Valence orientation and psych properties: Toward a typology of the psych alternation

2.1 Elicitation

It is a well-known issue that comparative elicitation-based linguistic studies usually rely on translational tasks when gathering material in their target languages. While understandable from an economic perspective, it is highly problematic for highly culturally informed semantic fields such as emotion. English especially has undergone somewhat of a reification, despite the fact that its emotion vocabulary is as much a cultural artifact as that of any other language (Wierzbicka 1992). In order to circumvent this problem, we chose to operationalize findings from anthropological and psychological research. Johnson-Laird and Oatley (1989) identified five basic, broadly defined emotions which are predicated on abstract universal antecedent events, i.e., contingencies which every human can be assumed to encounter in their lifetime (see also Boucher and Brandt 1981, Ekman 1994, 1999, Hupka et al. 1999, Turner 2007). While these are referred to using English labels in Table 1, note that their definitions are grounded in human social plans, representing junctures requiring a reaction.

Based on these definitions, we created simple scenarios with generic human referents and controlled for stimulus animacy. This yielded 5 × 2 = 10 scenarios such as the ones illustrated below, presented orally to consultants in a common language (English, German, or Spanish):

Speakers were then asked to imagine the given situations in a native language context and to empathize with the character representing the experiencer. Finally, a series of questions designed to highlight different parametric aspects of the target emotion (e.g., intensity, onset, social appraisal; cf. Johnson-Laird and Oatley 1989) were asked in order to facilitate broad lexical retrieval. Speakers provided verbs which they considered adequate descriptions of the situations in citation form as well as in naturalistic usage in simple declarative sentences. In order to elicit verbal alternations, consultants were also required to invert the structure of the latter, e.g., by rephrasing with a focus on the object of the original sentence. Additional information about distributive restrictions (e.g., stimulus animacy, register) as well as transparent interlexical relations were also recorded.

2.2 Sample and morphosyntactic analysis

The areal dispersal of valence orientation as identified by Nichols et al. (2004) bears directly on our central observation in (15). We therefore based our study on a convenience sample of eight typologically and genetically diverse languages from seven language families and four different linguistic macro areas, see (17). We chose languages that employ an array of different morphological strategies to achieve the psych alternation, see (11) and Table 2. Research on noncanonical behavior in the psych domain is available for all languages except Bété (17h).

We worked with one native speaker per language, gathering material across multiple sessions. Upon completion of the elicitation, we identified pairs conforming to the criteria for the psych alternation as given in (7) and coded the respective associated data points for morphological and syntactic patterns, yielding a crosslinguistic database of alternating psych predicates which will form the core piece of the remainder of the present discussion.

3.1 Overview

The distribution of base orientation in our sample is summarized in Table 2. A number of important observations are to be made here. Perhaps most strikingly, Bété was found to altogether lack structures which qualify as psych alternation according to the definition in (7). This, however, does not mean that Bété does not exhibit alternating psych structures at all – we will return to this issue at the end of this section.

For the languages in our sample which do conform to (7), the geographic patterns of the predominant strategies are in line with Nichols et al.’s (2004) observation for valence orientation in general. Detransitivization (i.e., a majority of EO basic forms) was found to be globally rare, with the critical exception of the Indo-European phylum. Both detransitivizing languages in the sample are Indo-European languages, while those with a preference for transitivization are found outside of Europe.2 ^[2] Moreover, Nichols et al. (2004) observed for their class of animate verbs (which encompasses experiential predicates because they by definition select for a sentient, and thus animate, participant) that languages with a propensity to detransitivize will disfavor causative derivations. This is also borne out, as the overview of morphological means in Table 2 shows: across Icelandic and Spanish, where reflexivization and middle voice formation prevail, only a single pair with an ES basic form is attested. By applying the classification in (10) to our set of languages in (17), we arrive at the distribution in Figure 1.

Figure 1

Distribution of basic forms in sample (n = 435).

Since the sample includes languages with a clear valence orientation (Icelandic, Spanish, Korean, Chinese, and Yucatec Maya) as well as less clear cases (Turkish and Finnish), it represents an interesting testing ground for our central observation as stated in (15). Before discussing individual languages, we need to address one caveat: the crosslinguistic study of noncanonical phenomena is notoriously difficult, because noncanonicity is language-specific by definition. It is construed as deviations from what is unmarked in a given language’s grammar, chiefly its syntax (see, e.g., Belletti and Rizzi 1988, Pesetsky 1995, Bayer 2004, Landau 2010). The uniting feature, as mentioned in the introduction, is the alignment of syntactic and semantic prominence in the experiencer argument. How this plays out, though, is dependent on the means and constraints set by the typological properties of a language. Hence, it is difficult to identify and operationalize a homogeneous set of phenomena which one would predict to find in every language that aligns with our central observation (15). This will become clear in the following account.

3.2 Valence orientation and psych properties in the sample languages

As mentioned previously, Icelandic and Spanish are two clear cases of a detransitivizing psych domain. This coincides with the robust psych properties found for both languages within and outside of the psych alternation. As seen in (12b), there is a restriction on psych passives in Icelandic in EO alternants. Further, the language is arguably infamous for its unmarked structures with oblique arguments (most commonly in accusative and dative case), as illustrated in (18).

Numerous studies have shown that the oblique arguments in constructions such as those in (18) pass all kinds of subjecthood tests and behave unlike objects with identical case marking (see, e.g., Höskuldur Þráinsson 1979, Zaenen et al. 1985, Halldór Ármann Sigurðsson 2004). More recently, an ongoing morphosyntactic change known as þágufallssýki “dative sickness” or þágufallshneigð “dative inclination” has begun to single out the subject-like obliques of psych verbs (Ásta Svavarsdóttir 1982, Smith 1994, Jóhanna Barðdal 2011), affecting accusative experiencers and turning them into datives3 ^[3] (Jóhannes Gísli Jónsson and Þórhallur Eyþórsson 2003). Speakers who manifest this phenomenon will produce (19) for (18a).

Even a few isolated ES verbs may be affected by this change for some speakers who will realize the canonical nominative experiencer argument in (20a) as a dative (20b):

Icelandic also has subject-like obliques which lack any clear psych meaning. These typically denote nonagentive themes. Crucially however, these are never affected by þágufallshneigð but may acquire structural nominative case instead (Þórhallur Eyþórsson 2000: 28). This is shown in (21).

This further attests to the special status of experiential arguments in the Icelandic system and the impact this category has on syntax.

Similarly, Spanish shows case variation in pronominal clitics which alternate between accusative and dative marking when used with psych verbs (Vázquez Rozas 2006, Fábregas et al. 2017), giving rise to a systematic link between basic EO alternants and forms outside the alternation. Accusative clitics are correlated with an eventive interpretation, whereas the use of dative clitics oblige a stative reading. As illustrated in (22), the choice of case in turn has a direct effect on the markedness of domain-specific word order patterns, facilitating the occurrence of oblique-first structures with dative experiencers (see, e.g., Jiménez-Fernández and Rozwadowska 2016), a structure which can only be licensed pragmatically for accusatives.

Moreover, Spanish exhibits noncanonical behavior in binding, extraction, and passive constructions both within and outside the psych alternation (Franco 1990, Marín 2014, Marín 2015).

In contrast, the psych domains of Chinese, Korean, and Yucatec Maya are highly transitivizing. The unrestricted availability of forward binding in Chinese EO verbs was already demonstrated in (13a). The behavior of the psych domain in this language was further investigated by Verhoeven (2010a) who showed that psych verbs do not differ from canonical transitives with regard to properties of lexical aspect, passive formation, nor binding of reflexives and reciprocals. It must be noted that Chinese has been shown to exhibit backward binding (Cheung and Larson 2015). Structures such as those in (23), which is based on Cheung and Larson (2015, 130), may be considered problematic for our claim.

However, the status of this phenomenon as a valid psych property has been questioned, as it may rather be attributable to pragmatic effects of logophoricity and givenness (Haspelmath 2001a, Bickel 2004, Landau 2010). Structures such as (23) thus do not pose a clear counterexample to our claim.

For Korean, evidence is somewhat scarcer. Verhoeven (2008) shows that EO verbs behave canonically in terms of passivization, topic marking, and word order. Temme and Verhoeven (2016) substantiate this last observation with experimental evidence. Following analyses by Ki-Seong Park (1993) and Shibatani and Chung (2001), we propose that the canonicity of these constructions can likely be attributed to the semantics of the causative light verb - 게 하다-key hata “do”, as shown in (24) from our own data.

This means of causativization can be construed as an agentive predicate and thus favors an agent subject (see also Fritz-Huechante et al. 2020).

Evidence for canonical psych-verb behavior in Yucatec Maya parallels that for Chinese and Korean (Verhoeven 2010b: 220ff). Passive formation is completely regular with transitive EO verbs (Verhoeven 2007: 249). These verbs also regularly allow forward binding, in parallel to example (13a) from Chinese, as shown in (25) (Verhoeven 2007: 256). Moreover, there are no noncanonical linearization preferences with transitive EO verbs.4 ^[4]

Example (25b) illustrates an instance of a psych construction recurrent in Yucatec Maya but not (systematically) present in the other languages of our sample, with the exception of Bété, which will be discussed in Section 3.3. It is best described in terms of Matisoff’s (1986) psi-collocations, which are very common in languages of Africa, South East Asia, and South America (cf. e.g., Craig 1977, McElhanon 1977, Ameka 1990, Bugenhagen 1990). It consists of a psycho-mate, i.e., a verb or adjective, and a psycho-noun which represents the experiential locus, usually a body part. The latter is possessed by the experiencer, i.e., the experiencer does not figure as a direct dependent of the psych predicate but rather is encoded as a possessor of the body part noun. Hence, with these structures, the alternating entity is not the experiencer, as in the cases presented in (11), but rather the body part noun phrase (NP). The alternation is shown in (26) with data we obtained using the method introduced in Section 2.1.

Most importantly for the present discussion, the possessor experiencer does not exhibit noncanonical syntactic properties, differently from what has been shown for possessor experiencers in other languages (Bickel 1997, Bickel 2004 on Tibeto-Burman languages).

The Turkish patterns are somewhat more heterogeneous, although the majority of basic forms is ES. The second largest group is formed by underspecified bases whose linking is rendered via auxiliary change. This is achieved via embedding under the light verbs olmak “be” for the ES alternant and etmek “do” for the EO alternant, as shown in (27). In a few items, univerbation with the light verb has created double-deriving pairs. The resulting alternation is illustrated in (28).

In terms of psych properties, research points toward canonical patterns in the domain (Kutscher 2009, Özsoy 2009, Verhoeven 2010b, 2014).

So far, the patterns of these two groups are in line with our observation in (15). Finnish presents a more complex case, as its basic forms are distributed across the different orientations, with ES basic forms being around 1.5 times as frequent as EO basic forms as well as a substantial subset of pairs originating from double derivation. It thus embodies the entire range of the valence orientation classification in (10), making it a touchstone for our central observation as given in (15). Examples of the derivational relations attested in our sample are given in (29).

We found that some verbs even participate in multiple strategies, forming multiple alternating pairs with one shared form. This is illustrated in (30) for the detransitivizing pair innostaa “excite” ∼ innostua “get excited”, which coexists with the transitivizing pair innostua “get excited” ∼ innostuttaa “make get excited”.

Note however that (30) does not entail that the relationships laid out here are mere artifacts of our method. As we illustrate in (31), the most basic forms only license a binary alternation whose derivational directionality is not predictable from the morphology of an individual alternant.

Interestingly, multiple valency orientation in the Finnish psych domain coincides with divided opinions on noncanonical psych-behavior in the literature. In light of the prominence-based motivation for the observation in (15), a priori one might expect psych properties to only occur within the set of EO basic forms, since both ES basic forms and double-deriving forms should have a clear, overtly marked functional allocation. Starting with our own elicited data, we find structural alternations such as those in (32) where the nonnominative experiencer can unmarkedly occur in subject position, while the stimulus is facultative.

Notably, according to our consultant, inversions such as (32b) are only available with inanimate and abstract stimuli. Clearly a noncanonical property, such partitive-initial structures are widely recognized in Finnish linguistics and are even referred to as “experiencer constructions” (see, e.g., Siiroinen 2005). However, our data show that while it is not entirely unrestricted, this inversion is found in EO basic forms, causativized ES basic forms as well as doubly derived EO forms, contra our expectations. We will return to this issue shortly. In the literature, Landau (2010) cites Finnish as one example of a language with noncanonical restrictions on passivization in stative psych verbs, although he only gives a few examples and does not explicate whether there is any interaction with the derivational makeup of the individual verbs. He concludes that these psych verbs must be unaccusative. However, Finnish passives do not straightforwardly compare to Indo-European structures despite superficial similarities, differing in a number of crucial respects. The most important difference is the fact that the underlying operation in Finnish passivization is one of argument suppression, while in Indo-European passivization, it is one of argument promotion (see, e.g., Shore 1988, Sakuma 2013). Consequently, Finnish passives are freely formed of predicates which have traditionally been classified as unaccusatives as well as copulas and thus cannot serve as a diagnostic in this regard (see Shore 1988). Nelson (1999) acknowledges this complication and evokes aspect and event structure to explain the exceptional linking properties and unaccusative behavior observed for causative EO verbs, i.e., verbs overtly derived from ES basic forms (see also Pylkkänen 2000). Depending on how these results are interpreted, this may be problematic for the central observation (15), or at least suggest that for some languages with substantial heterogeneity in the psych domain, the availability of stative readings may expand from the set of EO basic forms to derived EO forms, yielding “psych causatives”, seemingly creating a violation of the functional motivation given above for our proposed functional correlation (Pylkkänen 2000). Seeing as Finnish causatives outside of the psych domain seem to be typical representatives of this crosslinguistic category (cf. Brattico et al. 2007), this is especially surprising. However, at present, the exact impact of the derivational structure within the Finnish psych alternation on the (non-)canonical behavior has not been investigated systematically. Seeing as (15) was borne out for all other languages in our sample which showed a psych alternation, this is a crucial issue which needs to be studied in more detail.

Lastly, it is worth noting that both Nelson (1999) and Pylkkänen (2000) set up a four-way classification of the Finnish psych domain, focusing on pure states, inchoatives (i.e., ES verbs), and their respective EO causativizations (see also Siiroinen 1998), while our findings as summarized in (29)–(31) indicate that the situation is actually more complex than that, given the multitude of directed and nondirected derivational relations which hold among Finnish psych verbs.

3.3 The case of Bété

Let us now return to the seemingly exceptional case of Bété. Following Zogbo (1981: 72–7) and Koopman (1984, ch. 2), Bété has causative morphology and middle voice morphology. Both of these devices are frequently employed in the psych alternation by other languages (cf. the examples in (11) and the summary of our sample in Table 2). Bété’s failure to evince a psych alternation is thus not due to a complete lack of alternating expressions. Yet no pairs in our data satisfy criterion (7d) of the psych alternation, since they all violate the required inclusion of the stimulus as an argument. Another crucial difference from the alternations found in cases such as those shown in (11) is that, although structural alternations are formed, the experiencer appears to remain in the same or similar syntactic position across alternants, i.e., criterion (7b) requiring an ES ∼ EO alternation is not satisfied. Our proposal is that both of these observations are epiphenomena of the overall structure of the psych domain in this language: Bété lexicalizes psych meanings as complex constructions, unlike most other languages in our sample save for Yucatec Maya. Judging from our data, Bété constructions seem to fall into two major types, which we will refer to as types A and B. Both have the same basic makeup schematized in (33).

Type A constructions, illustrated in (34)–(35), closely resemble Bouchard’s (1995) nonincorporated psych verbs.

In terms of the schema in (33), they can be analyzed as follows: the nominal component N renders the semantic core of the predicate, i.e., the emotion itself, e.g., mʊ̋ná “joy”, jʊ́rʊ́ “anger”, ŋwa̋nɪ́ “happiness”, ŋāzɪ̄ “pity”, ŋānɔ̄ “fear”, and ŋāzɪ̄ “pity” (see also Verhoeven 2007: 51–4). The verbal component V is a semantically bleached light verb, e.g., sɪ̄ɓā “sting”, p(ʉ̄)lā “penetrate”, wʊ̋ súrú “pour onto”, mʉ̋ pálɪ́ “put into”, and lɪ̋ɓa̋ “hit”. In fact, we found that many of these light verbs combine with a multitude of different emotion nouns. Derivational directionality is lexically determined and follows from the choice of light verb. In alternants such as (34a) and (35b), the emotion noun takes the subject role. Crucially, the experiencer argument is then encoded as the object of the light verb. In (34a), the only felicitous way to specify the stimulus is via a free adjunct introduced by the equivalent of English “because”, while in middle constructions such as (35b), there is no structural option to include the stimulus (cf. Koopman 1984: 22f). Note that type A expressions already have a dyadic structure prior to the inclusion of the stimulus. When a stimulus is added as in (34b), a causative form of the transitive light verb is used. The stimulus is rendered as a causer subject while the emotion component becomes the causee, the result being a double-object construction.7 ^[7] Conversely, the stimulus is present in the underived alternant (35a), which also gives rise to a double-object construction. It is then removed via middle voice morphology. The important observation for our analysis is that the experiencer remains in an object position across the alternants.8 ^[8] In addition to causativization, Bété employs another double-object construction, achieved through applicative formation when adding a stimulus to the schema in (33), as illustrated in (36). Example (36a) is parallel to (34a) in featuring a transitive light verb lɪ̋ɓa̋ “hit”, an emotion noun in subject function and the experiencer as a (direct) object. In (36b) an applicative construction is used: the stimulus appears in the subject position, while the emotion component is rendered as an applied object in instrumental function (cf. Zogbo 1981: 74–5).9 ^[9] Crucially for our analysis, again the experiencer remains in an object position.

The three types of alternations (subsumed here as type A constructions) differ structurally from the pattern observed in languages such as those shown in (11), because the experiencer is always low in the structure, being coded as an object. What alternates is the emotion, i.e., the nominal component N from (33). This means that type A constructions fail another criterion for the psych alternation in addition to (7d) stimulus inclusion, namely, (7b) the requirement for an ES ∼ EO interchange. Since the central observation (15) is predicated on this definition, it does not transfer straightforwardly to type A expressions. Hence, such constructions in Bété, and likely other languages with similar patterns, present an altogether different dimension of variation as regards structural alternations in the psych domain (see also Rott et al. 2020) which cannot be classified as a psych alternation. However, the mechanisms underlying (15) should hold regardless, since they are independently motivated. The experiencer in type A is downgraded across the board, which suggests that noncanonical phenomena could develop or may be found, given further study. Their occurrence, however, should not be correlated with the morphological strategies employed to create the alternants. Rather, they should apply consistently in accordance with the downgraded status of the experiencer.

Type B, shown in (38)–(39), is structurally similar to the “psi-collocations” found in the Yucatec Maya data (see (26)).

It places the semantic burden on the psycho-mate, i.e., the verbal component V in (33), which surfaces as a metaphorically loaded action predicate such as ɟírí “cook”, sī “burn”, and mà dó “cut off”. The construction is then shifted into the psych domain by means of the nominal component N in (33), which is a typical psycho-noun such as dīgbə “heart ∼ soul” or drɪ́ “heart”. Other body parts are also found with certain specific psycho-mates, forming highly lexicalized collocations. According to our findings, the experiencer is most commonly realized as a possessive attribute to the NP headed by the psycho-noun, as in (37)–(38) (cf. Bickel 1997, Bickel 2004). The inclusion or the suppression of the stimulus is achieved through the same structures as introduced for type A in (34)–(35), namely, causativization and middle formation with the psycho-noun being the head of the NP alternating between subject and object function. Apart from the possessor experiencer structures, we also frequently found forms such as (39) in which the experiencer and the psycho-noun coexist without any overt possessive marking.11 ^[11] We propose that the experiencer in (39) is a dative-like verbal dependent of the external possessor type (cf. Payne and Barshi 1999, Lehmann et al. 2000: 55–6, Deal 2017). Causativization of the verb in (39b) allows the stimulus to be added as a causer subject, while the psycho-noun becomes the causee in the resulting double-object construction.

Some further support for the object status of the experiencer is provided by applicative morphology variants which we found for some overt possessor structures such as (40). Example (41a) shows its applicative variant, where the experiencer is licensed as an internal argument in goal function by the applicative suffix on the verb (cf. Zogbo 1981: 74, Koopman 1984: 23–4). The example in (41b) shows that the stimulus can be added as a subject which again results in a double-object construction, parallel to (39b).

Note that in (41) the applicative occurs across both alternants. In this regard, type B expressions necessarily differ from type A expressions such as (36), where the applicative contributes directly to the alternation.

If our overall analysis is correct, type B expressions parallel type A as well as our findings for Yucatec Maya in that the pivot of the alternation is the nominal component N rather than the experiencer, which remains in a syntactically low position across alternants. Further, throughout all pairs given for type B, one alternant consistently lacks the means to include the stimulus as an argument (i.e., (37a), (38b), (39a), (40), (41a)). This complete lack of a governed stimulus also sets these constructions apart from equivalents in Yucatec Maya such as (26a), where the stimulus argument is introduced by a preposition. A possible explanation which holds across both types A and B may be that Bété has an altogether different cutoff point on Verhoeven’s (2007) hierarchy for stimulus inclusion (see (6)). If true, this invites a reassessment of the stimulus inclusion criterion (7d), because a consistent lack of stimulus inclusion could be an independent typological feature. In contrast to type A expressions, the alternations with an overt adnominal possessor experiencer in (37)–(38) can be taken as a variation of the ES ∼ EO pattern since the experiencer consistently occurs as a nominal dependent of the NP headed by the metonymically related psycho-noun, which does change its syntactic function across the alternants (see also Goossens 1989, Verhoeven 2007). This places type B at an interesting position regarding our central observation (15) about the correlation between valence orientation and noncanonical behavior. In Yucatec Maya, the lack of noncanonical structures is expected due to its highly transitivizing psych domain. Bété actually employs a mixture of morphological strategies similar to Finnish, which does show psych properties which transcend the derivational patterns. While this aspect of Bété has yet to be investigated, there is ample crosslinguistic evidence of the acquisition of syntactic prominence in possessor experiencers (Bickel 1997, 2004) as well as oblique experiencers (cf. e.g., Seefranz-Montag 1983, Haspelmath 2001a). This suggests that noncanonical phenomena could potentially exist in Bété type B expressions as well.