1.1 Telicity

Natural languages are known to be sensitive to semantic differences between atomic (“count”) versus nonatomic (“mass”) reference in the nominal domain (Link 1983), and, analogously, between bounded (“telic”) versus unbounded (“atelic”) events in the verbal domain (Bach 1986), in ways which are both language dependent and also reflect general properties of the world we inhabit. Consider the nominal case first: for a felled tree on the back of a truck, a speaker of English can talk about the tree itself (atomic reference: tree is a count noun), or the wood that makes up the tree (nonatomic reference: wood is a mass noun). This choice will also affect the grammaticality of other morpho-syntactic elements. For example, English has several structures that are restricted to count reference (1) (including the plural suffix and count quantifiers like many), and others for mass reference (2) (including lack of plural morphology and mass quantifiers like much).

Since Bach (1986), the nominal domain has been analogized to the verbal domain through the notion of telicity, or boundedness of an event. Consider an hour-long event in which Mary carried the wood to a lumber yard in the back of her truck. This can be conceptualized either as an action of transportation with a clear goal (the lumber yard), and expressed with a clear boundary point in the telic predicate drive the wood to the lumber yard, or as an unbounded action of wood-carrying, expressed without a clear boundary point in the atelic predicate carry the wood in her truck. Again parallel with the nominal domain, elements of linguistic composition are sensitive to properties of telicity: for example, in English, verb phrase modifiers like in an hour can co-occur naturally with telic but not atelic predicates to describe how long the event lasted (3), while others like for an hour modify atelic predicates naturally but not telic predicates to the same semantic effect (4).

Across both the nominal domain and the verbal domain, several nonlinguistic and linguistic factors conspire to result in either mass vs. count status, or telic vs. atelic status. For example, while languages tend to categorize some nouns similarly (e.g. most languages classify water as mass and cat as count), other nouns vary from language to language (e.g. hair is mass in English and count in Italian). In addition, languages vary in what kind of structures are sensitive to these categories: in English plural marking and quantifiers distinguish mass/count, while in Mandarin different nominal classifiers are used for many of these same distinctions, and some languages seem to make little or no distinction between them at all (Lima 2014, but see Deal 2017; Scontras et al. 2017).

In the verbal domain, telicity is similarly based on several complex factors, including verbal lexical semantics as well as the argument structure of the sentence as a whole. For example, directed motion events in which the destination of an action is specified in the predicate generally lead to telic structures but the specific nature of the prepositional semantics matters: in English, Mary ran to the store is telic while Mary ran towards the store is atelic. In other languages, partitive case and conative particles preserve the general directionality of an action while removing the specificity of the ending, as in Finnish (I shot karhu-a [bear-partitive] means I shot in the direction of the bear, but I shot karhu-n [bear-accusative] means I shot the bear dead; Kiparsky 1998) or Dutch (She wrote aan haar proeschrift [conative-marker her thesis] means she worked on her dissertation but without the conative preposition aan, she finished it; Van Hout 1996). Identifying the telicity of a predicate, therefore, depends on a thorough understanding of many semantic details of a specific language.

While several morphological markers track with mass/count and telic/atelic distinctions, another parallelism between these two domains is that typically neither property is one that is marked directly with dedicated morphology. For example, English is not analyzed as having a morphological marker specifically for count noun status, despite -s generally being used to mark plural on count but not mass nouns, because singular count nouns (cat) lack the morpheme just as mass nouns do and some mass nouns even have count interpretations without allowing plural morphology, like furniture (Barner & Bale 2011). While there is a correlation between -s and count noun status, the relationship is indirect. For the telic/atelic distinction, there is similarly a trend for telic predicates to appear with perfective marking (Mary read the book) and for atelic predicates to appear with imperfective marking (Mary was playing) in both adult language and child language (Shirai & Andersen 1995). However, morphological aspect marking is clearly independent of the telicity of the predicate and all combinations of aspect and telicity are acceptable (5).

In fact, the differing interpretations of telic and atelic predicates with perfective and imperfective morphology constitute a classic linguistic test of telicity for English, the so-called Imperfective Paradox (Vendler 1967; Dowty 1979). This test notes that for atelic predicates, the imperfective version (5c) entails the perfective version (5d): the unbounded semantics of the atelic predicate are interpreted very similarly regardless of whether one takes an external (perfective) viewpoint or an internal (imperfective) viewpoint. By contrast, for telic predicates, shifting from an external to an internal viewpoint changes the entailments of the predicate itself. It is possible to continue (5a) with an explicit statement that the event’s culmination was not reached (but she left it at the airport and never found out how it ended) but such an explicit denial of the ending leads to a contradiction after perfective (5b). Thus, differential behavior of telic and atelic predicates under aspect morphology is one way that the predicate types can be identified.

A few cases of dedicated telic morphology have been claimed to occur in spoken languages. Most notably, Travis (2005a) explores the properties of three Malagasy (iso 639-3: mlg) verbal prefixes, tafa-, voa-, and (m)aha-,¹ which she argues realize telic meaning, based on comparisons like (6) and (7) (Travis 2005a: 395–6; drawn originally from Phillips 2000: 87).

Travis’s claim of morphologically marked telicity rests on the observation that in accomplishment verb constructions without these prefixes, the existence of a natural endpoint that is reached is implicated but defeasible (6). However, when one of the target prefixes is added, it is entailed that the endpoint has been reached (7). From examples of this sort she concludes that the relevant prefixes are telic morphemes: they change an atelic predicate into a telic one. However, the data leave room for interpretation. The structure of Travis’s examples is the same as the Imperfective Paradox (see example 5) and hinges on the fact that an explicit statement that the event’s culmination was not reached is allowed in (6) but is a contradiction in (7). The Imperfective Paradox, though, relies on the interaction of both telicity and perfectivity. The contradiction that arises in (7b) shows that the predicate in (7a) is both telic and perfective, but the acceptability of (6b) only requires one of those two values to be different. As presented, the defeasibility test in Malagasy could be evidence of a telic vs. atelic distinction (if the verb is consistently perfective), but also of an imperfective vs. perfective distinction (if the predicate is consistently telic). From the data that Travis provides, it is just as possible that (m)aha- and the other prefixes are perfective markers and their absence signals neutral, or even imperfective, marking for the sentence. This example highlights and foreshadows one problem that we also encounter in our investigation in this paper of ASL – while aspect and telicity are conceptually independent, in practice it can be difficult to separate telicity from aspect. The interpretative effect of a telic predicate in imperfective aspect shares many similarities to the interpretation of an atelic predicate. Nonetheless, even if we accept Travis’s argument, Malagasy is notable for its rarity. Together perhaps with the related language Tagalog (iso 639-3: tgl),² it is one of only a very few examples of (possibly) dedicated telic morphology that we are aware of as occurring in spoken languages.³ This is striking, considering the pervasiveness of boundedness in event conceptualization.

One final analogy between the nominal and verbal domains that we will see is relevant to our investigation in ASL is the possibility of semantic coercion: in the nominal domain, the presence of some morphological elements can coerce the opposite interpretation of the standard one; similarly, use of some modifiers coerces a telic interpretation from an otherwise atelic predicate, and vice versa. For example, (8) and (9) illustrate the typical uses (a) and atypical uses (b) of mass/count morphology. Turning to the verbal domain, we see the same kind of pattern: one continuation is natural while the other one is an unusual interpretation, forced by the specific argument choices in the predicates in (10) and (11).

The possibility of coercing both possible interpretations foreshadows another one of the problems we run into when diagnosing telicity properties of predicates in a new language: the boundary between forced and natural is not always such a stark one, and it is often possible to reconceptualize events in different ways depending on surrounding linguistic context. In this paper, we argue that, for several reasons, the relationship between verbal forms and telicity values in sign languages is much less straightforward and much less exceptional than has been suggested so far in several papers arguing for a unique status of telicity in sign languages. In the next section, we turn to a discussion about telicity in sign languages in more detail.

1.2 Telicity in sign languages

Given that the telic/atelic distinction is rarely if at all expressed via dedicated morphology in spoken languages, it is a striking exception, then, to encounter the claim that all sign languages encode abstract event structure universally, morphologically, and iconically – this is known as the Event Visibility Hypothesis (EVH) (Wilbur 2003; 2008; but cf. the weaker formulation for ASL in Malaia et al. 2013). The observation that underlies the EVH is that many signs that describe events with distinctive boundary points (e.g. DIE, STEAL) involve a bounded movement (which Wilbur defines as a change, across two prosodic time slots, in one of the following: handshape aperture, orientation, “setting change”, or “change of location with contact”). By contrast, many signs that describe unbounded events (e.g. PLAY, WORK) keep the hands constant along these dimensions through both timing slots (Wilbur 2008: 232). The intuition behind the semantics of the EVH is that these distinctive prosodic changes iconically represent the semantic boundary of the event and the lack of such changes implies a lack of boundary in the semantics. Figure 1 presents video stills (from ASL-LEX project, Caselli et al. 2017) of the contrast: PLAY describes unbounded events and the movement continues without a clear boundary, while STEAL describes bounded events and the movement ends abruptly.

Figure 1

Video stills of ASL verbs PLAY and STEAL.

Wilbur (2003; 2008) categorizes this change across the prosodic timing slots as an EndState marker and proposes that it has morphological status: its presence reflects the presence of a boundary point in the conceptualization of the event as it is expressed by a predicate, an overt manifestation of Ramchand’s (2008) res head marking the boundary (result) point. In other words, the EndState morpheme signals that the predicate is telic.

The Event Visibility Hypothesis is, in this formulation, stronger than the claim that there is a regular form-meaning link between the shift across prosodic timing slots and the conceptual structure. Wilbur (2008) explicitly argues that the EndState marker is an “affix morpheme” that specifically signals the inherent lexical/phrasal property of telicity (Wilbur 2008: 232, example 14), as above, and is not any kind of marker of perfectivity. As such, she distinguishes it from, for example, Rathmann’s (2005) proposal for the HOLD marker in ASL, which he suggests indicates an event was interrupted, or the sign FINISH analyzed as a marker of perfectivity (cf. Fischer & Gough 1999). Under Wilbur’s account of EndState as a telicity morpheme, it manifests as several allomorphs (since many different kinds of changes across the prosodic timing slots count as the EndState marker) and it is further subject to morpho/phonological constraints – most notably, it does not co-occur with spatial verbs. Crucially, she notes that few (possibly no) spoken languages have an explicit morphological marker of telicity (iconic or not) and that this is one point of difference between sign languages (as a class) and spoken languages, and further that this “…may provide the fundamental similarity that makes sign languages look more similar to each other…” (Wilbur 2008: 246).

Since the status of this EndState marker as an iconic morpheme puts sign languages in a unique typological class, there have been several extensions of this work to investigate the boundaries of both the morphemic and iconic components. Some of these extensions focus on the iconicity and its transparency, namely whether some regularity related to the EndState marker is available to both signers and nonsigners alike when perceiving sign language structure. For example, Malaia & Wilbur (2012) argue that there are physical characteristics of EndState that can be quantified – rapid deceleration, for instance – which seem to transparently encode the proposed boundary semantics of the morpheme through a bounded physical motion.

In a further investigation of this iconic component through a behavioral test of the transparency of EndState’s meaning, when adults who have had no exposure to a sign language were asked to guess a given sign’s meaning (in Italian (LIS), Dutch (NGT), and Turkish (TİD) sign languages), they guessed at above chance levels signs that matched the given sign’s telicity value (Strickland et al. 2015). In this study, nonsigners continued to pick meanings that matched based on telicity value even when the signs themselves were “nonce signs” that followed the phonological patterns of the sign languages (including presence/absence of an EndState allomorph) but did not actually have any meaning. Strickland et al. interpret their finding to point toward a possible encoding of telicity in “core cognition”, the kind of building blocks for cognition available for use in the languages of all humans (as well as pre-linguistic children and other animals). Under this view of telicity, it is thus unsurprising that unrelated sign languages throughout the world would make use of the same kind of telicity marking, and rather what is then quite surprising is that spoken languages do not do the same.

Kuhn (2017) focuses on another aspect of the EVH: its discrete nature. He argues that sign languages use an iconic mapping between their form and the expressed event structure that goes beyond the simple presence or absence of the EndState marker (in form) and boundary point (in meaning) to something that measures out the proportion of an event that occurs and matches it to the proportion of the movement performed in a sign. For example, the sign DIE can be produced without the realization of the EndState marker, and the interpretation is that the subject came close to dying, but did not yet die. Wilbur (2008) notes this as well, but Kuhn argues that the scopal properties of the structure proposed in the EVH do not predict readings that are available in ASL (12) and moreover, that the EVH does not capture further mappings between form and interpretation as seen in changes in acceleration and deceleration which ASL signers interpret as reflecting the speed and frequency of the events that they describe (13).

Kuhn proposes that the data can be better captured by an iconicity function which is present in the lexical entry for iconic predicates in ASL. For example, his semantic denotation for the verb CLOSE is shown in (14) (his example (29)), and we extrapolate that his denotation for DIE is as in (15).

This function Icon_φ requires that the distance that the phonetic form has traveled at a given time corresponds to the degree that a measure (of closure, of death, etc.) has changed in the event at the corresponding time, and moreover, that if the maximum of the distance is reached (Wilbur’s EndState), the corresponding maximum/endpoint of the degree of the event has been reached (Kuhn 2017: 19–20). This has an advantage in accounting better than the original EVH for meaningful differences expressed by both the intermediate forms of verbs (12) and variation in speed (13). It does, however, leave open the question of which verbs have Icon_φ in their semantic denotation, leaving this as simply a constraint present in (some number of) sign language verbs. Under this view, why some verbs, and why all and only sign languages, make use of such a function remains a puzzle, and it emphatically does not predict the existence of discrete morphological categories reflecting event structure.

Finally, Wright (2014) proposes yet a third linguistic mechanism by which sign languages are proposed to iconically encode telicity: through the repetition of movements within a sign that correspond to repeated atomic stages of processes. For example, events of drinking involve bringing a container to the mouth and events of eating involve bringing food to the mouth. In these cases, there is a minimal atomic unit that is a necessary condition on the event: the affected object needs to be brought to the mouth. The signs for these events iconically represent this minimal unit of the event; moreover, just as in the real world the events may include multiple instances of the minimal units (multiple sips from a cup, for example) so too can the sign be repeated. Wright argues that it is in fact the presence of these iconic signs for the atomic units that allows for persistent process – that is, atelic – interpretations on these predicates.

All of the above works take for granted that telicity is what is iconically reflected in the forms of verbs in sign languages, either due to an iconic boundary marker, an iconic function that portions out movements in the sign to correspond to progress of an event, or iconic repetition of stages in a process. We argue that while the observation of a relationship between event structure and form in sign languages seems to be supported in many ways, there is reason to doubt that telicity is always what is being reflected. In section 2 we argue that strong claims of “event visibility” have included limited tests of telicity as separate from, for example, grammatical aspect, and in Section 3 argue that these same claims have been focused on a limited set of verbs at the extremes in both categories, minimizing generalizability across the lexicon as a whole. We follow these observations with new data bearing on each point and then propose that these shortcomings have led to exaggerated differences between sign and spoken languages in both the morphological and iconic aspects of the Event Visibility Hypothesis. By providing new tests and data bearing on several points we add more nuance to the topic. We end by arguing that several factors have conspired to make event structure appear iconic in ASL, but when they are teased apart there are fewer surprising differences based on language modality than have been claimed so far.

2.1 Tests of telicity that depend on other languages

We begin by noting that in the current literature on telicity in sign languages, there is frequently not an independent (e.g. not English based, and not assumed from Endstate form) test of telicity for predicates categorized as such. Some investigations of the Event Visibility Hypothesis have not offered independent tests of the telicity of predicates at all. For example, Wilbur (2008: 221) discusses several tests proposed by Rothstein (2008) for states vs. dynamic events based on English, such as the in an hour/for an hour test that we discussed above, and implies that this should inform categorization in ASL but generally does not apply these tests within ASL or discuss which tests make sense in ASL and which do not. The paper provides a short list of verbs in ASL in each category based on Rathmann (2004), but they are not those used in discussion in the rest of that paper.

Strickland et al. (2015) do not discuss how they identified the signs they used as being telic or atelic, but rather seem to determine category membership in a circular manner: if a sign has the EndState marker, then it is “telic”, and if a sign does not, they say it is “atelic”. They do show that non-signers can successfully infer two categories and link them to English words which align with English tests of telicity, but whether the signs themselves are best (or solely) distinguished along the dimension of telicity is not determined in any independent way.

Others have employed telicity tests translated from English, but since these are language specific, it must be established that they correctly diagnose telicity in individual sign languages. Kuhn (2017) uses translated interpretations of English tests, while in fact showing that they do not function the same as in English. For example, Kuhn uses (16) (his example (39)) to demonstrate a contrast between one production of the verb SLEEP which he argues lacks the iconic function Icon_φ (16a) and another that expresses it (16b).

Interestingly, in ASL, ONE-HOUR may either mean ‘for one hour’ or ‘in one hour’, whichever is more consistent with the interpretation of the verb. Rathmann (2005) suggests that there is a form of ONE-HOUR that unambiguously specifies duration, but that regardless, overlap between forms means that the “for/in an hour” diagnostic does not provide a good test of telicity in ASL. Thus, the use of ONE-HOUR on its own does not work as a test of telicity in the same way for an hour and in an hour do in English.

Malaia & Wilbur (2012) investigated 50 ASL verb forms with four different tests stemming from English: an adverbial modification test (“for/in an hour” as in English), a conjunction countability test (“she did V(erb) on Sunday and on Monday”, counted as one event (atelic) or two (telic)), ‘almost’ modification (licensed for telic but not atelic predicates), and a stop/finish combinability test (‘stop’ combining with atelic and ‘finish’ combining with telic predicates). They found that 40 verb forms patterned the same with all four tests (categorizing 24 of their verbs as telic and 16 as atelic), and use the categorization to show that the kinematic signatures of the atelic and telic classes were different as measured using motion capture technology. This represents an important effort both to provide tests of telicity that are independent of verb form, and to expand lexical coverage. However, more work is needed to determine whether the translated tests diagnose telicity in ASL, as they do in English; Malaia & Wilbur do not discuss their application of these tests in detail.

A much stronger form of evidence for various forms of the EVH would be to develop a test of telicity that is independent of the form of the verb and is either specific to ASL or is language independent, so that it is possible to ask whether this test does or does not necessarily track the EndState morpheme. Here we will first discuss some ASL-specific tests (Section 2.2), and then discuss a test that is language independent (Section 2.3), and is consistent with Wilbur’s (2008) distinction of telicity/atelicity as one of hetero- vs. homogeneity in semantics.

2.2 ASL tests of telicity

Rather than assuming that ASL signs have the same semantic properties as their typical English glosses, or that the tests work the same way, it would be preferable to develop ASL-specific tests (and so on for other (sign) languages). In his dissertation, Rathmann (2005: Chapter 3) proposes precisely this, after noting the difficulty of transferring linguistic tests of telicity to ASL. He analyzes two constructions (NEED X TIME vs. STILL Y) which, parallel to the English “in an hour/for an hour” tests, can each be used naturally with only one telicity value.

The NEED X TIME construction ((17), Rathmann’s example (35)) specifies the time needed to complete an event and it is felicitous with telic predicates which have a designated boundary point (17a) but not with atelic predicates that contain no designated ending (17b). Rathmann notes that this construction is a close equivalent to English “take X time”, which is also restricted to telic predicates (18). Thus, we can consider the ability to felicitously appear as a complement of NEED X TIME as a test of telicity: telic predicates can do it but atelic predicates cannot.

The STILL Y construction indicates that an event is ongoing and Rathmann argues ((19), his example (36)) that only atelic predicates (19a) and not telic predicates (19b) can felicitously be complements of STILL Y. The equivalent verb in English selects for an imperfective form, regardless of its telicity value (20).

As Rathmann himself notes, these tests are not bulletproof in that, like other linguistic tests for telicity/atelicity (and as we saw above, similar to linguistic tests for mass/count), they are susceptible to coercion effects. For example, he notes that (17b) can be well-formed, but if so implies that events of walking always take 45 minutes, which is counteracted by our real world knowledge.

A key feature of Rathmann’s approach is that it appreciates the fact that telicity is not property of verbs in isolation, but depends instead on properties of the entire predicate, including all the verb’s associated arguments. As far as we know, Rathmann’s telicity tests provide the most extensive examination of the semantics of telicity in ASL independent of verb form, and form a small part of a much richer investigation into event structure in sign languages (Rathmann 2005). His work also highlights the importance of not using apparent translation equivalents across languages, as even very similar words (like STILL/still) may work differently (but see section 5 below). More generally, given the vigor with which the EVH and related notions have been investigated recently, we want to add further, ideally converging, evidence for telicity values in ASL, one which does not draw from English-specific tests and works on a wide spectrum of predicates.

2.3 Language-independent conceptual test

We suggest focusing on a conceptual distinction between bounded and unbounded events – the homogeneity distinction – and that may allow us to distinguish between telic and atelic predicates across languages. To start, we return to our original analogy, comparing the telicity distinction in the verbal domain with the mass/count distinction in the nominal domain. Consider wood and tree. If we take some amount of wood in a truck, and then look at some arbitrary subpart of that wood, can we still describe it as wood? In this case, we can. Mass nouns like wood have the property of being divisive, namely, they can be increasingly divided and still the overall term applies. This contrasts with tree: if we have something which we can describe as a tree, and we take some arbitrary subpart of that tree, is it still possible to call it a tree? Probably not – count nouns are not divisive/homogeneous (Link 1983; Chierchia 1998).

The same notions hold in the verbal domain (Bach 1986): if we take an arbitrary subpart of the event described with the telic predicate in (21) drove the wood to the lumber yard, the same description would not accurately describe any arbitrary subpart of the event – for any subpart of the event that does not include the endpoint, the event requires a different description. This contrasts with the atelic predicate in (22) – any subpart of the event described by carried the wood in her truck can also be accurately described by that same predicate. In other words, the structure of the described events for atelic predicates, but not telic predicates, is homogeneous.

So, let us consider how to test homogeneity first with the English verb write in different predicate forms (wrote, wrote an essay). Imagine a world in which Mary is very busy and ambitious, but she has one hour free each day to work on her writing. She joins a writing group to help her use this time wisely. This week she is working on an essay, which she started on Monday and finished on Friday (23). If we ask what Mary did during this whole week in the writing group, you can truthfully say, She wrote. If you ask what she did on Tuesday in the writing group, it is also truthful to say, She wrote. In the case of the simple predicate wrote (perfective form, no direct object), a subpart of the event still counts as the same kind of event, so we can say that wrote passes the homogeneity test, and thus by this diagnostic, is atelic. By contrast, in the same context, if you ask what Mary did in her writing group that week, one can say, She wrote her essay, but it is not truthful to say during her Tuesday writing time that, She wrote her essay. With the addition of the direct object, in the same aspectual form, wrote an essay becomes telic, specifying its endpoint. Any arbitrary sub-part of the event that does not include the endpoint of the event, therefore, cannot be described with that predicate. In other words, whether a predicate that describes the whole can also describe the subparts is a diagnostic of whether the event is conceptualized as having a boundary point (telic) or not (atelic).

This homogeneity test itself is conceptual in so much as it depends on our understanding of the structure of the event being described and not merely the grammaticality of an added linguistic element/phrase. However, what makes a particular predicate and context evoke one kind of event structure over another is very much dependent on linguistic properties. For example, the contrast between wrote and wrote an essay relies on two language-specific dimensions of English to evoke the telic/atelic distinction: the addition of a direct object and the use of perfective aspect.

Regarding direct objects in English, the intransitive verb write is atelic but adding a direct object her essay creates a telic predicate. The presence/absence of a direct object is highly correlated with whether an event is telic/atelic, but specific linguistic properties of both the verb and the object also play an important role. For example, in English, a direct object that is a mass noun rarely makes a predicate telic (Mary wrote stuff is still atelic) and not all verbs are as flexible with their argument structure as write. The verb construct for example requires a direct object and there is no intransitive construct in English that is atelic. Part of knowing English (or any specific language) is knowing which verbs permit/require/refuse direct objects and what the grammatical signals are (if any) of mass and count nouns. But learning any language requires learning the nuances of specific lexical items. One of the critical reasons to use a conceptual/linguistic test like homogeneity is that it allows these differences to emerge across languages and does not require assumptions about them to start.

The second language-specific element invoked in the homogeneity test in English is that the tested examples were all in perfective aspect (i.e. the simple past tense). As was noted for the sentences in (5), telic predicates in perfective aspect entail their completion point. This makes them non-homogeneous: telic predicates in perfective aspect will “fail” our homogeneity test because arbitrary subparts of the event do not include the ending point which the perfective aspect requires be part of the predicate’s interpretation. However, if a telic predicate appears with imperfective aspect (Mary was writing an essay) then the entailment of completion is removed and it is possible to describe an arbitrary subpart with the same predicate. The interpretive effective of the imperfective aspect collapses the homogeneity distinction between telic and atelic predicates so that if one is interested in the event structure properties of the predicate itself, then at least in English it is critical not to test the predicate using imperfective aspect. We note that the imperfective aspect does not eliminate all differences between the two kinds of predicates – telic predicates continue to identify a boundary point even if the aspectual information does not require it to be considered in evaluating truth conditions; aspect and telicity are different kinds of information. In the next section we apply the homogeneity test to various predicates in ASL, and in doing so attempt to use aspectually neutral forms, staying away from any known markers of imperfectivity. However, we will return to precisely this issue later in the paper in section 5, where we will ultimately argue that what initially looks very much like a “telicity” distinction based on form in some predicates in ASL is not yet possible to separate from markers of aspect.

2.4 Homogeneity test applied to predicates in ASL

In this section, we discuss applying the homogeneity test in ASL as another independent test of telicity. We note that the conjunction test in Malaia & Wilbur (2012) is similar in spirit to our test since it too centers on how (or whether) sub-parts of an event can be combined into a single event. However, it is not clear the extent to which it relied on changes to the target verbs, the associated arguments, or the contextual scenarios provided. It seems likely, however, that in many cases it too was a test of homogeneity.

For the data we present here, we met with three Deaf signers of ASL, one a multi-generational native Deaf signer and two Deaf signers from hearing families who learned ASL at very young ages. The signers did not know each other prior to the testing session and interacted with each other for only a short time prior to participation. Thus, we are confident each gave independent judgments that were not influenced by long-term accommodation to the other participants. In each interview, we investigated in ASL whether the given predicates are interpreted as homogeneous, defined in the same way as we did for English in the previous section: that is, can the same description be used to describe an entire event as well as an arbitrary subpart of the event?

To evaluate homogeneity in our meetings with consultants we used two types of probes. In the first method of probing for homogeneity, we described a single character’s behavior and asked whether a description containing the verb form under consideration was true for a subset of the action. We illustrate this with two predicates that seem to fall into the extreme categories that have been the focus of most previous work on the Event Visibility Hypothesis, PLAY and STEAL. PLAY has no EndState marker in its form (e.g. no abrupt deceleration) and has been categorized as atelic (24), while STEAL has an EndState marker (an abrupt end to the single internal movement) and has been categorized as telic (25); neither can appear in a form with the opposite EndState value. In the case of PLAY, we considered a context presented as in sentence (24). We asked whether if we take any subpart of this playing-in-room event, it is possible to describe it in the same way, and all three signers reported that it was. In contrast, for STEAL we presented a capture the flag scenario and the sentence in (25) and asked whether we take any subpart of this event, whether it could be described in the same way, and all three signers reported that it could not. For these verbs, we were unable to identify any contexts for which PLAY could be used to describe only the whole event, or for which STEAL could be used to describe any subpart of the event.

In our second method of probing for homogeneity, we tested each verb in two contexts, which we label here as Whole and Subpart, as in (26) and (27). Presentation of these contexts was followed by a question. The focus here is on the form of the verb in the question, and whether it can be used to describe both the Whole and the Subpart contexts, or only one of the contexts. We asked about the characters in each context: Who performed the action? To help with disambiguation, the characters in the Whole and Subpart contexts differed in gender. Associating the two contexts with different characters allowed us to present both contexts at the same time, which was maximally efficient.

Consultant responses were thus either that both characters performed the action described by the question, or that only one did. If the consultants responded that both characters performed the action, we interpreted the verb form in the question as occurring in an atelic predicate. If the consultants responded that only one character performed the action (always the character in the Whole context, reflecting how telicity-aspect entailments work), we interpreted the verb form in the question as occurring in a telic predicate.

To illustrate, in the case of PLAY, we considered a context presented as in sentence (26), in which the amount of time for the event in the Subpart context was a subset of the time for the event in the Whole context. Both characters were reported as fulfilling the question containing the verb. (Note that as with all contexts in this paper we present an English translation, but all discussions were in ASL.)

For probing a telic value, we considered a situation where one character (Whole) completes the entire process to a completion point, while the other character (Subpart) stops short of completion. For STEAL, we considered the situation described in (27). All signers report that only the first character fulfills the question that contains the verb in (27).

Our first method for probing is the most direct (especially since it is more explicit regarding the subset being of the same event), but also requires a high degree of confidence in mutual understanding of the context and the question. The second method shares more structure in common with an experimental trial, but leaves more room for misinterpretation about the relationship between the events that the two characters participate in (and their goals, etc). Two consultant discussions proceeded naturally using the first method; the third primarily used the second method of probing. Results from the two methods converged across all lexical items that we tested.

Our examples illustrate that when applied to STEAL and PLAY, tests of homogeneity are able to diagnose the verbs’ predicates as telic and atelic, respectively, in line with previous reports of these verbs. They thus serve as a proof-of-concept illustration of the test. Notably, STEAL has the EndState form as defined by Wilbur (2008) and PLAY lacks it, so they are consistent with what is expected based on the EVH. But this form-meaning correspondence is hardly surprising: these are precisely the kind of verbs that have been the focus of most of the literature on the EVH. We note that these verbs are also particularly extreme cases: PLAY describes a process that is difficult, if not impossible, to conceive of as telic even with a different argument structure, while STEAL is a punctual event, and possibly just as difficult to conceptualize as atelic. More generally, PLAY and STEAL differ along a wide variety of semantic dimensions, making it especially challenging to be sure what the specific contribution of EndState is across these cases. Additionally, since these verbs do not have both EndState and non-EndState forms, it is not possible to determine whether EndState is a true morphological marker, or something reflecting inherent lexical semantics (Aktionsart), or something similar to a phonaestheme (e.g. English glisten, glitter, gleam, glimmer), in which there is some relationship between form and meaning, but the shared element (e.g. gl-) is not quite fully a morpheme. A more informative test of the EVH would be to find a single verb that allows for the creation of a minimal pair – with and without EndState – and for tracking the homogeneity inferences across both halves of the pair. The next section focuses on precisely these types of verbs.

2.5 Homogeneity test applied to alternating predicates in ASL

One example of such an alternating verb is WRITE, illustrated in Figure 2 (video stills from ASL-LEX project, Caselli et al. 2017). In one form of this verb, which we will call WRITE_1, the movement involves one sideways movement/single swipe across the nondominant hand. Because this movement has a clear rapid deceleration to an endpoint, Wilbur (2008) would analyze this morphologically as having the morpheme EndState. We can ask whether predicates containing WRITE_1 pass or fail the homogeneity test, following the reasoning outlined above for write in English. If WRITE_1 contains a morpheme (EndState) that expresses telicity, we should expect it to fail this test of homogeneity.

Figure 2

The verb WRITE occurs in multiple forms that differ on EndState status.

This predicate indeed fails the conceptual homogeneity test (28): while IX-Mary WRITE_1 ESSAY describes the entire week-long writing event accurately, a subpart of that event (namely, Mary’s writing on Tuesday) is not accurately described by the same string. We thus conclude that the form WRITE_1 can be used to express a telic predicate. In addition, because WRITE_1 fails homogeneity, we can also be confident that it is not in an imperfective form (see section 5 for further discussion).

Note that in ASL, the addition of the direct object (ESSAY) is optional and does not seem to affect the value for this test, differing in this way from English, as we might expect given that ASL is well known for null arguments, both subjects and objects (Lillo-Martin 1986).

Now, we consider a different form, WRITE_2. By any account this sign is morphologically related to WRITE_1 in that it has the same lexically-specified handshape for both hands, orientation, and location, and a closely related meaning. The only difference is that now the movement involves multiple smaller sideways swipes of the dominant over the nondominant hand. Although as far as we are aware the existing literature on the Event Visibility Hypothesis does not focus any attention on verbs like WRITE which seem to have two forms, nevertheless when we apply Wilbur’s (2008) description of EndState form, this sign would clearly not be categorized as having the morpheme EndState. Just as with WRITE_1, we can ask whether WRITE_2 passes the homogeneity test, as it is predicted to be atelic and thus interpreted as homogeneous. As (29) illustrates, this predicate indeed passes the test for homogeneity because it is possible to use the same predicate to describe both the entire time period (29a), and a subset of that time period (29b), consistent with the EVH.

The same results obtained from the alternative, two-character elicitation method, as shown in (30):

Since WRITE_1 and WRITE_2 differ in whether they have the EndState marker and the semantic distinction tracks this difference of form, we find a new piece of preliminary evidence to support the EVH. Alternating verbs of this sort are crucial evidence for the claim that telicity marking is morphological: verbs that have only the form with EndState (e.g. STEAL), or only the form without EndState (e.g. PLAY), offer only weak evidence of morphological status – even if the form always corresponds to the expected telicity value. As we argue further below (Section 4), true morphological structure is characterized by substitutional relations, known as paradigmatic contrast, and examples like WRITE_1 and WRITE_2 provide a demonstration of paradigmatic contrast as it relates to the status of the EndState form.

To sum up our discussion of homogeneity, we have so far applied this test to only a couple of verbs of the sort used by Wilbur (2003; 2008), Strickland et al. (2005), and Kuhn (2017) (PLAY and STEAL), as well as to a type of verb not previously discussed in the literature – WRITE, which has alternating, paired forms. In both cases we find a mapping consistent with the EVH between the presence of the EndState marker on the verb and categorization of a predicate as telic. Conversely, and just as importantly, we also find a one-to-one mapping between absence of the EndState marker on the verb and categorization of a predicate as atelic (an example of “significative absence” in the sense of Stump 1997). However, the surprising claim that ASL provides a morphological marker of telicity should be put to further tests, both with more lexical coverage and addressing the possibility of alternative interpretations of the form, specifically aspect marking. In order to really conclude that EndState is a morpheme expressing telicity, we would need to establish that the form-meaning relationship found in WRITE_1 and WRITE_2 is in fact a pattern repeated in other verbs, and that the difference in meaning is really one of telicity and not of aspect. In addition, exploration of a wider set of verbs allows us to test whether the EVH holds of most (if not all) verbs, as previously claimed, or only in certain subparts of the lexicon. Lexical coverage helps to establish the extent to which a telicity (or aspect) contrast is a pervasive and central organizing principle of the verbal system. We turn to lexical coverage in the next section (Section 3).

3.1 Need for lexical coverage

Evidence both from extreme verbs like PLAY and STEAL and from alternating verbs like WRITE does seem to suggest that a change in form tracks changes in meaning, and we have provided no evidence so far that does not support the EVH. However, there is reason to think that precisely on a topic like telicity that combines world knowledge and language-specific properties, increased lexical coverage is important. Recall our original analogy, between telicity in the verbal domain and mass/count in the nominal domain. For most nouns in English, application of plural number marking like -s is generally possible for count nouns, and inability to take plural morphology holds for mass nouns. However, not only do we not analyze the plural morpheme as the instantiation of underlying atomicity (count semantics), we also know that a few paradigmatic examples of each category like cat and milk do not exhaust the possible categories of interest within the mass/count domain; for example, there are nouns like stone which alternate between mass and count depending on linguistic markers like number morphology (how much stone vs. how many stones). These may be analogized to verbs like ASL WRITE, which can take two forms with interpretative differences. Moreover, if telicity is really a grammatical marker it should appear on enough verbs to suggest that it has reached some level of generality, even if some exceptions exist. It is important then for investigating claims of morphological “event visibility” to increase lexical coverage by investigating several more verbs and asking two questions: (1) what forms can the verb take (with and/or without EndState)? and (b) what interpretations can its predicates have (telic and/or atelic)? We will focus first on our method for testing robustness of possible forms (Section 3.2), and then discuss possible meanings for these forms (Section 3.3).

3.2 Testing for available forms in ASL

To ask what possible meanings each form of a verb might take, we first needed to generate an inventory of verbs and document which can appear in which possible forms. Can all verbs take both forms? Clearly not – PLAY and STEAL are prime examples. Can no verbs take both forms? Also clearly not – we already saw that the verb WRITE could appear both with and without EndState. Moreover, in Wilbur’s description, EndState has a wide variety of allomorphs that are seemingly phonologically conditioned. To test the range of possible forms in ASL, we created a list of 23 verbs in ASL, based on intuitions that some may vary more than others. Most of these items have not been previously analyzed for telicity in the literature: READ, WRITE, STEAL, DIE, PAINT, LEARN, DRINK, SLEEP, IMAGINE, THINK, SWIM, SKATE, DRIVE, TYPE, SKI, LECTURE, DESTROY, BUILD, PLAY, STORYTELL, STUDY, BREATHE, and DANCE.

We then asked the same three Deaf signers in the same interviews about homogeneity of PLAY, STEAL, and WRITE discussed above whether each verb was available in a form with EndState, and a form without it. One challenge with the question of form is that a verb may superficially appear not to have an EndState marker (abrupt deceleration) by virtue of marking pluractionality, which is done via repetition (Klima & Bellugi 1979; Rathmann 2005; Kuhn & Aristodemo 2017). In other words, by defining EndState as rapid deceleration, descriptions involving multiple events will look like they are lacking EndState, even if perhaps the linguistic encoding of each individual event is telic. Therefore, for these tests of forms we advised signers to consider only cases with single arguments (subject, objects, and indirect objects), at a single point in time, in order to minimize pluractionality marking.

Our results in this larger set showed robust agreement in form judgments across all three signers. Verbs were categorized as shown in Table 1, with 8 verbs judged to have only a form consistent with EndState, 10 verbs judged to have only a form that is inconsistent with EndState, and 5 alternating verbs that could appear both with and without the EndState form. Even before we consider the semantic contexts in which the verbs can appear, the fact that the verbs fall into three form classes (including several alternators) already suggests that the distributional facts for EndState are more complicated than has been recognized in previous work.

We are not yet aware of any large database of verbs in ASL that includes the relevant semantic properties, but we chose a variety of verbs that seemed to have the potential to show a variety of patterns. However, because we drew our small sample with the intention of identifying verbs of various types, it would be premature to draw any conclusions about the distribution of these three classes of verbs within the wider language. That said, it is notable that signers’ judgements on forms were generally consistent and robust: while there was some discussion about whether certain forms could exist independent of other marks of aspect (e.g. followed by a perfective marker FINISH), in general most verbs had clear judgments about what possible forms they could take. These results are thus consistent with the kinematic study of Malaia & Wilbur (2012) in that our informants could also reliably identify the presence or absence of the EndState marker. Moreover, they extend that work because the informants here were probed about the acceptability of the different forms regardless of the verb (or predicate) semantics, thus allowing for an independent assessment of the connection between the form and potential meanings.

3.3 Testing for available meanings in ASL across several verbs

After determining the range of possible forms for each verb, we then asked about each allowable form, whether it can appear in a predicate with a telic interpretation, and whether it can appear in a predicate with an atelic interpretation, with the telicity value determined by our homogeneity tests. These semantic judgments were somewhat less robust than the form judgments (primarily in confidence by signers) as might be expected of semantic judgments in an area with potential for coercion, but nevertheless distinct categories emerged after clarifying discussions, several of which began to break the one-to-one correspondence we have seen so far between EndState forms and telic interpretations. A summary of these judgments are shown in Table 2; the next sections discuss each subcategory in turn.

3.4 Summary of the judgment data

Our goal of gathering data from a range of verbs was to determine if the EVH held widely across ASL. With respect to the EndState form, the data are quite clear: signers are able to make consistent and clear judgments about which verbs do and do not require EndState, as well as which verbs can appear both with and without the form. However, in asking whether the forms were linked to the predicted telicity value, the results were far more mixed. While we found many verbs in ASL that were consistent with the EVH, and encouraging results from alternating verbs, we also found some notable counter-examples. At a minimum, these counter-examples demonstrate that EndState is not the only means within ASL to signal telicity, and likewise, that EndState does not always signal telicity. In fact, some of the examples violate the EVH wholesale. “Mismatches” between morphological form and meaning (e.g. syncretism, deponency) are not uncommon to see within a language, although if there are enough such cases, they raise doubts about whether one has the right characterization of the pattern. Moreover, they speak to the additional component of “event visibility”, namely that EndState is iconic. To the extent that the marker is supposed to be transparently available to users of the language, the examples that we found which have non-isomorphic form-meaning relations are even less expected.