Nichols et al. (2004: 168) in defining ergative “not as dominant alignment but as salient presence of morphological ergativity” in at least one of the three categories in Table 6, which identifies Cabécar and Basque as ergative languages. Some sample languages (Yucatec Maya, Georgian, Hindi, Marathi) possess complex alignment systems including split verbal person marking and/or tripartite or active-inactive case marking. None of these languages fits the above definition of accusative alignment. Since all of them do possess ergative morphology in at least one of the three categories in Table 6 they are identified as possessing ergative traits.

Nichols et al. (2004) report a number of (significant) correlations between the directionality types in (7) and morphological alignment, which are supported by our data. All detransitivizing languages of our sample (Spanish, Romanian, Serbian, German, and Icelandic) are accusative languages instantiating Nichols et al.’s (2004: 168) observation that the reduced type favors accusativity. Furthermore, the directed types in (7), i.e., augmentation and reduction, favor accusative alignment, which is also descriptively visible in our sample: the predominantly transitivizing languages Nafsan, Korean, Tamil, Khoekhoegowab and Malagasy are accusative languages. Finally, our sample seems to support the correlation between the undirected type in (7), which includes ‘neutral’ and ‘indeterminate’ alternations, as explained in Section 2, and ergative alignment. All languages classified as ergative or possessing ergative traits in Table 6 have either predominantly undirected pairs (Basque, Hindi) or belong to the more mixed type but with a substantial amount of undirected pairs (Georgian, Cabécar, Marathi, Yucatec Maya), see Figure 2, Figure 4. The more specific alternations present in these languages include auxiliary change (Basque, Hindi, Marathi) and double derivation (Georgian, Cabécar, Yucatec Maya^[20]).

In order to inspect the impact of alignment on the choice of alternation strategy in more detail, we contrast languages without ergative traits (i.e. the languages identified as accusative and neutral in Table 6) with languages with ergative traits (i.e. the languages labeled as ergative or displaying ergative traits in Table 6). Figure 6 presents the averaged proportions of each strategy in these two language groups. Since these averages relate to the proportions, the differences between languages in the number of examined verbs are outbalanced. Figure 6 shows that in languages without ergative traits, all strategies are frequently represented, with an advantage for the augmented strategy (augmented: 44.6%; reduced: 29.6%; undirected: 25.7%). Languages with ergative traits crucially differ: the undirected strategy is the preferred option (68.2%), the reduced strategy is marginally represented (2.2%; only three instances in the Cabécar data), while the augmented strategy still occurs often (29.6%).

Figure 6:

Directionality of alternation type per alignment type.

In order to test the statistical significance of these differences we fitted generalized linear mixed-effects models on the data. In order to get informative results about the threefold distinction between augmented, reduced and undirected strategies, we analyzed the data in two steps. In a first analysis, we examined whether Alignment (languages with vs. without ergative traits) has an impact on the choice of directed (augmented/reduced) or undirected strategies (dependent variable). The statistical model contains the fixed effect of Alignment (languages without ergative traits as a baseline; languages with ergative traits as level of interest). The individual Languages were added to the model as a random variable, which means that the statistical model examines whether the effect of alignment type holds true beyond the variation that is due to the different languages (see SM 2, Section 5.3). This analysis (see Table 7) shows a negative effect of Alignment (−3.58), which means that the directed strategies (augmented/reduced) occur less often in languages with ergative traits (see the increase of ‘undirected’ in languages with ergative traits in Figure 6). A log-likelihood test was performed testing whether a model without the fixed factor Alignment has a significant impact on the model fit. The result of this test is significant (p-value below 0.05 in Table 7), which means that Alignment has a significant effect on the choice between ‘directed’ and ‘undirected’ in our data.

In a second analysis, we examined the choice of direction, i.e. the choice between augmented or reduced strategies in case that a direction is selected (i.e., excluding ‘undirected’). A generalized mixed-effects model with the same parameters as above was fitted to the data. The dependent variable is the choice of augmented (baseline) or reduced (level of interest), the role of Alignment is a fixed factor, and the role of Languages is treated as a random factor. The result in Table 8 shows that Alignment has a negative impact on this choice (−5.89), which corresponds to the decrease of instances of the reduced strategy in languages with ergative traits. The log-likelihood test is again significant (p-value below 0.05 in Table 8), confirming that Alignment has a significant effect on the occurrence of reduced pairs, such that they significantly decrease in languages with ergative traits.

The above data show that the augmented pairs occur frequently both in languages with and in languages without ergative traits. Augmentation seems to be the most natural alternation given the functional makeup of the experiential domain in terms of the prominence of the experiencer and the stativity of the situation. Assuming that a psych situation primarily renders a state of the experiencer directed towards a stimulus, as delineated in Section 1, the experiencer-oriented lexicalization, as present in the intransitive ES predicates seems to be the structurally most direct way of encoding the plain alternant^[21] with the induced alternant to be formally (more) marked. This is supported in Nichols et al. (2004: 172f) with respect to their ‘animate verbs’, which cross-linguistically favor the basic lexicalization of the plain alternant, while the induced alternant tends to be derived.

The results in Table 8 indicate that reduction is significantly less frequent in languages with ergative traits compared to languages without ergative traits. Disfavoring reduced alternation pairs is also in line with the prerequisites of ergative alignment, at least in those cases where the ergative is associated with causer/agenthood (Legate 2012; Woolford 1997). In a reduced alternation pair the transitive verb is the basic lexical form, which does not seem to (optimally) fit the prerequisites of a psych situation. Such a coding does not only run against the abovementioned universal lexicalization preference for ‘animate’ verbs to form augmented pairs but would also equip the stimulus with causer/agent properties in the basic lexicalization. We know from detransitivizing languages (e.g. Spanish, Romanian, Serbian, German, Icelandic from our sample) that the (basic) transitive structures show non-canonical psych-specific semantic properties, among them stativity and non-agentivity (e.g. Belletti and Rizzi 1988; Dowty 1991; Landau 2010; Pesetsky 1995; Verhoeven 2015 among many others). The latter would not be in line with the thematic properties of an ergative argument.

Finally, the significant association of languages with ergative traits with the undirected alternation type, as significantly confirmed in Table 7, seems to come as a surprise, at least at first sight. Ergative alignment is expected to naturally go hand in hand with augmentation under the assumption that the observed alternation is a kind of actor-adding alternation, which is indeed the case, as explained before (see again Nichols et al.’s 2004 assumptions on their ‘animate verbs’). In this case a basic intransitive encoding and a marked transitive encoding would not only be visible on the verb but additionally by adding an ergative argument, while the absolutive argument remains constant across the alternation. Most languages with ergative traits in our sample (i.e. Yucatec Maya, Cabécar, Georgian, Hindi, Marathi; see Figure 2) indeed display augmented pairs to some extent, Basque being the only exclusively underspecified language (as mentioned above, the average of augmented pairs is 29.6% for the languages with ergative traits of our sample).

Nichols et al. (2004: 169) speculate about the association of ergative alignment with the undirected alternation type^[22] by referring to the ‘obsolete’ hypothesis that basically or originally the ergative structure is intransitive, adding the ergative as an adjunct. They argue that the ‘indeterminate’ mechanisms, as present in ambitransitive pairs (cf. the English example in (6) and the Chinese example in (13)) and pairs whose alternants are only distinguished by conjugation class (cf. the Hebrew pair in (14)) would be in line with such a conceptualization in so far as both alternants, the intransitive AND the transitive ones are morphologically unmarked. However, given that the ergative languages of our sample all belong to the ‘neutral’ type – they are either double deriving or show auxiliary/light verb change – this explanation is not applicable. In case of auxiliary/light verb change (see example (12) from Basque, example (26) from Marathi and example (28) from Hindi) there is indeed a common form used in both alternants. However, this is most often a nominal/nominalized form, either an adjective/participle or a noun, while the transitivity alternation is due to the auxiliaries or light verbs, which are intransitive or transitive, respectively (cf. Creissels and Mounole 2019).

(28)

Hindi

a.

plain
Vah	aurat	apn-ī	gāṛī	sē	parēśān	h-u-ī.
that	woman	refl-f	vehicle	abl	irritated	become-pst-f
‘That woman got irritated because of her car.’

b.

induced
Us-k-ī	gāṛī	nē	us	aurat	kō	parēśān	ki-y-ā.
that.obl-gen-f	vehicle	erg	that.obl	woman	dat	irritated	do-pst-m
‘Her car irritated that woman.’

Further, the languages which evince double derivation (Cabécar, Georgian, Yucatec Maya) do not conform with the aforementioned explanation since they display overt morphology both in the transitive and the intransitive form, see (30b–c) for Georgian, (31) for Yucatec Maya, and (42a–b) for Cabécar. Similar to the aforementioned cases of auxiliary/light verb change, Georgian and Yucatec Maya mostly take nominal (adjectival or noun) roots as bases in their double deriving pairs, adding suffixes, i.e. causatives/factitives or inchoatives/passives, which are at the same time verbalizers. Corresponding cases are also present in Cabécar, as exemplified in (42a–b), where a causative and a middle form are derived from the adjective dokó ‘ugly’ resulting in the transitive EO verb dokówa̱ ‘disgust’ and the intransitive ES verb dokóna̱ ‘get disgusted’. Most undirected pairs however belong to the most pervasive verb class in Cabécar, i.e. verbs formed from equipollent roots, which require the causative and the middle suffixes in order to be inflected as transitive and intransitive verbs, respectively (González Campos and Lehmann 2018, Sect. 10.2.2.4; González Campos and Obando Martínez 2018), as illustrated in (29). In the psych domain (as in general) these roots seem to be mostly non-productive but may historically relate to productively used nominal roots (González Campos and Obando Martínez 2018).

Overall, these cases do not fit the assumption of a basic intransitive type, which is also used for the transitive type (via ambitransitivity). Hence, this explanation is not valid for the significant association of languages with ergative traits with the undirected alternation type in the psych domain. The detailed inspection of the data rather points to the dominant role of the nominal origin of psych concepts in undirected alternation pairs (here: the neutral type, see Section 2), which is however not specific to ergative languages. The positive correlation of languages with ergative traits and undirected alternation could thus be epiphenomenal to the nominal makeup of the psych domain.

4.2.3 Pattern heterogeneity and the problem of dyadic alternations

One of the major ways in which psych verbs in our sample deviate from the expected patterns is via a stronger admixture of strategies in a number of languages. A closer look at the data reveals that one of the reasons for this is that some languages actually create a triad of forms rather than a binary, opposing a stative, an inchoative and a causative form.

As the examples in (30)–(32) show, languages vary as to which form is the most morphologically simple: while it is the stative form in Georgian and Yucatec Maya, Mapudungun lexicalizes the inchoative meaning as the most basic form. All forms oppose a causative/factitive derivation which uniformly serves as the induced alternant. Given the binary opposition proposed by Nichols et al.’s (2004) typology, the way the involved forms constellate is not surprising: the induced alternant has a clear functional delineation, as it renders the semantic causation of whatever the plain form encodes. In contrast, the plain alternant is underspecified for lexical aspect. This fuzziness does not pose a serious problem for Nichols et al.’s (2004) account, in part due to the presence of eventive predicates in their list (cf. the animate subset given in (3)). However, it is of major import in the psych domain because of its ontological components. One of the core components of every psych verb we laid out in (2) is a mental state, i.e., the psychological content at the core predicate. The experiencer holding a certain mental state due to exposure to the stimulus thus is a non-eventive situation (cf. Pylkkänen 2000).^[25] As the authors note themselves (Nichols et al. 2004: 156f.), languages show quite a bit of variation in the lexicalization of concepts with some inherent stativity. Most languages typically render this either immediately in a stative plain alternant, or mediated via the addition of a temporal boundary, resulting in an inchoative plain alternant. For the European set of languages where reduced pairs prevail, there is no immediate effect of stativity versus inchoativity, because this configuration renders the mental state as an induced predicate first. Thus, even though languages of this type may also distinguish inchoative forms from stative ones, as shown in (33)–(34), this does not impact their overall classification, as the morphological relationship between the base (i.e. the (a) forms) and the derived form (the (b) or (c) forms) remains one of reduction.

In strongly underspecified languages, this triadic setup is similarly unproblematic with respect to typologization, as all three alternants are of equal morphological complexity. The coexistence of a stative and an inchoative plain exponent which oppose a singular induced form merely increases the overall size of the domain. Such a situation is attested in Irish in our sample, as illustrated in (35), which resolves into the two undirected tuples (35a) ∼ (35c) and (35b) ∼ (35c).

For transitivizing languages however, the fact that both statives and inchoatives can provide a suitable binary alternant to induced predicates creates a systematic problem. If an inchoative derived from a stative is juxtaposed as the plain alternant with a causativized stative as an induced alternant, the two equally morphologically complex items yield an undirected pair, as seen in the (b) and (c) forms in (30)–(32). However, taking the stative as the plain form, the same contrast results in an augmented pair, as with (a) and (c) in (30)–(32). Due to the binary coding system in our dataset, this then yields two coexisting pairs of different types but with a shared induced alternant, creating the impression of a heterogeneous domain for these languages. Crucially, valence orientation in the psych domain may thus fall out from the need to identify a single plain alternant. Note again that the patterns from Mapudungun (32) show that this cannot be resolved by simply adopting the stative alternant as the basic underlying form in the alternation (cf. Piñón 2001). In light of the robust yet language-specific preference for either statives or inchoatives in other languages, it seems clear that no a priori choice can be made. Rather, it seems that due to the unique event semantics of the psych domain with its inclination for stativity, an adequate typologization should accommodate all three forms if possible. One solution could be to subsume underspecified languages and transitivizing ones under a single type, distinguishing instead of three valence orientation types a causativizing from a causative-less alternation (the detransitivizing type), crossed with a factor coding the availability of a stative and an inchoative reading (to the logical exclusion of the feature combination [−stative, −inchoative], since this would entail no alternation at all). The special status of the detransitivizing type is supported by the survey of morphological means in Section 4.2.1, which showed that augmented pairs and undirected pairs employ the same kinds of morphemes, while we found that reduced pairs are formed in ways that rarely or never occur in pairs with equal formal complexity across alternants. However, exploring the ramifications of such a two-dimensional typology is beyond the scope of the current contribution and should be addressed in future research.

The exact conditions for the occurrence of a triadic system in the psych domain are as yet unclear. Our limited sample offers some evidence that overall morphological richness in the verbal domain and a predominance of agglutinative (or non-fusional) morphology could play a role, since this facilitates the distinct expression of both the inherent stativity as well as temporal boundaries such as the onset of the mental state. In many other languages however, only one of these mechanisms tends to be preferred, although individual verbs may differ (cf. Verhoeven 2010). Such preferences may in turn be specific to the psych domain, as our Turkish data shows. The dominant derivational relation here is augmentation (n = 44, 68.75% of our sample), often from an overt anticausative with an inchoative reading (n = 27, 61.36% of the augmenting subsample):

The intransitive verbalizer marker -lAn- ^[26] is itself internally complex, consisting of the general verbalizer -lA- and a morpheme -n- which has been analyzed as either a passive marker (Bağrıaçık 2018) or a reflexive marker (Kornfilt 1997; see also Göksel and Kerslake 2005: 56f.), both typical sources for anticausatives (Haspelmath 1990). Strikingly, although the verbalizer on its own is highly productive outside the psych domain, as the examples in (37) (based on Kornfilt 1997: 453f.) show, it is completely absent from our sample, and forms such as **endişelemek are considered ungrammatical by speakers (as opposed to the licit infinitive endişelenmek of (36a), -mAk being an infinitival marker).

One reason for this might be the fact that the verbalizer -lA- is highly polysemous, creating verbs with a large variety of aspectual semantics and argument linking and different valency patterns (Bağrıaçık 2018), which could mean that the anticausative marker is required to express the stative nature of the psych situation. It is worth noting that verbs derived in this way prefer overt causativization via the suffix -dIr-, although there are other affixal constellations where verbs can form corresponding transitives either via verbalizer -lA- or the complex transitive/causative verbalizer -lAt- (Göksel and Kerslake 2005: 56f.). The former would create a reduced alternation while the latter would yield an undirected pattern. The differentiating factor is that the psych verbs formed in this way are largely derived from nouns, which according to Göksel and Kerslake (2005: 56f.) bars the immediate derivation of corresponding transitives from the root. Thus, while Turkish has a large number of morphemes which could contribute to a triad of forms on par with (30)–(34), this does not happen due to idiosyncratic grammatical restrictions. Rather, it seems that many of the anticausatives can be used with both stative and inchoative readings. The remainder of the augmenting set does not use the morpheme -lAn-, and is either inherently stative/inchoative (e.g. kızmak ‘be ∼ get angry’, korkmak ‘be afraid’, coşmak ‘get excited’) or has other morphological markers with similar functions (e.g. sevinmek ‘be happy’ from sev- ‘love’ via an intransitive reflexive verbalizer -In). Etymological research shows that such items tend to derive from non-psych verbs via metaphorical extension, a common trajectory of psych verb formation (e.g. ‘get angry’ < ‘get warm’ for kızmak; ‘get excited’ < ‘boil up, overflow’ for coşmak; see also Klein and Kutscher 2002; Kutscher 2009; Verhoeven 2007). The second largest group in Turkish is the set of undirected verbal pairs (n = 12, 18.75%). These, too, are deeply shaped by their denominal nature, as the examples in (38)–(39) illustrate.

Essentially in light verb constructions similar to examples (15) and (26), we find a constellation of statives (the (a) alternants) and causatives (the (b) alternants) as a product of the verbalization strategy of the psych nominals. Word formation thus plays a central role in the patterns of the Turkish psych domain and may at least in part explain why, although Turkish would appear to be a good candidate for a triadic psych alternation, the internal structures of the domain do not yield this pattern. Additionally, observe that the univerbation in example (39) (also attested in our dataset for the pair mahvolmak ‘be devastated’ ∼ mahvetmek ‘devastate’) may shed some light on how the periphrastic accommodation of stative nominal psych expressions may over time give rise to undirected alternations even if a language makes heavy use of directed alternations elsewhere (cf. also the more marginal existence of such pairs in detransitivizing languages, e.g. German Angst haben, Spanish tener miedo ‘be scared’, literally ‘have fear’ ∼ Angst machen, dar miedo ‘scare’, literally ‘make/give fear’).

4.2.4 The impact of word formation and lexical restrictions

Word formation was also found to be a key factor for other languages with a more mixed picture. Since we have discussed the patterns of Finnish elsewhere (see Rott et al. 2020), we will only give a brief summary here: In short, this language employs similar morphemes for the purpose of psych verb formation as well as creating the alternation (chiefly causatives, factitives and inchoatives for the former, causatives and inchoatives for the latter). Often, this leads to multiple derivations occurring on the same root, creating a heterogeneous picture.

The psych nominal in (40) gives rise to two pairs of opposite directionality, a reduced pair ((40a) ∼ (40b)) and an augmented pair ((40b) ∼ (40c)). Conversely, the psych nominal in (41) only creates one undirected pair via the exchange of the suffixes. The exact choice of strategy appears to be lexically determined (see Rott et al. 2020), and may be related to more fine-grained semantic distinctions of both the psych situation to be lexicalized as well as the lexical material enlisted for the purpose (e.g. future vs. past orientation in emotional domains such as fear or the increased likelihood of certain emotions such as anger or disgust to elicit an outward action). Nearly half of our Finnish sample was found to be synchronically denominal (n = 48, 48.44%) and instantiated different constellations of these strategies, as can be seen in Figure 2. Deverbal items were also well-attested (n = 12, 20%). A similar situation can be found in Cabécar, where different verbalization strategies may apply and combine (42), creating both an undirected pair ((42a) ∼ (42b)) and an augmented pair ((42b) ∼ (42c)) which share an inchoative alternant.

The Hungarian psych domain is characterized by strategies in complementary distribution, i.e. the same form was rarely involved in the formation of multiple pairs, but rather different bases were lexically determined to take different arrays of suffixes (with different allomorphs for each morphological operation, a characteristic trait of Hungarian, cf. Nilsen Márkus 2015), as seen in the undirected pair in (43), the reduced pair in (44), and the augmented pair in (45).

Compared to Finnish, Hungarian seems to draw more heavily on categorially underspecified bases in its psych domain (n = ∼22, ∼46%), which may factor in the more pronounced trend toward lexical specification, as the derivational relations are much more synchronically opaque.^[28]

In all of the above circumstances, word formation and concomitant lexical information heavily impacts the patterns of the alternation and contributes to domain-specific heterogeneity.