Stability and Change in Intransitive Argument Structure

Abstract: The argument structure of verbs is pretty uniform across languages. Thus, verbs of `falling’ involve a Theme and an optional Causer and verbs of `working’ an Agent. Aspect is relevant to that uniformity as well since the former verbs will be telic and the latter durative. Stative verbs form a third main class. I first show that, when (spoken/written) languages change, the basic argument structure and aspect don’t change for most unaccusatives and unergatives. There are, however, systematic reports (e.g. Rosen 1984; Keller & Sorace 2003; Randall et al 2004) that certain verbs are unergative in one language and unaccusative in another and that verbs alternate between different aspects (e.g. Levin & Rappaport Hovav 2014). I examine a few verbs diachronically that are ambiguous in the Keller & Sorace work, i.e. verbs of continuation and of controlled motional process, and conclude that a more fine-grained system is helpful.


Introduction
, Vendler (1976), Levin & Rappaport Hovav (1995), and numerous others divide the Aktionsart of verbs into different kinds. In this paper, verbs are divided into three main aspectual types, manner (durative aspect), result (telic aspect), and state (stative aspect) and these aspects determine the theta-roles. Unergative verbs, such as work and play, are durative and their basic theta roles are an Agent; unaccusatives, such as arrive and fall, are telic and their basic theta roles are a Theme and an optional Causer may render them causative2. Subject experiencer verbs, such as fear, and (many) copular verbs are stative and have a Theme and optional Experiencer.
Argument structure and aspect remain fairly stable in language change, as I have argued in van Gelderen (2018). Causative-inchoative verbs have been the subject of much earlier research (e.g. Kulikov 2009, Narrog 2009, Ottosson 2013. This work shows that, as the causative suffix ceases to be transparent in the history of English, the unaccusative and causative classes converge into a class of labile verbs. Their aspect remains stable. Typical durative verbs, such as plegan `play' and creopan `crawl', keep their Aktionsart as well. The theta-role changes to verbs include additions of Causers to unaccusatives and of Themes to unergatives, keeping the Aktionsart stable. There are also verbs, the psych-verbs, that change their aspect from durative to telic and then to stative and their theta-roles change accordingly (Allen 1995;van Gelderen 2014). That work shows that the stative alternative is the least marked and that languages change towards stative psych-verbs; similarly, sign languages have the stative variant (Oomen 2017) and first language acquisition (van Gelderen 2018) starts with this. Psych-verbs will not be discussed here.
The outline is as follows. Section 2 reports earlier results from van Gelderen (2018) that show that telic and durative verbs generally keep their original lexical aspect. It also introduces the Sorace Hierarchy that shows a gradient split between unergatives and unaccusatives. Section 3 Iists systematic reports in the literature (e.g. Rosen 1984) that certain verbs are unergative in one language and unaccusative in another or that one verb includes both a durative and telic meaning (e.g. Levin & Rappaport & Hovav 2014). After these reviews, sections 4 and 5 discuss a few instances of ambiguous Aktionsart, namely verbs of controlled motion and of continuation. With controlled motion, the conclusion is that, for `swim', the agentive feature comes to predominate whereas, for `speed', the telic one does. Verbs of continuation are stative and therefore do not involve a telic feature but their agentive feature is not strong either. Of these, `float' is ambiguous and `remain' unaccusative.
The data discussed in this paper are obtained from dictionaries, i.e. Bosworth & Toller (B&T), Oxford English Dictionary (OED), Middle English Dictionary (MED), and Dictionary of Old English (DOE), and from corpora, e.g. Corpus of Contemporary American English (COCA) and Corpus of Historical American English (COHA). The sources used in these corpora and dictionaries will not be cited in the references. Perlmutter (1978) makes a very important distinction between intransitives that have come to be called unergative and those that have come to be called unaccusative. Unergative verbs have Agents that are animate and act deliberately whereas unaccusative verbs have Themes that are either animate or inanimate and undergo the action. The characteristics listed in Table 1 present well-known contrasts between these two verb types in Germanic and Romance. have perfect c.

The aspectual stability of intransitives
Theme can be added (e.g. She danced a dance) -er can generally not be added (e.g. *arriver) be perfect ( in older English, German, Dutch, Italian) Theme cannot be added (e.g. *The bus arrived a passenger) Dowty (1991) characterizes unergatives as atelic and unaccusatives as telic and Tenny (1987: 264) writes that unergatives "tend to describe non-delimited events, while unaccusatives tend to describe delimited events". In this way, the choice of argument is connected to that of lexical aspect. Likewise, many other scholars see the unergative-unaccusative distinction as completely aspectual, e.g. Zaenen (1988). Abraham (1990) uses the terms terminative/biphasic versus monophasic to distinguish two crucial classes, which I will refer to as telic versus durative.
Unergative verbs have durative aspect with an obligatory Agent and, in the history of English, acquire transitive-like properties, keeping their Agent and durative aspect, while allowing the possibility to add a Theme. This is shown in the change from the earlier intransitive `climb' (1), the only Old English occurrence of climb in the OED, to the later transitive (2) use. (1) Gif hit unwitan aenige hwile healdað butan haeftum, If it (light) unwise any time holds without fetters hit ðurh hrof waedeð, bryceð and baerneð boldgetimbru, it through roof wades, breaks and burns timbers seomað steap and geap, stigeð on lenge, clymmeð on gecyndo hangs steep and high, rises in length, climbs in nature. If an unwise person holds it (light) without bounds, it will go through the roof and break and burn the timbers (of a house); it hangs steep and high and rises and climbs in nature.' (DOE, Solomon and Saturn 412-6, Dobbie, 1942) (2) a.  (3), and develop into causatives by adding a Causer in (4). The unaccusative meaning is also kept, as in (5).
(3) aefter gereordunge hi aemtian heora raedingum oððe on sealmum. after repast they empty their readings or (spend it) on psalms 'After repast, they free themselves for readings or psalms.' (Rule of St Benet,Logeman,48,82.13) (4) Hugo empties his pockets of screws, springs, and other tiny metal pieces. (COCA 2012) (5) His eyes emptied, his jaw went slack and he mumbled. (COCA 1993) The unaccusative verb arrive `to come to shore' is a loan from French and is initially both unaccusative, as in (6) and (7), and causative, as in (8). The causative use of arrive remains until the 1650s but then, like other change of location verbs, becomes impossible. The basic aspect and Theme-focus remains the same. So far, the durative aspect has been seen to go with an Agent whereas telic aspect needs a Theme. Of course, the distribution in two classes of intransitive verbs is more complex. Sorace's (2000) Hierarchy considers seven classes of verbs that form a continuum between unaccusative at the top of Table 2 and unergative at the bottom. I have added the aspect in the right hand column and this shows the three basic ones, telic, stative, and durative.  Sorace (2000) and Keller & Sorace (2003) recognize core from non-core unaccusative and unergative verbs: the core ones consistently display the characteristics that are expected according to Table 1. In Table 2, the verbs in the top row (`Change of Location') and in the bottom row (`Controlled non-motional process) represent the core unaccusatives and unergatives, respectively. Other verb types, e.g. controlled motional process and the continuation of a pre-existing state do not. I will discuss two of each of the non-core groups that Keller & Sorace identify as not conforming (bolded in Table 2), namely swim and speed in the controlled motion group and float and remain in the continuation group3. Although telicity is crucial in distinguishing the core classes, motion can be telic or not, thus making a controlled motional process less of a core unergative. In the same way, telicity is less relevant in verbs of continuation of existence. Accordingly, these verbs are ambiguous unless an agentive feature is explicitly added.
In the course of the history of English, unaccusative verbs not only acquire a Causer theta-role, but they are also develop into copular verbs. Because the Theme theta-role is central to unaccusatives, i.e. the intransitives in the upper half of Table 2, these verbs are typically reanalyzed by language learners as copulas. The verb become is originally a telic unaccusative in (9) but it is later used as a (telic) copula in (10).

(9)
Hannibal to þam lande becom. unaccusative Hannibal to that land came 'Hannibal came to that land.' (OED, Orosius, Bately 100.17) (10) [Of] tristrem and hys leif ysote, How he for here be-com a sote copula `About Tristram and his love Isolde, how he become a fool for her.' (Cotton,(17)(18)also in Fairfax) Having shown in this section that intransitives, like climb, empty, and become, don't change their durative or telic aspect when they develop into transitives and copular verbs, I now first turn to claims in the literature that some intransitives are durative in one language but telic in another and that verbs can be poly-aspectual before looking at examples of these in the history of English. Rosen (1984) argues that languages differ as to how the intransitive class is divided. Using Davies' (1981) work on Choctaw, she compares Choctaw and Italian `die' and `sweat'. The former is unergative in Choctaw but unaccusative in Italian whereas the latter is unergative in Italian and unaccusative in Choctaw. Similarly worrisome, verbs of uncontrolled process, such as sneeze and snore, are unergative in Dutch and Italian, but unaccusative in Eastern Pomo. In response to Rosen, Sorace (2000: 877) argues that languages make different decisions about verbs like `sweat',`sneeze', and `snore' because, as uncontrolled process verbs, they are in the middle of the Hierarchy in Table 2 and can be more or less agentive. She doesn't discuss the change of state verb `die' but Dąbrowska (2015) shows that English die fails many of the typical unaccusativity tests and argues it is not an unaccusative verb. I therefore assume there is some crosslinguistic variation with verbs of uncontrolled process and, with Dąbrowska, that `die' is unclear as to what category if belongs to.

Cross-linguistic differences
Rosen also argues that the "class of motion verbs presents special complications" (Rosen 1984: 66) because some express manner of motion and others result. Not only do some express manner in one language and result in another, as also known from Talmy's (1985) work, but specific verbs can be coerced by adding adverbials of manner or result. Dutch lopen `walk' is durative in (11), shown by the auxiliary `have', but it can be coerced into a telic verb in (12), in which case the auxiliary changes to `be'.
Dutch unergative coerced to telic through goal I am there walked `I walked there.' Table 2 and the cut-off point may be different. The verb `sweat' is an uncontrolled process and can therefore either be seen on the unergative side (if process is emphasized) or on the unaccusative side (if the lack of an agent or the presence of a goal is emphasized).

Languages indeed choose a boundary between the verbs in
Another example of a verb that may have two meanings that differ aspectually is clean. Levin & Rappaport Hovav (2014) point out that this verb is ambiguous between a manner (i.e. durative) reading, as in (13), and a result (i.e. telic) reading, as in (14). (13) He was cleaning after he cooked (but it remained dirty). (14) The machine cleaned the carpet beautifully.
Note that in (13), there is just an Agent, not a Theme, and a progressive -ing coercing it towards manner/ telic aspect. In (14), there is a Causer and Theme with a telic reading. I will argue that the basic aspect of clean is telic; the history of English sheds light on this topic, as it will on some other verbs in sections 4 and 5. Old English has a verb claensian `make clean' that Modern English still retains as cleanse but which has narrowed its meaning to `make completely clean'. The verb clean is a relatively recent addition, according to the OED. Apart from a mention in a lexicon from the 1450s, the earliest examples are given in (15) and (16) and these are all telic, thus suggesting that this aspect is basic.
The English Frigats are now so well Fitted and Cleaned, that none of the Algerines they meet with, escape them. (OED, 1681, London Gaz. No Looking in COHA, the frequency of the verb clean increases dramatically after this introduction. In the period from 1810 to 1840, there are 59 instances of the verb clean and they are all telic; from 2000 to 2010, there are over 10 times as many such verbs (750) and most remain telic. Data from first language acquisition confirms this telic basic aspect. Eve's files (Brown 1973) show seven instances of the verb clean and they are all telic, as in I clean it off, you clean it off, and we better clean them, between the ages of 1;6 and 2;3. It therefore appears that the telic meaning is relatively basic. In short, clean is a verb that is a telic verb, due to its adjectival origin, but which can be coerced into a manner meaning, which is durative. Randall et al. (2004) show that speakers and learners seem to operate with a system of features in which certain features are ranked over others in a particular language. German, for instance, uses locomotion to distinguish unergatives and unaccusatives but Dutch does not. Below, we will see that, in the history of English, the feature agentivity is more prominent with the verb `swim' but not with the other verbs of the same type. In earlier but related work, Keller & Sorace (2003) conduct a series of experiments testing German native speakers on the `have' versus `be' distinction with the seven verb groups presented in Table 2. As mentioned in section 2, they find two unexpected verb classes: verbs of continuation (`stay', `survive') prefer `have' and verbs of controlled motion (`walk', `swim') tend to be combined with `be'. Accordingly, rather than just invoking telicity for the higher group and agentivity for the lower group, there are other factors, namely motion. These basic concepts are important parts of the Faculty of Language and may be pre-linguistic.
Because I argue that argument structure is pretty uniform, in this section, I have discussed a few verb types that seem to differ cross-linguistically in their Aktionsart. First, the verbs of uncontrolled process are genuinely ambiguous between durative and telic, which is not unexpected from their position in the Hierarchy. Second, the verb `die' may not fit in this hierarchy, for unclear reasons. Third, certain motion verbs express either goal or manner and, last, seemingly ambiguous verbs, like `clean', still have a basic aspect, namely telic. In the remainder of this paper, four Old and Middle English verbs from two classes of verbs in Table 2 will be examined to see how they develop and which features seem to play a role.

Origin and changes in verbs of controlled motional process
In this section, I look at two intransitive verbs that are ambiguous between unaccusative and unergative meanings and how that plays itself out in language change. I selected swim and speed from this group that also includes walk, run, crawl, wade, and stride. Swim is a non-core unergative in Table 2, i.e. a controlled motional process.4 I start by listing all Old English meanings of this verb and then turn to another verb in this group, speed, which originates in Middle English and for which I examine some MED examples. Buck (1949: 680) writes that notions of `float, swim, and sail' are closely related in Indo-European and this carries over into Old English. Thus, Bosworth & Toller's (B&T) Old English Dictionary lists three main uses of swimman `swim': (a) "of living creatures moving in or on water", (b) "of a vessel moving on water", and (c) "of lying on the surface of water". This definition includes agentive and non-agentive meanings, i.e. verbs with an Agent theta-role (involving a durative manner verb), as in (17), verbs with a Theme theta-role (involving a continuation of state rather than a telic change of location), as in (18) There is literature on the typology of verbs of aquamotion, e.g. Koptjevskaja-Tamm et al. 2010. I haven't considered this because those verbs are from all classes in Table 2 and I wanted to specifically look at the non-core verbs.  (17) to (19), the other instances of the Old English verb swimman appear in (A1) to (A20). I have organized them according to the three descriptions in B&T with a fourth category for verbs that seem to fall outside these definitions. Examples from the same text are grouped together.
Examples (A1) to (A13) are agentive and durative, while (A14) is agentive with coerced telicity (through `out'). Evidence for the unergativity of this group is, for instance, the perfect auxiliary have in (A8b), and a deontic modal in (A14b) and (A14c), typical of an Agent. The second group has only two known instances, (18) above and (A15). The context of (A15) makes this agentive, as does the deontic modal mihte. The third group includes the examples in (A16) to (A18). Many have a deontic modal, e.g. all instances in (A16) have wille. This may be to coerce a more agentive meaning of the basic `continue to be on the water'. The last group includes (A19), for which I am not sure if `swim' or `float' is the appropriate translation and (A20) which is a metaphorical use of `float'.
The results for the four meanings of Old English `swim' are summarized in Table 3. Meaning (a) is the durative, agentive meaning and its use predominates. Meanings (b) and (c) are related, a Theme is floating with no goal reached; some are coerced into agentive meanings. Taking that coercion into account, 23 are durative, 3 telic (A14abc), 3 stative (marking a continuation of state), and 2 are unclear.   Table 3 shows that an agentive meaning predominates, which means a durative aspect prevails. That `swim' is unergative in Old English is corroborated by a number of other phenomena (of Table 1). There is one instance, i.e. (A8b), of a `have' auxiliary, as mentioned, and none with `be'. In addition, according to B&T, there are also a number of adjectives with a meaning of `able to swim', e.g. swimmendlic and sund. These are typically agentive. The aspectual meaning doesn't change much in the Middle English period as evidenced in the MED. The MED keeps the durative/agentive category (a) but collapses the Theme-based classes (b) and (c) and adds a new third category, namely `have an abundance of.' By 1600, the durative, agentive meaning of `swim' prevails, however. For instance, Shakespeare's occurrences with swim are durative, as shown in (20) and (21), except for one which involves a vessel in (22)  The agentive suffix -er occurs twice, as in (23), another indication of basic unergativity.
As two spent Swimmers, that doe cling together. (Shakespeare, Macbeth 1,2,8) In short, the verb swim is initially unergative with an Agent although, infrequently, it has a Theme without a telic aspect. It changes at the end of Middle English into a mainly durative, agentive, unergative verb. The next verb in this group, speed, is initially ambiguous but then ends up more telic. In Old English, the intransitive (ge)spedan is not a motion verb; it has the meaning of `to succeed, prosper', as in (24) (DOE, Ker, 1957 331) In Middle English, it comes to be used as a motion verb in (26) and (27), and would be classified as `controlled process' in Table 2, a manner of motion verb with the meaning `to travel swiftly'. As in the case of `swim,' we would therefore expect some ambiguities because motion may imply a goal and this is the case. From the use of the causative in (26), it seems telic (unaccusative/causative) and that is emphasized by the particle ut. However, there is a perfective auxiliary have, more indicative of unergative. The sense in (27)  In Modern English, this verb is still ambiguous, unergative in that speeder exists in (28) and unaccusative in that many instances of causatives, as in (29) and (30) Looking at the first 50 (relevant) instances of the verb `speed' in COCA, we find the results in Table 4. Since `swim' and `speed' are motion verbs, I should say something on the relevance of the work of Talmy (1985), Fanego (2012), and Huber (2017). Talmy classifies languages as to whether their motion verbs encode the path or the manner of an event and Fanego and Huber show that English, throughout its history, has had more manner of motion verbs with the path being indicated by an adverb or particle. This classification emphasizes duration and telicity but ignores agentivity, which appeared to be very relevant in Table 2, and is therefore not further explored. In short, two verbs from the non-core unergative group of controlled motional process in Table 2, swim and speed, show variable behavior in the history of English, the former settling on an unergative meaning and the latter on an ambiguous Modern English pattern, although with the unaccusative meaning being dominant.

Origin and changes in verbs of continuation
In this section, I examine two verbs from the group of continuation of a pre-existing state, namely float and remain. This group also includes last, stay, survive, persist, persevere, stand, lie, and rest. As with the previous two verbs, the first one appears in Old English whereas the second only does in Middle English. For these verbs I have looked at all instances in the DOE but have not listed them all.
A word closely related in meaning to swim is float, flotian in Old English. Bosworth & Toller's definition for flotian is simply `to float.' In Old English, this verb is unaccusative in (31) in having a Theme and a Goal, but it is also agentive, in (32). It develops causative uses in (33), as expected if it is unaccusative. The basic meaning of the verb is continuation of a state, i.e. not very telic, as in (34) and (35), and some are ambiguous between `float' and `swim', as in (36), which is emphasized by the deontic modal meahte.  (37) and (38)  field the people of Persia thus began take.heart `They went as quick as deer inside and moved as fire in a field and the people of Persia began to take heart.' (MED, c1400 Kyng Alisaunder 2436 This ambiguity between Agent and Theme as theta-role for the intransitive `float' is not unexpected from verbs with a meaning of `continuation of a pre-existing state'. Although Keller & Sorace (2003) have shown that there is a preference for the auxiliary `have' in German and Dutch, there are very few agentive uses of floater in English, as in (39). we can't put all of our eggs in a basket that might float down the river (COCA 2017) Another unaccusative verb like float is remain in that its focus is a Theme argument, not an Agent, but it is stative, not telic, and in that sense is not a core unaccusative. The earliest examples from the MED and OED are from late Middle English, provided in (41) and (42) I have not found instances of this verb used with the auxiliary `be', which would be indicative of unaccusative use, or of -er, indicative of unergative use. This verb also develops into a copula in the 16 th century, as (46) shows, and stays quite stable in its stative aspect. The copula use now predominates in modern corpus data, e.g. COCA.
In this section, I have looked at float and remain to see if they are aspectually ambiguous. Both are mainly unaccusative, focussing on a Theme, but the former includes motion that could be either durative or telic.
With remain, motion is not relevant, making this verb more clearly unaccusative.

Conclusion
Intransitives can be divided into unergatives and unaccusatives but, as many people have shown, the division is gradient, with some verbs displaying core and some non-core characteristics. In this paper, I have examined two verbs each from two non-core verb groups. The results corroborate the work in Keller & Sorace (2003) who show that native speakers of German have variant `have'/`be' selection (and impersonal passive use) with these verbs as well as cross-dialectal differences. Among the verbs of controlled motion, the diachronic data show the durative `swim' initially to be somewhat ambiguous between unaccusative and unergative but it settles on the latter in the modern period, which is consistent with its basically durative aspect. The verb `speed' is initially ambiguous and continues to be although the telic sense is more prevalent. The verbs of continuation that are examined are float and remain. Of these, float may involve motion so is more ambiguous than the more clearly unaccusative remain. The reason behind the instability of these two verb classes has been touched upon above: controlled motional process verbs (unlike non-motional ones) have movement inherent in their meaning with an emphasis on a moved element and a possible Goal. This makes that class of verbs more like unaccusatives. Stative verbs of continuation by definition reach no goal but involve no Agent either. The features that play a role in dividing the unergative and unaccusative class involves: Agent, Theme, Goal (telic), and motion. The text in which example (18) appears has an alternative with `float' just one line before, namely (i).