2. Encoding directed motion with one verb only

We first consider the encoding options available for describing directed motion events in clauses with only one verb. We begin in section 2.1 with constructions in which the verb encodes manner, and enumerate the possibilities for how path can be encoded. We also highlight non-motion uses of some of these path encoding resources, especially in encoding aspect and result. In section 2.2 we turn to constructions in which the verb encodes path, and enumerate the ways in which manner is encoded, again drawing parallels with non-motion events. These two classes of data constitute the primary evidence used by Talmy and others to support a two-way typology. This is unsurprising, since if a clause has only one verb which can encode either path or manner, but not both, two language types should arise. In section 2.3 we summarize and review a range of attested motion event encoding options with a single verb that do not clearly fit Talmy's typology, mostly involving types of satellites (in the broad sense), which again have non-motion uses. The picture that emerges is that (a) motion event encoding devices are rarely if ever dedicated only to encoding such events and (b) the range and diversity of exceptional data is sufficient to warrant a larger reassessment of the two-way typology.

Before we begin, we briefly reevaluate Talmy's notion of satellite. We start with his definition:

Satellites on this conception include English particles, Germanic (in)separable prefixes, Russian verb prefixes, Chinese coverbs, and Atsugewi non-inflectional affixes. For example, Talmy considers each element marked by a ◂ to be a satellite in the following English sentence:

1. (4) Come ◂right ◂back ◂down ◂out from up in there!

   (Talmy Reference Talmy and Shopen1985: 102, ex. (60))

Crucially, the satellite is distinct from the preposition, which often occurs with the satellite or verb and takes as its object the ground element with respect to which the path is defined. For example, in (5) out is a satellite, while of is a preposition (indicated by >by Talmy):

1. (5) I ran ◂out of >the house.

   (Talmy Reference Talmy and Shopen1985: 103, ex. (62a))

Talmy (Reference Talmy and Shopen1985: 105) proposes a single diagnostic for distinguishing a preposition from a satellite: the ground is optional with a satellite, but not with a preposition. Thus, some English satellites also double as prepositions (e.g. in (the house), on (the roof)), while some prepositions are only prepositions (e.g. into *(the house)), and some satellites are only satellites (e.g. forth (*the house)). A satellite, then, is a sister to the verb root and does not require the obligatory presence of a ground element. Although the directional affixes of some languages may meet this criterion, we now suggest that more generally it fails to pick out a natural class of elements across languages. (For further, partially overlapping discussion see Filipović Reference Filipović2007: 33ff.)

First, the English elements that Talmy labels satellites are not always sisters to the verb, at least not to the exclusion of the ground. Applying the it-clefting constituency test to (5) shows that the satellite+ground PP combination out of the house is a constituent excluding the verb:

1. (6)

   1. (a) ?It was out of the house that I ran, not into the house.

   2. (b) *It was out that I ran of the house, not in.

The slight oddness of (6a) most likely arises because goal phrases are preferred right after manner-of-motion verbs (Nikitina Reference Nikitina and Asbury2008). In the comparable sentences with a path verb in (7) the different status of the two sentences with respect to it-clefting is clear:

1. (7)

   1. (a) It was out of the house that I went, not into the house.

   2. (b) *It was out that I went of the house, not in.

By this diagnostic, out of the house is a constituent. Thus, out alone is not a sister to the verb root; rather, of the house is a complement of out, and the entire PP out of the house is a sister to run, effectively nullifying the distinction between satellites and prepositions. In fact, particles have been subsumed under the class of prepositions, as in Jackendoff's (Reference Jackendoff, Anderson and Kiparsky1973) proposal that they are ‘intransitive’ prepositions; see also Emonds (Reference Emonds1972) and Svenonius (Reference Svenonius, Reuland, Bhattacharya and Spathas2007), among others.

Second, it seems semantically unmotivated to distinguish obligatory vs. optional ground elements. In John ran in, though a specific ground is not expressed, one is understood. In (8), both in (the house) (a satellite+ground) and to the store (a preposition+ground) indicate the goal of motion and often they are apparently alternate expressions of the same semantic content (Nikitina Reference Nikitina and Asbury2008):

1. (8)

   1. (a) John ran in (the house).

   2. (b) John ran to the store.

However, for Talmy (8a) and (8b) represent typologically distinct methods of encoding path: as a satellite and as non-verb non-satellite, suggesting that in addition to V-framed and S-framed languages, there might also be A(dposition)-framed languages, something surely not intended. Thus, we suggest that PP not be excluded from the notion of satellite, thereby recognizing a wider range of path encoding options than under a strict interpretation of Talmy's typology.

Talmy's definition also has implications for manner satellites. Talmy (Reference Talmy2000: 222) notes that V-framed languages ‘map [manner] either onto a satellite or into an adjunct, typically an adpositional phrase or a gerundive-type constituent’. For example, Talmy (Reference Talmy and Shopen1985: 111) contrasts manner satellites in Nez Perce, which are verbal affixes, with gerundive manner clauses in Spanish:

1. (9) ?ipsqi- ‘walking’, wilé·- ‘running’, wat- ‘wading’, siwi- ‘swimming-on-surface’, tuk ^we- ‘swimming-within-liquid’, we·- ‘flying’

   (Nez Perce – cf. Talmy Reference Talmy and Shopen1985; 111, ex.(82))

1. (10) Entró corriendo/volando/nadando/… a la cueva.Footnote ⁵ entered.3sg.past running/flying/swimming/… to the cave ‘S/he entered running/flying/swimming/… the cave.’

   (Spanish – cf. Talmy Reference Talmy and Shopen1985: 111, ex. (83))

However, there is no apparent reason to distinguish the two: syntactically neither is the main verb (root) and semantically both indicate manner. Thus, in what follows, we employ the term ‘satellite’ in a broader sense: any constituent that is sister to or adjoined to the verb (root). When this notion and Talmy's narrower notion need to be distinguished, we will indicate it overtly.

3. Encoding directed motion with two or more verbs

We turn now to monoclausal constructions with more than one verb. Languages with such constructions are not accommodated by Talmy's typology, so it is not surprising that some researchers (Slobin Reference Slobin, Strömqvist and Verhoeven2004b, Zlatev & Yangklang Reference Zlatev, Yangklang, Strömqvist and Verhoeven2004) posit a third class of E(quipollently)-framed languages to deal with a subset of such constructions; however, as we show, even this elaboration of the typology is empirically inadequate. We look first at canonical E-framed languages – in particular serial verb languages – and then at other types of multiverb constructions found across languages.

4. The role of morpholexical and morphosyntactic resources

The data surveyed in the previous sections show a wide variety of encoding possibilities for motion events that do not fit comfortably into a two- or three-way typology. They also show that some options involve one motion verb, while others involve two or more. The former are the focus of Talmy's work; the latter have figured in work that extends his typology. The question is why the number of verbs should play a role in determining the available encoding options for directed motion events. We argue that constraints as in (3), repeated here, shed light on this question:

1. (44)

   1. (a) Verb is the only clause-obligatory lexical category.

   2. (b) A verb may lexicalize only one of manner and path.

The requirement in (44a) reflects the more general requirement that all main clauses contain a verb (excepting copular constructions in some languages). Similarly, (44b) is an instance of a more general constraint on how much and what type of semantic information can be packaged into a verb meaning. Following Levin & Rappaport Hovav (Reference Levin, Hovav, Spencer and Zwicky1998) (see also Pinker Reference Pinker1989 and Grimshaw Reference Grimshaw2005), a verb's meaning can be thought of as being composed of two distinct facets. One is an event schema, built from a small, universal set of primitives (e.g. causation, process, change-of-state, change-of-location, existence) that represent the verb's basic event type. The second facet, and the one relevant here, is some idiosyncratic semantic material, now often referred to as the ‘root’ (after Pesetsky Reference Pesetsky1995), which crucially distinguishes among semantically related verbs. Roots fall into a limited set of ‘ontological types’; two of the most important are manner – an indication of how a particular action is performed – and result – an indication of the result state or location of the action.

Moreover, various researchers have argued that the categories of result and goal (a subpart of a path) are manifestations of a single more basic category, or perhaps reducible to one another. Talmy (Reference Talmy2000: chapter 3) himself sees both as types of Core Schema in an event, loosely the component that determines the event's temporal structure. More specific arguments that goal and result represent a single category are based on their comparable contributions to the aspectual properties of the predicate, including telicity (Tenny Reference Tenny1987, Reference Tenny1994; Dowty Reference Dowty1991; Krifka Reference Krifka and Rothstein1998; Hay et al. Reference Hay, Kennedy, Levin, Matthews and Strolovitch1999; Rappaport Hovav & Levin in press) and durativity (Wechsler Reference Wechsler2001, Reference Wechsler, Erteschik-Shir and Rapoport2005; Beavers Reference Beavers2002, Reference Beavers2006, Reference Beavers, Dölling, Heyde-Zybatow and Schäfer2008b). A second similarity comes from argument realization: figures of motion events and patients of change-of-state events tend to be realized as direct internal arguments (Rappaport & Levin Reference Rappaport, Levin and Wilkins1988, Dowty Reference Dowty1991, Baker Reference Baker and Haegeman1997, Krifka Reference Krifka and Rothstein1998, Beavers Reference Beavers2006), while paths and results are realized as obliques or as secondary predicates, as discussed in section 2.1. Indeed, this similarity is a key motivation for the localist hypothesis (Gruber Reference Gruber1965; Lyons Reference Lyons1967; Anderson Reference Anderson1971; Jackendoff Reference Jackendoff1972, Reference Jackendoff1983; DeLancey Reference DeLancey2000), although this hypothesis takes path as basic (as Talmy Reference Talmy2000 seems to do).Footnote ¹³ Thus, following the common assumption that coming to be in/at a location is like coming to be in/at a state and vice versa, (44b) is just an instance of a more general constraint proposed by Levin & Rappaport Hovav (Reference Levin and Hovav1991, Reference Levin, Hovav and Roca1992) and Rappaport Hovav & Levin (in press) that while a verb root may lexicalize manner or result, it may not lexicalize both simultaneously;Footnote ¹⁴ a separate statement is not needed for motion verbs. Assuming that languages can lexicalize only one of manner or result in the verb, a two-way typology is the logical outcome for sentences with one verb, explaining the appeal of Talmy's typology. However, once languages with multiverb constructions are taken into account, positing a third class of E-framed languages appears to be a natural next step.

Furthermore, the goal of a motion event can be expressed using any encoding option that can convey the appropriate semantics. Most obviously, goals are expressed using dedicated goal markers, such as English to, but there are two alternative semantic notions that allow for goal construals, giving rise to alternative expressions of this notion: boundary and location. The notion of boundary was introduced in section 2.3.1, in the context of until-markers. When occurring with a spatial complement, such markers, which are neither aspectual nor result-denoting in nature, are understood as contributing a goal because the spatial complement must be understood as a boundary. In addition, as discussed in section 2.1.2 some apparent goal markers are in fact markers of location. The use of location markers to express goals is not surprising since a goal is still fundamentally a location, albeit the final location in a motion event. This characteristic of goals is reflected in analyses that decompose directional adpositions into layered PPs with a directional head selecting for a locative head (see van Riemsdijk Reference Riemsdijk, Pinkster and Genee1990, Rooryck Reference Rooryck, Thráinsson, Epstein and Peter1996, Koopman Reference Koopman and Koopman2000, den Dikken Reference den Dikken2003, Svenonius Reference Svenonius, Reuland, Bhattacharya and Spathas2007, van Riemsdijk & Huijbregts Reference Riemsdijk, Huijbregts, Karimi, Samiian and Wilkins2008, among others). Thus, the expression of goals as boundaries or locations represent other, less recognized, semantic perspectives on the notion of goal, which allow for additional encoding options. Looking at paths from this perspective, we see that paths are expressible as property scales that measure changes, entities with physical extent, or a series of locations. Again, there is no reason to treat path as a unique category subject to unique constraints.

With this background, we turn to the consequences of (44) for motion event encoding. The options for expressing a given event in a given language fall into two main classes: manner in the verb or path in the verb (a third class is discussed below). Each determines a different set of possibilities for encoding or combining both manner and path in a clause (setting aside goals as locations until section 5):

1. (45)

   1. (a) Path as V: If path is expressed in V for a given expression, then

      1. (i) if the language has monoclausal multiverb constructions, manner may also be expressed as a V.

      2. (ii) if the language has manner adverbials (ideophones, subordinate clauses, adverbs), these may encode manner.

   2. (b) Manner as V: If manner is expressed in V for a given expression, then

      1. (i) if the language has monoclausal multiverb constructions, path may also be expressed as a V.

      2. (ii) if the language has appropriate result satellites (affixes, applicatives, semantic cases, adpositions, particles), these may encode path.

      3. (iii) if the language has until-markers, these may be used to encode path.

The encoding of the meaning component not expressed in the verb depends on available language-specific resources for encoding and combining manners and results in a clause. These resources need not be specific to motion events. For example, French and Japanese share a number of crucial properties regarding morpholexical and morphosyntactic inventories. Both lack applicative morphemes (cf. Tswana), aspectual affixes (cf. Russian), particles (cf. German), semantic cases (cf. Finnish), bipartite verb stems (cf. Klamath), and result satellites (cf. English). Therefore, there are only two possibilities for encoding path left in both languages from those discussed in sections 2–3: path verbs and until-markers. Likewise, there are only two options in both languages for encoding manner: manner verbs and subordinate adverbial clauses. Both languages exploit all four options.

We cannot, however, predict which options will be available in a given language. For example, either French or Japanese could have lacked until-markers (though we know of no such languages) or path verbs (as may be the case in Russian). Nor can we predict which resources of the ones available will actually be employed for encoding motion events. Although Japanese allows until-markers to encode path, the availability of until-markers is not sufficient for their use in path encoding. For example, few English speakers accept until in the expression of goals (cf. John strolled to/??until the park). Furthermore, even if a particular resource is both available in a language and exploited in its motion constructions, this does not predict how it interacts with other resources. Rather, the available set of combinatorial processes for putting these resources together represents a further dimension of variation (Bouchard Reference Bouchard1995; Pustejovsky & Busa Reference Pustejovsky, Busa, Bertinetto, Bianchi, Higginbotham and Squartini1995; Cummins Reference Cummins1996, Reference Cummins1998; Song & Levin Reference Song and Levin1998). For example, we have shown that at least the following combinatorial options are exploited by different languages for encoding manner and path without using path satellites (in the narrow, Talmyan sense):

1. (46)

   

Again, the possible options for encoding motion events are determined by general properties of a language; they are not specific to these events alone. For example, despite their very similar morpholexical inventories, Japanese allows VV compounds and V-te-V complex predicates, while French does not. Thus, Japanese allows V- and E-framed options, while French only has the former, although both languages also have until-markers, a type of S-framed option. Thus, this difference has significant consequences for the encoding options available to each language.

Interestingly, depending on the resources available to it, a language may even allow both canonical S- and V-framed constructions. For example, both English and Hebrew (the latter sometimes classified as V-framed; Slobin Reference Slobin, Strömqvist and Verhoeven2004b) have manner verbs, path verbs, manner adverbial participles, and goal adpositions, yielding both canonical encoding types, as shown in the Hebrew examples (47) and their English translations.

1. (47)

   1. (a) ha-kelev zaxal la-meluna. the-dog crawled to.the-doghouse ‘The dog crawled into the doghouse.’

   2. (b) ha-kelev nixnas la-meluna bi-zxila. the-dog entered to.the-doghouse in-crawl_N ‘The dog entered the doghouse crawling.’

      (Hebrew – Itamar Francez, p.c.)

Indeed, in addition to the options in (45), there is a seldom discussed third option: encoding neither manner nor path in the main verb, but rather encoding both as satellites. English instantiates this option as in (48a) (cf. the more commonly discussed options in (48b, c)).

1. (48)

   1. (a) John moved stealthily out of the bedroom.

      (manner=adverb, path=adposition)

   2. (b) John stole out of the bedroom.

      (manner=V, path=adposition)

   3. (c) John left/exited the bedroom stealthily.

      (path=V, manner=adverb)

This third possibility arises simply from the resources available in English, including path, manner, and pure motion verbs, plus result and manner satellites.

What emerges is a more varied picture of motion event encoding both within and across languages. Languages will share similarities in how motion is encoded only in as much as they share types of manner, result, boundary, and argument marking resources and combinatorial processes for putting the pieces together. We should not, then, expect a small number of language types with respect to motion constructions, but rather as many types of languages as there are combinations of relevant resources, and indeed the data discussed in the preceding sections attest to this. Despite this, not all options for motion event encoding have equal status, as we discuss in section 6. First, though, we expand our approach to include non-grammatical factors that contribute to motion event encoding.

5. Locations with directional interpretations: the role of pragmatics

As discussed in section 2.3.3, the literature on V-framed languages includes a growing number of examples of manner verbs with PP dependents which are used to encode directed motion events – an apparently S-framed encoding strategy. We take as our starting point two factors which allow us to compare these data directly to comparable examples found in S-framed languages. First, researchers who note such examples in V-framed languages often simultaneously note that they involve only a handful of manner-of-motion verbs (Alonge Reference Alonge, Agostiniani, Bonucci, Giannecchini, Lorenzi and Reali1997, Folli & Ramchand Reference Folli, Ramchand, Verkuyl, de Swart and Hout2005, Zubizarreta & Oh Reference Zubizarreta and Oh2007, Folli Reference Folli and Asbury2008, Gehrke Reference Gehrke2008), and second, the PPs in these examples are invariably locative in nature (see section 2.1.2). These two factors are also relevant for S-framed languages, where the directional interpretation of locative PPs is often attributed to pragmatic inference involving the verb, the locative adposition, and context. We suggest that the same explanation is available for the V-framed language examples: directional interpretations of locative adpositions should be available with the appropriate pragmatic support even in the absence of morphosyntactic devices for directly expressing direction in a PP. (See also Levin et al. Reference Levin, Beavers and Tham2009 for further developments of this proposal.)

Studies of English and some other S-framed languages have pointed out that in some circumstances locative phrases can be understood as goals rather than locations, both with directed motion verbs and with manner-of-motion verbs (see Thomas Reference Thomas2004, Gehrke 2007, and Nikitina Reference Nikitina and Asbury2008 on English; Gehrke 2007 on Dutch; Biberauer & Folli Reference Biberauer, Folli, Courzet, Demirdache and Wauguier-Gravelines2004 on Afrikaans; Tungseth Reference Tungseth, Kempchinsky and Slabakova2004, Reference Tungseth2008 on Norwegian; Israeli Reference Israeli2004 on Russian; Nedashkivska Reference Nedashkivska2001 on Ukrainian). In English, for example, in and on are locative, contrasting with into and onto, which are inherently goal-markers (although see footnote 4). Yet in and on may receive goal interpretations in certain contexts. For example, such an interpretation is available for (49a) if John is standing just outside the room and for (49b) if Kim is standing next to the bed.

1. (49)

   1. (a) John walked in the room.

   2. (b) Kim jumped on the bed.

Specifically, locative phrases are understood as goals precisely in those contexts that allow a reader or hearer to infer that a goal interpretation is intended. Evidence comes from an extensive corpus study by Nikitina (Reference Nikitina and Asbury2008), which identifies some contextual factors that facilitate a directional interpretation of in. For example, a goal interpretation of the PPs in (49) is unavailable if John or Kim was standing some distance from the relevant location (e.g. down a long hallway) (and similarly of course for a location interpretation).

Furthermore, Nikitina points out that verbs that are inherently punctual and thus naturally describe a transition are more likely to be found with in goal PPs than verbs that describe a process. In particular, in is found less often with manner-of-motion verbs (which tend to describe processes with duration) than with directed motion verbs (which are more likely to allow punctual, transition readings); see also Thomas (Reference Thomas2004). Interestingly, the manner-of-motion verbs that Gehrke (2007, Reference Gehrke2008) cites as showing the comparable phenomenon in Dutch are punctual, while those that disallow it are durative.Footnote ¹⁵ Turning to the complements of such prepositions, in is found more often with ‘containers’ – locations with well-defined boundaries, such as rooms, pools, boxes, and cars – than it is with ‘areas’ – locations that lack such boundaries, such as forests, neighborhoods, fields. As Nikitina points out, it is more plausible to infer a punctual transition into a container than an area, thus allowing for a focus on the result location rather than on the extended path of motion. These semantic effects are clear evidence against treating prepositions in English such as in as having both locative and directional readings, since then lexical ambiguity would be expected to be more consistently available. The directional interpretation of these PPs, then, is better understood by positing a pragmatic basis.

On a pragmatic account, we expect that location phrases should have comparable directional interpretations in path languages, as long as contextual support is available. We argue that this is so, and therefore that locative PPs should not be used as evidence that these languages express motion in a way that clashes with their Talmyan type. As mentioned in section 2.3.3, Pourcel & Kopecka (Reference Kopecka, Hickmann and Robert2006: 35) and Stringer (Reference Stringer2003: 46) note that in French the location marker dans can receive a goal-marking interpretation in the right context, as shown in (50) and (51) (repeated from (31) and (32) in section 2.3.3). Not only is this ‘non-canonical’ use of dans in (51a) possible, but in the context of a mother shouting to her children it is actually more natural-sounding than the equivalent ‘canonical’ V-framed expression of the same meaning in (51b) (cf. the oddity of the English translation).

1. (50) Il court dans le jardin. he runs in the garden ‘He runs into the garden.’

   (French – Pourcel & Kopecka Reference Pourcel and Kopecka2006; 35)

1. (51)

   1. (a) Allez, courons dans la maison! go.2pl, run.1pl in the house ‘Come on, let's run in the house!’

   2. (b) ?#Allez, entrons dans la maison en courant! go.2pl, enter.1pl in the house in running ‘Come on, let's enter the house running!’

      (French – Stringer Reference Stringer2003: 46, ex. (7))

Kopecka (Reference Kopecka2009) reports on an in-depth corpus study examining the factors favoring such interpretations in V-framed French. Her study confirms that the factors relevant to English extend to French. For instance, as shown in (50) and (51), dans ‘in’ is more likely to be understood as into with locations that can be viewed as delimited – that is, as being ‘containers’ rather than simply ‘areas’. Although Kopecka limits her study to ten manner-of-motion verbs, she finds that the verbs denoting manner of motion that are most likely to produce displacement are those that are most likely to be found with locative phrases understood as goals. This dovetails with Allen et al.'s (Reference Allen, Özyürek, Kita, Brown, Furman, Ishizuka and Fujii2007) observation that manner-of-motion verbs can be subdivided into those that describe forms of motion that necessarily produce displacement to some goal, such as running, walking, flying, and perhaps jumping and rolling, and those that do not, such as dancing. Verbs of the first type involve manners characteristic of animate entities and are typically used with the intent of reaching a goal, suggesting that they may indeed at least implicate a notion of path, though they do not lexicalize direction or result, despite claims that they do by Alonge (Reference Alonge, Agostiniani, Bonucci, Giannecchini, Lorenzi and Reali1997) and Folli & Ramchand (Reference Folli, Ramchand, Verkuyl, de Swart and Hout2005) for Italian and Fábregas (Reference Fábregas2007) for Spanish; see also Mateu (Reference Mateu and Asbury2008: 246, n. 29). None of these studies, unfortunately, considers the durative/punctual criterion, although Baicchi (Reference Baicchi and Papi2005: 514) notes that ‘immediacy’ or ‘suddenness’ of the overall event is a property of many comparable Italian examples.

These observations are confirmed in corpus studies of Spanish by Martínez Vázquez (Reference Martínez Vázquez2001) and Fábregas (Reference Fábregas2007). Fábregas notes that a is found with a directional interpretation precisely with those manner verbs which imply displacement, despite the purported V-framed status of Spanish. Similar results emerge from Martínez Vázquez's study, which systematically explores the range of semantic subclasses of motion verbs in Spanish; she reports on all types of phrases understood as goals, including those expressed with until-markers, but importantly cites a fair number of examples with a, typically found with those manner-of-motion verbs that imply displacement (see Fábregas Reference Fábregas2007: 168–169, ex. (3) for further examples):

1. (52)

   1. (a) … deslizándose a las habitaciones de las bailarinas … slipping to the rooms of the dancers ‘slipping into the dancers’ rooms'

   2. (b) … volaron a Mar de Plata … flew to Mar de Plata ‘they flew to Mar de Plata’

      (Spanish – Martínez Vázquez Reference Martínez Vázquez2001: 51–52, exx. (101), (112))

Stringer (Reference Stringer2003, Reference Stringer and Saint-Dizier2006) notes similar data in colloquial Japanese, also involving manner verbs that implicate displacement, although he notes that the judgments are variable (see Beavers Reference Beavers2008a: 305–309 for further discussion):

1. (53)

   1. (a) Akira-wa umi-no-naka-ni hashitta. Akira-top sea-gen-inside-to ran ‘Akira ran into the sea.’

   2. (b) Hidari-ni tobu. left-to leaps ‘(He) leaps to the left.’

      (Stringer Reference Stringer2003: 46–53, exx. (5), (35c))

Independent of a language's Talmyan type, a pragmatic inference of directed motion can arise when context facilitates it. Although a full account of how such interpretations arise is still necessary, a theory of motion event encoding must be flexible enough to allow for them. Our approach more easily accommodates the use of inference to attribute goal interpretations to locative expressions than an approach that explicitly posits a two- or three-way typology, since we make no predictions about the presence or absence of such interpretations beyond the fact that putatively V-framed languages tend to lack lexicalized path encoding satellites. However, nothing prevents interpretive processes that allow locative expressions to take on such meanings in a given context, creating apparent S-framed behavior.

6. Conclusion: revisiting Talmy's typology

We have argued that the range of attested crosslinguistic diversity in motion event encoding points to a much richer typology of languages than typically assumed. However, the options exploited in a given language are constrained by the more general manner, result, boundary, and location encoding resources available to it, and the resources available for putting them together. Thus, the crosslinguistic diversity in motion event encoding can effectively be reduced to a more basic form of typological diversity. In this concluding section we return to Talmyan typologies, and suggest that they may be a by-product of the interaction of more basic typological parameters with factors affecting how the relevant resources are used.

Although a particular language may have multiple options available for encoding manner and path, some may be preferred on independent grounds, for example due to morphosyntactic complexity or to preferences for certain types of lexemes over others within the lexical inventory of a language. We begin by considering morphosyntactic complexity. The use of encoding options that are less complex – and, thus, presumably easier to process – is preferred to the use of more complex ones. As a consequence, a language might appear to have a more limited set of encoding options available than it actually has. Consider (54), which includes several acceptable descriptions in Japanese of an event of John running to the station in which both manner and path are encoded:

1. (54)

   1. (a) John-wa eki-ni itta. John-top station-to went ‘John went to the station.’

   2. (b) John-wa eki-ni hashitte-itta. John-top station-to running-went ‘John went running to the station.’

   3. (c) John-wa eki-made hashitta. John-top station-until ran ‘John ran to the station.’

   4. (d) John-wa hashitte eki-ni itta. John-top running station-to went ‘John went to the station running.’

      (Japanese – Yoneyama Reference Yoneyama1986: 2, ex. (4))

Presumably, (54a) is the least morphosyntactically complex event description, involving a single path verb that entails or selects for the goal PP, while (54b) involves a V-te-V complex predicate of the type discussed in section 3.2, and (54c) makes use of an adjunct PP headed by an until-marker, and thus involves the boundary/goal inference discussed in section 2.3.1, adding semantic complexity. The option in (54d), despite showing the -te participial form also found in (54c), involves a subordinate participial clause; thus, it is truly biclausal (Matsumoto Reference Matsumoto1996: chapter 9), as reflected in the lack of adjacency of the two verbs, and represents a grammatically more complex option. Of these options, (54a) is most preferable on complexity grounds, while (54d) is least preferable. The other two options are both suboptimal, but better than (54d). This suggests that Japanese favors V-framed encoding, with tendencies towards E-framed encoding. Romance languages have a similar spectrum of options with one exception: there are no equipollent forms. Given the more limited options, French appears to conform more closely to the V-framed ‘ideal’.

Similarly, the putative S-framed tendencies noted for English may arise because the canonical S-framed pattern, manner verb plus path satellite, is presumably the least marked of the available options; these were illustrated in (48) and are repeated here. The example in (55b) avoids the manner adverbials required if manner is expressed via a satellite, as in the other two sentences.

1. (55)

   1. (a) John moved stealthily out of the bedroom.

      (manner=adverb, path=adposition)

   2. (b) John stole out of the bedroom.

      (manner=V, path=adposition)

   3. (c) John left/exited the bedroom stealthily.

      (path=V, manner=adverb)

Thus, the relative complexity of available options may favor some over others, in turn yielding strong tendencies within a language that may masquerade as categorical constraints.

The issue of complexity arises in a second form with respect to the optionality of expression of some components of motion events. Optionality is inherent in Talmy's verb/satellite contrast: satellites of all types – whether path or manner encoding – are generally optional, unlike the main verb. If satellites are optional, then the question arises of when they need to be expressed. This question has been addressed by a comparative study of manner encoding in English, an S-framed language, and Greek, a V-framed language. Papafragou et al. (Reference Papafragou, Massey and Gleitman2004) show that the frequency of manner encoding is dependent not just on language type but also on whether the manner in question is ‘inferable’ or ‘opaque’. Given a scene with a man walking up the stairs (where walking is a canonical and thus easily inferable way to go up stairs), English speakers tended to use manner verbs (with or without path PPs), while Greek speakers tended to simply use path verbs, leaving the manner unexpressed, as might be expected given the two languages' Talmyan types. However, when Greek speakers were presented with scenes with unexpected manners of motion (e.g. a plane flying upside down), the frequency of manner encoding (either through complex manner adverbials or manner verbs) increased significantly. This suggests that there may be a preference for choosing encoding options that avoid the use of satellites, especially more morphosyntactically or semantically complex satellites such as until-markers or subordinate clauses which are typically used in V-framed languages if manner is in the verb. This observation is also made by Slobin (Reference Slobin, Shibatani and Thompson1996), who notes the considerable loss of manner information in English-to-Spanish novel translations. It appears that speakers avoid satellites when possible; their use depends on how necessary the meaning components they would encode are to the event description, as well as how inferable these components are from context. A dispreference for satellites may in turn move certain languages towards the more predominant use of either V- or S-framed encoding options, even if other options are available.

Preferences for some encoding options in a given language may also arise due to the shape of its verb lexicon, as verbs are the linchpin both in previous typologies and in our approach. Although nearly every language has both path and manner verbs, languages differ significantly as to how many verbs of each type they have. A language may prefer certain types of motion descriptions depending on its having a greater number of path vs. manner verbs. Most languages have basic path verbs such as English come and go, but there is more variability as to whether they have available path verbs that encode further directional or orientational information, such as approach (i.e. ‘go towards’) or enter (i.e. ‘go in’). Crosslinguistic differences in verb inventories are even more pronounced for manner verbs. Most languages have verbs describing very basic manners of motion like walk, run, fly, and swim, but fewer also provide highly contentful manner-of-motion verbs such as amble ‘walk in a leisurely manner’, jog ‘run for exercise (or) at a slow and regular pace’, waltz ‘dance to a three-beat rhythm’, and the like (Wienold Reference Wienold, Egli, Pause, Schwarze, Stechow and Wienold1995, Slobin Reference Slobin, Niemeier and Dirven2000; see also section 2.2). It seems plausible that the encoding options preferred in a certain language would be those that exploit its lexicon to the fullest.Footnote ¹⁶

There is also crosslinguistic variation in which meaning components are encoded in a motion verb. For instance, in Atsugewi (Hokan; Northern California), rather than path and manner, a motion verb may encode properties of the figure. Thus the Atsugewi verb root -lup- is used to describe the movement or location of ‘a small shiny spherical object (e.g. a round candy, an eyeball, a hailstone) …’ (Talmy Reference Talmy2000: 57–58). Atsugewi is classified by Talmy as S-framed, as path is not specified by its motion verb roots, yet its motion verbs lexicalize meanings quite different from the verbs in the more familiar Indo-European S-framed languages such as English, German, and Russian. Classifying them all as S-framed tells us that the verb root does not lexicalize path, but says little about what it actually does contain.Footnote ¹⁷

Thus, Talmy's typology results from numerous converging factors, including the overlap of path/manner (or rather result/manner) encoding in the verb, the verb's obligatoriness, and the independent availability of various means of encoding manners and paths, combined with preferences for certain non-verbal encoding possibilities over others. A similar conclusion is reached by Slobin (Reference Slobin, Shibatani and Thompson1996), who argues that languages may exhibit a range of encoding options, though only some of these options will be viewed as relatively simple or colloquial (due in part to preferences a language may exhibit for certain encoding options because of its typological status). These options will become preferred for speakers of the language as part of the development of a canonical rhetorical strategy, with other available options being dispreferred. We agree in spirit with Slobin's reasoning, but suggest that although there is crosslinguistic variation (as in the availability of until-markers, compounding, or serialization), not just any option is in principle possible in any language. Rather, variation follows from more basic motion-independent resources a language has, so that on a language-by-language basis we can still make certain clear predictions about what options a language may allow.

This conclusion has ramifications for proposals that Talmy's two-way typology arises from a ‘macroparameter’, as in Mateu & Rigau (Reference Mateu, Rigau and Alexiadou2002). Mclntyre (Reference McIntyre2004), Zubizarreta & Oh (Reference Zubizarreta and Oh2007), and especially Snyder (Reference Snyder1995a, Reference Snyder2001). We focus on Snyder's proposal, which has been perhaps the most influential, particularly in work on language acquisition (e.g. Liceras & Díaz Reference Liceras and Díaz2000, Snyder Reference Snyder2001, Slabakova Reference Slabakova2002). Snyder introduces the Compounding Parameter, which differentiates languages according to whether they allow productive noun-noun (NN) compounding; he then proposes a strong connection between the availability of NN compounding and a cluster of phenomena said to involve ‘complex predicate formation’, including the availability of directional complements to manner-of-motion verbs and resultative phrases. In this way, Snyder connects the Compounding Parameter to Talmy's typology. If Talmy's typology is indeed epiphenomenal, then accounts based on such parameters are called into question. In fact, Snyder's Compounding Parameter has been criticized: Guevara & Scalise (Reference Guevara, Scalise, Scalise, Magni and Bisetto2009: 123) question it on morphological grounds, while Son (Reference Son2007) presents data showing that the availability of NN compounding can be dissociated from the availability of resultatives (see also Mateu Reference Mateu and Asbury2008: 245, n. 26).

In sum, we propose that the wide variation in motion event encoding falls out from general constraints on how manner and path may be encoded in language, together with independent properties of the morpholexical inventories and morphosyntactic resources of particular languages. Although we suggest that Talmy's typology is an epiphenomenon, it emerges because the lexical category verb may encode either manner or path, but not both simultaneously, forcing a language to choose to encode one meaning component in the verb and one outside it: that is, to choose a V- or an S-framed option. The natural question our study poses is whether comparable explanations will prove to be applicable to other apparent typological differences between languages (cf. Hale & Keyser Reference Hale, Keyser, Mendikoetxea and Uribe-Etxebarria1997, Reference Hale and Keyser1998; Koontz-Garboden Reference Koontz-Garboden, Booij and Marle2006 on state-derived inchoatives/causatives; Harley Reference Harley1995, Reference Harley1997; Hoekstra Reference Hoekstra, Haider, Olsen and Vikner1995; Siewierska Reference Siewierska1998; Snyder Reference Snyder1995a, b; Levin Reference Levin2008 on the dative alternation; Folli & Ramchand Reference Folli, Ramchand, Verkuyl, de Swart and Hout2005 on resultatives/goal expressions; Beavers Reference Beavers2006, Reference Beavers, Friedman and Ito2009b on argument/oblique alternations and argument realization patterns). We believe that they will, and we hope that this work will provide further impetus to the necessary investigations.