Aspectual reduplication in Sign Language of the Netherlands: reconsidering phonological constraints and aspectual distinctions

2.1 General background

Previous work has shown that two types of aspect can be distinguished: lexical aspect and grammatical aspect. First, lexical aspect (also called “situation aspect” by Smith [1997]) refers to inherent properties of the verb, or, in the words of Filip (2012: 721), it is “a semantic category that concerns properties of eventualities (in the sense of Bach 1981) expressed by verbs”. These “properties” generally refer to an end or boundary that is present in the lexical structure of some classes of verbs, but not in others, known as the basic distinction between telic verbs, which have a clear endpoint or goal, and atelic verbs, which do not (Filip [2012]; and see Garey [1957] for the telic/atelic distinction). Two other fundamental concepts are change of state (i.e., whether there is a transition from one state to another) and temporal extent (i.e., whether the event is punctual or whether there is some temporal extent). Based on these three properties, verbs can be divided into different classes (for instance, states or processes) (Filip [2012], and references therein such as Bach [1986]; Comrie [1976]; Dowty [1979]; Vendler [1957]). The present study will not be concerned with lexical aspect, and solely focuses on grammatical aspect – although, of course, the two are closely related (see Binnick [2012a] for more discussion).^[1]

Second, grammatical or verbal aspect (also called “viewpoint aspect” by Smith [1997]) is not an inherent lexical property of the verb, but can be described as “a sub-system belonging to the grammar of a particular language” (Binnick 2012b: 32). Comrie (1976: 3) defined it as “different ways of viewing the internal temporal constituency of a situation”. Taking English as an example, De Swart (2012: 753) compares When Bill came to the office, Sarah left through the back door to When Bill came to the office, Sarah was leaving through the back door – the progressive verb form in the second sentence changes the interpretation of the sentence and is part of the English inflectional system: it functions as a grammatical aspect marker. It is this type of aspect, i.e., grammatical aspect, that this study focuses on – specifically, continuative, habitual, and iterative aspect.

Within the domain of grammatical aspect, a fundamental distinction is that between perfective and imperfective aspect. Perfective viewpoints focus on the situation in its entirety, looking at the situation from outside, while imperfective viewpoints focus on part of a situation and its internal structure (see, e.g., Comrie 1976; Gvozdanović 2012; Smith 1997). While this distinction is grammaticalized in some languages (e.g., Russian and Spanish), this is not the case for all languages (e.g., English) (Comrie 1976; see also Gvozdanović 2012). Still, it has been noted that the distinction between these two viewpoints by means of different verbal inflectional forms is “fairly stable across languages” (Deo 2012: 161). Many languages have a single category that expresses the imperfective. Bybee (1985), for instance, noted that the distinction between perfective and imperfective is the most common inflectional aspectual distinction made in her sample of 50 languages. However, as we will see in Section 2.2, languages may also have further formalized distinctions, such as habitual and continuous aspect, which are usually assumed to be subtypes of the imperfective (Comrie [1976: 25]; see also, e.g., Carlson [2012] on habitual aspect; Mair [2012] on continuous aspect). The distinction between habitual/continuous is the second most common one in Bybee’s (1985) sample.

In the following, I first set the scene by describing previous studies on spoken languages, focusing on the three grammatical aspect types under investigation in the present research (Section 2.2). Then, I go into the two types of realization of grammatical aspect that have been identified for sign languages: modulating the verb sign (Section 2.3.1) and free functional elements (Section 2.3.2) (see also Pfau et al. [2012] for an overview). Finally, I turn to previous studies on NGT (Section 2.4), the language under investigation here.

2.2 Continuative, habitual, and iterative aspect in spoken languages

The present study focuses on continuative, habitual, and iterative aspect. Here, I define each aspect type, and briefly discuss how they are commonly expressed in different spoken languages.

According to Comrie (1976), the imperfective aspect can be further divided into habitual and continuous aspect, as mentioned above. In turn, the continuous is further divided into progressive and nonprogressive, where the progressive has been described as “the idea that an event is progressing dynamically over a time frame opened up by an utterance” (Mair 2012: 803). This time frame has been defined by Klein (1994) as the “topic time”, i.e., “the time span to which the speaker’s claim on this occasion is confined” (Klein 1994: 4). “Nonprogressive” includes situations in progress or nonprogressive states (see Mair [2012: 808], also for a critical evaluation of this distinction).

Here, I will generally focus on the continuous. Various languages have developed a grammatical category to express these semantics, which I will refer to as continuative aspect here.^[2] An example is given in (1), which shows that continuative aspect is expressed by a prefix on the verb in Zapotec.

Apart from marking on the verb, continuative aspect has also been found to be marked by periphrastic constructions or by adverbs or particles (see Mair 2012). For instance, in Russian the lexical adverb dolgo expresses ‘for a long time’ (Smith 1997: 251).

The other type of imperfective aspect concerns habituals, which “describe a situation that is characteristic of an extended period of time” (Comrie 1976: 28) – they are imperfective, because the situation is characteristic of the whole period. In the Zapotec example (2a), the habitual aspect is marked on the verb by the bound morpheme ru – comparing (2a) to (1) illustrates that habitual and continuous meaning are contrasted inflectionally in Zapotec, i.e., by means of the bound morphemes ru- and ku-, respectively (Bybee 1985). Cross-linguistically it is common to express habituality with a verbal affix, but it may also take the form of, for instance, an auxiliary or a periphrastic construction (Carlson 2012).

Comrie (1976) notes that habituality should be distinguished from iterativity (the repetition of a situation), because repetition alone is not sufficient to use an imperfective or habitual form. He points out that, if a situation is repeated a few times, each repetition can be viewed as a single instance, i.e., a situation on its own, which can be referred to with the perfective form. Further, according to Comrie (1976), a habitual form can refer to situations without iterativity, as in (2b), where English used to is considered a habitual marker (although note that Binnick [2005] actually argues against English “used to” as a habitual marker – I will not go into this discussion here). For further discussion of the difference between habitual and iterative aspect, see Bertinetto and Lenci (2012).

Bybee (1985) shows that some languages (15 out of 50 in her sample) employ a verbal marker to give iterative meaning to the verb, i.e., iterative aspect, as illustrated for Kiway in (2c), where the affix -ti expresses iterativity. Further, iterativity can be expressed by similar means as those mentioned above for the other two aspect types, for instance, adverbials (Bertinetto and Lenci 2012).

Finally, a grammatical aspect marker that commonly occurs not only in spoken languages, but also in sign languages, is reduplication. A spoken language example is given in (3).

The next section will illustrate the importance of reduplication for sign language aspect marking.

2.3 Aspect in sign languages

This section focuses on grammatical aspect in sign languages, which is mainly expressed by verbal inflection (specifically, reduplication) and free-standing markers. These will be discussed in turn in Sections 2.3.1 and 2.3.2, respectively. Section 2.4 outlines previous studies on aspect in NGT specifically, and finally, Section 2.5 details the research questions the present study addresses.

2.4 Previous studies on aspect in NGT

The first to describe the encoding of aspectual distinctions in NGT were Hoiting and Slobin (2001), focusing on habitual and continuative aspect in this language. According to them, continuative aspect describes an action that is ongoing (e.g., “He’s going on working [at the moment]” [p. 128]), and is marked on the verb sign by “three repetitions of an elliptical modulation accompanied by pursed lips and a slight blowing gesture” (p. 127). Habitual aspect, which they define as describing an ongoing action that occurs habitually (e.g., “He always works on and on” [p. 128]), is marked on the verb sign by “a slower elliptical modulation accompanied by gaze aversion, lax lips with protruding tongue, and slowly circling head movement” (p. 127). Note that both definitions are somewhat unconventional: in the definition of continuative aspect, there is no reference point (e.g., Klein’s [1994] “topic time”; see the definition given in Section 2.2), and in the definition of habitual aspect, there is a notion of continuity, implied by the word ‘ongoing’ – traditional definitions of habituality (and the one we adopt in the present study), however, do not include such a notion (see, again, Section 2.2). Hoiting and Slobin do not provide examples of the elliptical modulation. Yet, they do identify phonological constraints.

For the reader to appreciate these constraints, it is necessary to introduce the sublexical (i.e., phonological) building blocks of signs, which have been argued to function like segments in spoken languages (sometimes called “parameters”). Simplifying somewhat, the building blocks that have been identified are the handshape (hand configuration), place of articulation, and movement of the sign (Sandler 1989; Stokoe 1960; for an overview, see Fenlon et al. [2017]). The latter comes in two types: path movement and hand-internal movement. If a sign has a path movement, the hands move from one location to another. Internal movement involves a change in handshape or in orientation. Another kind of internal movement is so-called secondary or trilled movement, which involves fast repetitions of orientation or handshape changes, or wiggling of the fingers. Internal movement and path movement may combine, or occur on their own (Brentari 1990; Sandler and Lillo-Martin 2006; Wilbur 1987). Evidence for the phonological significance of these building blocks comes from minimal pairs; for instance, two signs may differ in terms of their movement alone. Furthermore, just like segments in spoken languages, these building blocks can be described in terms of distinctive features, which are organized in feature hierarchies. Various phonological models have been put forward (e.g., Brentari 1998; Sandler 1989), to which I will return in Section 5.1.

According to Hoiting and Slobin (2001), aspectual marking in NGT is phonologically constrained, the relevant building blocks being movement and place of articulation: verbs with internal movement and/or body contact cannot undergo the elliptical modulation. These verbs thus remain uninflected, but are followed by an aspectual particle which they gloss by means of the Dutch word door (lit. ‘through’), which takes on the inflection instead. This particle is borrowed from spoken Dutch, where it can be used with a similar meaning with some verbs (e.g., Hij loopt door ‘He continues to walk’). The sign is illustrated in Figure 1. In the following, I will gloss this sign as cont (since later studies have shown that this particle expresses continuity, but not habituality, as we will see below). An example where a verb with body contact (try, which contacts the nose) is combined with cont, which then takes on the inflection, is given in (7).

Figure 1:

cont in NGT (van Boven 2018: 7), video still taken from the Corpus NGT (Crasborn et al. 2008; Crasborn and Zwitserlood 2008).

A few recent studies have called into question the analysis of continuative and habitual marking offered by Hoiting and Slobin. First, Oomen (2016) conducted an elicitation task with one native signer. This task followed the general format of Dahl’s (1985) questionnaire, originally developed to elicit tense, mood, and aspect (TMA) in spoken languages, but which was later also used to investigate SSL (Bergman and Dahl 1994). This questionnaire consists of approximately 200 sentences preceded by a question or context sentence that is supposed to trigger specific TMA marking (if available in the language); the verbs in the target sentences are given in their infinitival forms as in (8). Sentences are presented in English and informants are asked to translate the sentences to a target language (Dahl 1985).

Oomen (2016) used this questionnaire in slightly adapted form: she brought the number of sentences down to 66, translated the sentences into Dutch, and adapted the items to the purpose of her study, such that they would (i) trigger continuative and habitual aspect and (ii) target verbs that were selected based on their phonological specifications.

Oomen’s results suggest that both aspect types are expressed in NGT by reduplication combined with non-manual markers. Specifically, continuative aspect was “consistently marked by means of a relatively slow reduplication of the verb’s movement and a synchronous back-and-forth movement of the head or body” (Oomen 2016: 43), while habitual aspect involved reduplication and synchronous left-to-right head and body movements. Habitual marking only occurred in past contexts. Notably, cont did not appear in the data at all: verbs with internal movement or body contact were inflected for continuative and habitual aspect in the same manner as those without these constraining features. Oomen offers regional variation as a potential explanation for the differences between these two studies: she reported on an informant from Amsterdam, while Hoiting and Slobin included informants from Groningen. It is well-known that there are lexical differences between the NGT dialects from these regions (Schermer 2004), but Oomen is the first to suggest a grammatical (i.e., inflectional) difference. It is also important to note that Oomen defined habituality in a more traditional way than Hoiting and Slobin, that is, excluding the notion of continuity.

In a subsequent study, van Boven (2018) presents a descriptive analysis of cont based on naturalistic corpus data, searching for this sign (i.e., the gloss door) in the Corpus NGT (Crasborn et al. 2008; Crasborn and Zwitserlood 2008). The results again challenge Hoiting and Slobin’s description, as well as Oomen’s observations: cont does occur in the data, but it co-occurs with a wide variety of verbs, not only with those involving body contact or internal movement. It also appears in various syntactic positions, sometimes preceding rather than following the verb. cont was mostly used to express continuative aspect.

Finally, habitual aspect in NGT was further investigated by van Boven and Oomen (2021). They, too, investigated naturalistic corpus data from the Corpus NGT. They searched for any sentence that refers to an event that is regularly repeated over an extended period of time. They conducted specific searches in the Dutch translations available in the corpus, searching for particles and adverbials that normally occur in Dutch habitual sentences (e.g., regelmatig ‘regularly’, per week ‘a week’, and elke dag ‘every day’). In total, they included 106 sentences in their data set, which, according to them, have habitual meaning. They then analyzed how this habitual meaning is encoded in the sentence. Since the present study builds on their methodology, and actually includes these data, more details about the search procedure are given in Section 3.1.

While they could not identify clear non-manual markers, van Boven and Oomen found that habitual meaning can be realized by reduplicating the verb, as well as by several adverbials. These two types of manual markers sometimes occur together (26.4 % of the 106 analyzed sentences), but also on their own (40.6 % for adverbs; 19.8 % for reduplication). Moreover, manual marking is not obligatory (13.2 % of the sentences does not involve any manual marking). (9a) shows a sentence with both reduplication and the adverbial always, while (9b) illustrates a sentence without any manual marking.

As for cont, van Boven and Oomen (2021) argue that it does not express habitual meaning: it occurs in only two out of 106 sentences, and in both cases, it triggers an additional continuative reading. Therefore, they conclude that cont actually encodes continuative rather than habitual aspect, which aligns with the fact (i) that Hoiting and Slobin (2001) maintained a definition of habituality that includes the notion of continuity, and (ii) that van Boven (2018) found this sign to mostly express continuity.

In the corpus data, habitual marking occurs with verbs with a variety of phonological properties, as well as both in past and non-past contexts. Van Boven and Oomen offer a methodological explanation for the differences between their study and the previous studies on NGT: corpus data display more variation than elicited data.

2.5 The present study

Strategies for encoding habitual and continuative aspect in NGT have been investigated previously. From these studies, it seems clear that reduplication plays a role. Yet, many questions remain unanswered, since there is considerable variation in the results. Moreover, none of the previous studies takes into account iterative aspect, while in other sign languages, iterative aspect has often been found to be encoded by reduplication as well. The present study therefore aims at answering the following questions:

1. Can reduplication of the predicate encode habitual, continuative, and iterative aspect in NGT?

2. Can different types of reduplication be distinguished for different types of aspect in NGT?

3. Which restrictions on aspectual reduplication can be identified in NGT?

   1. Are there phonological restrictions (internal movement/body contact)?

   2. Is there a relation between cont and reduplication for continuative aspect?

   3. Is there a relation between tense and reduplication for habitual aspect?

Previous studies on aspect in NGT included only one (Oomen 2016) or a few informants (Hoiting and Slobin 2001) from one sign region and focused only on elicited data that sometimes involved translations – and thus potential influence – from spoken Dutch (Oomen 2016). Some studies included naturalistic corpus data (van Boven 2018; van Boven and Oomen 2021), but none of the previous studies on NGT systematically analyzed all three aspect types that are potentially expressed by reduplication. Therefore, the current study is the first to comprehensively analyze and compare all three aspect types, combining naturalistic corpus data with data elicitation, in order to provide a comprehensive description of aspectual reduplication in the language.

3.1 Corpus search

The starting point of this study was a search in the Corpus NGT (Crasborn et al. 2008; Crasborn and Zwitserlood 2008), which contains recordings of 92 deaf NGT signers (age 17–84 years). The signers in the corpus performed several tasks, such as discussing certain topics and retelling video clips, resulting in semi-spontaneous monologues and dialogues.^[8] The corpus contains over 70 h of video data, and part of these data have been transcribed by fluent signers, using the annotation tool ELAN (Crasborn and Sloetjes 2008). The available transcriptions include annotations on gloss tiers, where signs are glossed on separate tiers for the dominant and non-dominant hand, and a translation tier, which includes Dutch translations (Crasborn et al. 2008).

The present study includes the corpus data collected for habitual aspect by van Boven and Oomen (2021), as reported above. In order to complement these data with constructions involving iterative and continuative aspect, I adopted the same strategy, i.e., I included sentences in my data set based on their meaning: the Corpus NGT is not annotated for aspectual distinctions or reduplication, and for this reason, searches had to be conducted on the translation tier, searching for particles and phrases that often occur in Dutch continuative and iterative sentences. Additionally, for continuative aspect, one specific gloss was searched for (hele ‘whole’), as this sign was expected to often occur in continuative sentences (e.g., in the context of the phrase hele dag ‘the whole day’). Table 1 provides an overview of the specific search terms, and specifies how many search hits there were in total per aspect type,^[9] and how many sentences were included in the end for each aspect type.

Each sentence that potentially involved the relevant aspect type was included. The inclusion criteria for the habituals are explained in van Boven and Oomen (2021: 165–166), but are briefly repeated here. Adopting definitions of habituals by Comrie (1976) and Rathmann (2005), van Boven and Oomen searched for any sentence that references an event that is regularly repeated over an extended period of time (see also the definitions in Sections 2.2 and 2.3). As Table 1 shows, 106 sentences met this criterion, and representative examples are given in (10) (van Boven and Oomen 2021: 166). For all of the examples extracted from the corpus, the corpus file number (CNGTxxxx), signer number (Sxxx), and begin time of the example (m:s.ms) are provided.

Table 1 also shows that 112 sentences showed up in the searches for habituals, but did not meet the inclusion criteria – i.e., they did not involve a systematically recurring event, and therefore were excluded. Two representative examples are given in (11) (van Boven and Oomen 2021: 166). Both showed up in the searches because the Dutch translations contain altijd ‘always’.

For continuative aspect, I adopted the definition given in Section 2.2, i.e., the sentence was included when “an event is progressing dynamically over a time frame opened up by an utterance” (Mair 2012: 803), and also when a situation is in progress. In total, 106 sentences met these criteria, as Table 1 shows. Two representative examples are given in (12).

As is evident from Table 1, 414 sentences showed up in the searches for continuative aspect but did not meet the criteria, and were thus excluded. Representative examples are given in (13) – these do not involve ongoing, progressing events or situations. These sentences showed up in the search hits because their Dutch translations contained (koppelen) aan het ‘connect to’ (13a) and (ingehaald) door ‘caught up with by’ (13b).

Iterative aspect had not previously been investigated for NGT. Therefore, I assumed the definition proposed by Rathmann (2005: 37), which is also given in Section 2.2.1 above: “The iterative morpheme contributes the meaning that multiple instances of the eventuality unfold in their own interval. A break is possible between each interval”. It was thus important that an event occurred multiple times, each instance with its own starting point (similar definitions are given in, e.g., Comrie [1976] and Bybee [1985]). In total, 28 sentences met these criteria (see Table 1), and representative examples are given in (14).

However, the vast majority of the sentences that showed up in the searches for iterative aspect, 241 in total (see Table 1), did not involve multiple occurrences of an event and were therefore excluded from the data set. Two representative examples are given in (15); both found their way into the search hits because the Dutch translations contain opnieuw ‘again’.

3.2 Data elicitation

In order to complement the corpus data, an elicitation task was designed, aimed at eliciting NGT sentences that contain continuative, habitual, and iterative aspectual marking.

3.2.2 Stimuli

The elicitation task aimed at eliciting aspect marking on six different NGT verbs with different phonological features, as shown in Table 2.^[10] For each verb, participants were presented with six items (i.e., there were 36 stimuli in total): two for each aspect type, once in a non-past context and once in a past context (note that NGT does not mark tense on the verb; rather, the tense information was provided by time adverbials such as yesterday).

Table 2:

Verbs included in the elicitation task.

Verb type
Body-anchored	Internal movement	No potentially constraining features
sleep	melt	clean
hug	talk	swim

Each elicitation item consisted of two parts: (i) a picture denoting the verb, for instance, a man who is swimming, and (ii) a question in NGT about that picture, preceded by a brief context, for instance, “This man lives close to the beach. What has he been doing the past few hours?”; see Figure 2.

Figure 2:

Example of an elicitation item: “[This man lives close to the beach.] What has he been doing for hours?”. Targeted answer: “The man has been swimming for hours.” (Stimulus picture from Dreamstime.com, Image ID: 75474659, Copyright Mimagephotography, https://www.dreamstime.com/mimagephotography_info).

Participants were asked to answer the question based on the picture they saw. Importantly, the question was phrased such that it would elicit an answer that contains marking for one of the aspectual distinctions under investigation. That is, the preceding context and the question specify how often a certain event occurs and/or how long the event lasts. Neither the context nor the question contained the target verb (in the example “to swim”), in order to avoid influencing the answers as much as possible. Participants were instructed to answer in complete sentences (in our example, an expected answer would be translated as ‘The man has been swimming for hours’).

Additionally, the base form of each verb was elicited (six items in total), in order to be able to compare the verbs in the sentences potentially marked for aspect to their base form articulated by the same participant. This was done by simply asking what the person on the picture is doing. Therefore, in total, the participants were presented with 42 elicitation questions and pictures. An English translation of the elicitation task, including instructions, is openly available (van Boven 2023). All items and instructions were presented by a deaf NGT signer, who was also consulted when designing the elicitation items.

Some of the contexts presented to the participants were based on the TMA-questionnaire developed by Dahl (1985) and adapted for NGT by Oomen (2016). That is, I based the items for continuative and habitual aspect on the contexts in this questionnaire for those verbs that can be depicted on a picture (i.e., sleep, hug, talk, and clean) – although where necessary, some changes were made – and complemented these with two new verbs (melt, swim), as well as new contexts for iterative aspect.

The stimuli were piloted with one deaf NGT signer, after which six stimuli were adapted since they did not elicit the targeted aspect types. The adaptations were done in consultation with a deaf signer.

3.3 Data annotation and analysis

The data annotation was done for both data sets in ELAN (Crasborn and Sloetjes 2008) – see Section 3.3.1. The statistical analyses were conducted in R (R Core Team 2008) and are introduced in Section 3.3.2.

3.3.1 Data annotation

The annotation tool ELAN (Crasborn and Sloetjes 2008) was used to annotate both data sets. To illustrate, a screenshot of the annotations is shown in Figure 3. For both data sets, I annotated (i) the aspect type of the sentence, (ii) whether the predicate was reduplicated, (iii) whether the sentence contained an adverb marking the aspect type, (iv) any non-manual markers potentially expressing aspect, and (v) any additional comments or relevant information about the phonological features of the predicate. For the elicited data, I additionally annotated (vi) any difference between the targeted aspect type and the aspect type actually produced by the participant, (vii) whether the sentence is past or non-past, and (viii) whether the participant added cont to the sentence. All data annotations for both data sets are openly available (van Boven 2023).

Figure 3:

Example of the annotations in ELAN.

For the habituals in the corpus data, I adopted the relevant annotations already made by van Boven and Oomen (2021) (i.e., on reduplication, adverbs, non-manual markers, past tense, and comments).^[12] I added annotations of the relevant phonological features.

As for the elicited data, I did not only annotate productions including the targeted verbs, but also sentences in which participants produced a verb that was not targeted, but which still involved one of the aspects. Sometimes, participants produced the targeted verb meaning, but used a different variant of that verb than was expected (e.g., the target form of melt involves internal movement, but some participants produced a variant without internal movement, but with body contact; see also Footnote 10) – these were also included.

4.1 Reduplication of the predicate

Table 4 shows how often reduplication is used to express an aspectual distinction in the corpus data. Clearly, reduplication of the predicate occurs, but not always: for habitual and continuative sentences, about half of the predicates is reduplicated in the corpus data, while for iteratives in the same data set about 30 % of the sentences does not involve a reduplicated predicate.

For the elicited data, some sentences could not be analyzed for reduplication, since for some predicates, no corresponding base form produced by the same participant could be elicited – be it because the elicitation of the base form did not succeed, or because the participants inflected a verb for aspect that was different from the targeted verb. For this reason, 14 continuative, 25 habitual, and 10 iterative sentences could not be analyzed for reduplication. Table 5 shows the results for reduplication of the predicate in the remaining elicited sentences – for habitual and continuative sentences in the data, about 20 % involves predicate reduplication, while this percentage is, again, about 70 % for iterative aspect.

As described in Section 3.2.2, it was analyzed statistically whether there is a difference between aspect types and between data sets in terms of reduplication. There was no significant difference between habitual and continuative aspect (p = 0.41); however, participants were five times more likely to use reduplication with iterative aspect than with habitual and continuative (odds ratio = 5.47, p < 0.001, z = 4.86, 95 percent confidence interval from 2.8 to 10.87). While I cannot conclude anything about the difference between habituals and continuatives in terms of likelihood of reduplication, iterative aspect is thus clearly more likely to be encoded by predicate reduplication than habitual and continuative aspect. Moreover, there was a significant effect of data set: participants were two times more likely to use reduplication in the corpus data than in the elicited data (odds ratio = 2.17, p = 0.03, z = 2.15, 95 percent confidence interval from 1.07 to 4.4). No significant interaction between data set and aspect type was found (p = 0.08 for continuative/habitual compared to iterative; p = 0.8 for continuative compared to habitual). An overview of all statistical data for the fixed effects is provided in Appendix B, and see van Boven (2023) for the complete analysis in R.

4.1.1 Phonological restrictions on reduplication

Following Hoiting and Slobin (2001), it may be expected that predicates that are not reduplicated are body-anchored and/or involve internal movement. In total, 68 body-anchored verbs, 37 internal-movement verbs, and 67 verbs without constraining features were elicited. Recall, however, that not all elicited verbs could be analyzed for reduplication. Table 6 shows the numbers excluding those verbs that could not be analyzed for reduplication.

Table 6:

Verb types in the elicited data; numbers after exclusion of predicates that could not be analyzed for reduplication.

Verb type	N (analyzed)
Body-anchored	56
Internal movement	18
No constraining features	49

Table 5 showed that out of the 123 elicited predicates that could be analyzed for reduplication, only 53 were in fact reduplicated. We can thus analyze whether the 70 non-reduplicated predicates involve internal movement or are body-anchored. However, this is not the case: out of the 70 non-reduplicated predicates, only 15 involve internal movement, and 26 are body-anchored. The other 29 verbs are not specified for one of the constraining features, suggesting that reduplication is optional even for verbs that are unconstrained; see Table 7.

Table 7:

Potentially constraining features and reduplication in the elicited data; numbers after exclusion of predicates that could not be analyzed for reduplication.

Body-anchored/internal movement	Reduplicated	Not reduplicated
Yes	33	41
No	20	29

The Pearson Chi-Squared Test I conducted in R (R Core Team 2008), in order to investigate the relation between potentially constraining features (body-anchoredness/internal movement) and reduplication (i.e., Table 7), shows no significant relation between the two variables (X ² (1, N = 123) = 0.05, p = 0.82). I cannot conclude anything as to whether there is a relation between the potentially constraining features on the one hand and predicate reduplication/zero marking on the other. An overview of all statistical data is provided in Appendix B, and see van Boven (2023) for the complete analysis in R.

We can, however, check for each aspect type whether predicates that involve internal movement or are body-anchored are reduplicated, and which ones these are. Table 8 shows the results.

Table 8:

Reduplication of “constrained” predicates per aspect type in the elicited data (BA = body-anchored, IM = internal movement); shaded cells highlight unexpected patterns.

Verb type	N	Continuative	Habitual	Iterative
Reduplicated BA	30	4 (13 %)	5 (17 %)	21 (70 %)
Reduplicated IM	3	0 (0 %)	0 (0 %)	3 (75 %)

As Table 8 shows, the phonological restrictions previously identified for habitual and continuative aspect do not hold for iteratives in the elicited data: 24 predicates that are body-anchored or involve internal movement are reduplicated for this aspect type. For continuative and habitual aspect in the elicited data, the previously identified constraint on reduplicating internal-movement predicates seems to hold. However, for the phonological feature body-anchoredness, the picture is less clear, as there is a total of nine cases where a body-anchored predicate is reduplicated (the shaded cells in Table 8), namely hug (n  = 3), sleep (n = 1), and melt (n = 5). Although at first glance, it thus seems that these contradict the phonological constraint previously identified, it is worth inspecting them further.

The five instances of melt involve a variant of this verb that was analyzed as body-anchored, because the non-dominant hand is the place of articulation, while the dominant hand moves. Here, the verb actually does not contact the body, but rather the other hand – and in fact, the hands are often in proximity to each other, rather than actually making contact. This verb and its reduplicated form are illustrated in Figure 4a. For hug, it appears that it actually cannot be reduplicated in the expected way, that is, by repeating the entire path movement performed by the hands. Still, some participants employ an alternative strategy to reduplicate this body-anchored verb. The data suggest that hug is lexically specified for a short, swaying body movement. Some participants reduplicate the verb by repeating this body movement rather than the path movement of the hands. The hands thus retain contact with the body, as illustrated in Figure 4b. Finally, for sleep, it appears that the repetition may well have been a slip of the hand (but see an alternative explanation below: it contacts not the trunk, but the head).

Figure 4:

Reduplication of body-anchored melt (a) and hug (b).

For the corpus data, it was more challenging to analyze potential effects of phonological features on reduplication, since it was obviously impossible to deliberately include base forms with different phonological specifications in this data set. Still, when we take a closer look at the corpus data in light of the patterns observed in the elicited data, some striking observations can be made. For both continuative and habitual contexts, it appears that body-anchored predicates can be reduplicated by repeating the path movement, but this only holds for a specific type of body-anchoredness in the corpus, namely predicates that contact the non-dominant hand, e.g., play-soccer in Figure 5a (much like melt in the elicited data), or the head, e.g., cry in Figure 5b (much like sleep in the elicited data).

Figure 5:

“Body-anchored” predicates play-soccer (a) and cry (b). For aspectual inflection, both can be reduplicated by repeating the path movement.

As for predicates with internal movement in the corpus (e.g., handshape change in talk in Figure 6a), it appears that they are not reduplicated for habitual and continuative aspect, much like the elicited data. There is, however, one exception: wiggling of the fingers (as in fingerspell in Figure 6b and type) does not block aspectual reduplication.

Figure 6:

“Internal-movement” predicates talk (a) and fingerspell (b). For aspectual inflection, only fingerspell can be reduplicated by repeating the path movement.

Thus, the data suggest that, in the context of aspectual marking, both body-anchoredness and internal movement should be defined more narrowly – in fact, body-anchoredness could be re-defined as the location feature [trunk], while internal movement only comprises a change in finger position (or aperture, e.g., in the case of talk a change from open to closed). Moreover, as for predicates involving [trunk], it appears that only reduplication of path movement is blocked rather than all types of repetition (as evidenced by the repetition of body movements for hug in Figure 4b). These refinements should be taken with due caution, however, as our data do not involve negative evidence, and the data extracted from the corpus do not systematically include verbs for all phonological specifications. It remains to be investigated whether these constraints also hold outside of the current data set.

Finally, another important note should be added: the phonological restrictions identified in the data only apply when reduplication expresses specific types of aspect only. In some of the corpus sentences, reduplication seems to also mark multiple arguments (object and/or subject) on the verb, i.e., plurality, as in (16), and it is sometimes ambiguous whether reduplication actually encodes plurality of arguments, aspect, or both.

(16)

[…] hearing talk.2h.alt+ through-each-other theater palms.up

‘[…] All of the hearing people were (continuously) talking over each other during their play.’

[CNGT0294; S018; 02:09.81]

It seems that in cases in which reduplication (also) expresses plurality, verbs with an opening or closing of the hand can be reduplicated, as evidenced by the reduplication of talk in (16), which involves a handshape change – for this type of marking (but not for “pure” aspectual marking), the second hand is often added, and sometimes the hands move in alternation, as (16) also demonstrates. This type of reduplication occurred with internal-movement verbs where the hands open and close, like talk, observe, and accept. For corpus data, it is notoriously difficult to disentangle the functions that reduplication of the predicate may have in a specific sentence. This may also partly explain why participants were two times more likely to reduplicate the predicate in the corpus data than in the elicited data (see above), as such cases, in which reduplication may have multiple functions, have also been included in the corpus data, while other potential meanings of reduplication were controlled for in the elicited data (see also Section 5.3 for this and further possible explanations for the difference).

4.1.3 Movement characteristics

Cross-linguistically, different aspect types have been found to be distinguished from each other by differentiating the manner and rate of the movement of the reduplication cycles (e.g., Klima and Bellugi 1979; Rathmann 2005).^[14] In the elicited data, we observe one striking difference between iterative aspect on the one hand, and habitual and continuative aspect on the other: in the former, the movement cycles are often separated from each other by means of pauses, which is not the case for the latter two aspect types, where the movement cycles are uninterrupted. Iterative reduplication with pauses is illustrated in Figure 7.

Figure 7:

Iterative reduplication of clean with a pause in between reduplication cycles.

Finally, a characteristic of aspectual reduplication which had not been described in previous studies but which appeared in the data is spatial displacement of the predicate: sometimes, the reduplicants are articulated at different locations in the signing space. This is illustrated for swim in Figure 8, where there are four articulations of the verb, alternately on the left and right side of the signer.

Figure 8:

Iterative reduplication of swim with displacement of the reduplicants in space.

4.2 Other aspectual markers in NGT

Apart from reduplication of the predicate, two other aspectual markers could be identified in the data: the sign glossed as cont (Section 4.2.1) and some non-manual markers (Section 4.2.2).

5.1 Reconsidering phonological constraints on aspectual reduplication in NGT

The present study offers a new perspective on the phonological constraints on aspectual marking in NGT that have previously been put forward by Hoiting and Slobin (2001). Considering all aspect types in the elicited data together, no significant relationship was found between the phonological features of the predicates (i.e., internal movement/body-anchored or not) and reduplication, and thus I cannot conclude anything as to whether there is a relationship. Still, if we have a closer look at specifically the body-anchored and internal-movement predicates in imperfective contexts in our data, it seems that the constraints “body-anchored” and “internal movement” require a more precise definition, as proposed in Section 4.1.1. Independent evidence for the more specific definition of these constraints comes from phonological models developed for sign languages. First, for the “body-anchored” constraint, I argued that in the current data predicates articulated on the trunk are phonologically constrained, while predicates contacting other body parts are not. In phonological models of sign languages – e.g., Sandler’s (1989) Hand Tier model and Brentari’s (1998) Prosodic model – it is assumed that signs are specified for a major location, the trunk or torso being one of them, as opposed to, for instance, the non-dominant hand and the head. The fact that these major locations are distinguished in these models supports the proposal that one major location, the trunk, blocks reduplication, while the others do not.

Second, for internal movement, I noted that, in our data, handshape change blocks reduplication, but finger wiggling does not. In the phonological models mentioned above, finger wiggling and handshape change are represented differently, despite the fact that both are hand-internal movements. Recall that finger wiggling has been subsumed under trilled movement, and it is represented as such in both models. Sandler (1993: 252) defines trilled movement as “rapidly repeated hand-internal movement”, and in her Hand Tier model, signs with trilled movement have a feature [trill]. In Brentari’s (1998: 164) Prosodic model, trilled movement is a so-called “articulator-free feature”. Handshape change, however, is not represented by specific features in the phonological models. In the Hand Tier model, a change in finger position is represented by branching at the finger position node (see, e.g., Sandler 1993; Sandler and Lillo-Martin 2006). In the Prosodic model, one underlying handshape is specified, while the other (redundant) handshape can be predicted from the opposing value of the underlying shape at the aperture node (Brentari 1998). The fact that finger position/aperture changes on the one hand, and trilled movement on the other, are represented separately in both models is in line with the observation that only one of them blocks reduplication. To further investigate this, it would be interesting to test whether other types of trilled movement (such as circling, rubbing, and hooking, as identified by Brentari) also block reduplication in NGT.

Based on these observations, I propose to revise the previously identified phonological constraints on aspectual reduplication in NGT (Hoiting and Slobin 2001), as in (20).

Interestingly, these constraints do not apply to iteratives in the data set. Indeed, as mentioned in Section 4.1, participants were five times more likely to use reduplication with iterative aspect than with habitual and continuative aspect, which may result from this difference in terms of constraints on reduplication. Possibly, combining a change in finger position with movement repetition is too complex when the movement cycles are uninterrupted, as is true for habituals and continuatives. In iteratives, the reduplication cycles are separated from each other by means of pauses, making the movement combinations less complex. As for the [trunk] location, it is less evident why a combination with repeated movement would yield a phonologically too complex form (in fact, there are verbs in NGT which are lexically specified for both body-anchoredness and repeated movement, such as be-scared, which contacts the trunk). Still, the resistance of body-anchored signs to undergo reduplication appears to be a more general phenomenon: for plural reduplication, van Boven (2021) found that NGT nouns specified for inherent repetition or body contact are less likely to be reduplicated than nouns without those features, although reduplication was not completely blocked. In other sign languages, body-anchoredness has been observed to completely block plural reduplication (Pfau and Steinbach [2005] for German Sign Language; Pizzuto and Corazza [1996] for Italian Sign Language; Sutton-Spence and Woll [1999] for BSL).

The fact that iterative reduplication is never blocked in the data suggests that there is an interaction between the phonological make-up of a sign and the aspect type: the phonological constraints are specific to the inflectional morpheme. An interaction between the phonology, morphology, and reduplication type has also been described by Harley and Leyva (2009) for Hiaki, an Uto-Aztecan language, although the findings do not exactly correspond to what is found here for NGT. They report five different reduplication types, and the application of two of the reduplicative allomorphs (the disyllabic and closed-syllable reduplication) is determined by a combination of not only phonological, but also morphological (transitivity-marking suffixes) properties of the stem.

Yet, it is not immediately clear which morphological distinction we are dealing with in NGT: iterative versus habitual versus continuative, or perfective versus imperfective more generally. This matter is the focus of the next section.

5.2 Reconsidering grammaticalized aspectual distinctions in NGT

The data in this study were collected from a semantic starting point, that is, the sentences were divided into habitual, continuative, and iterative aspect based on their meaning, and only after that, their form was analyzed. In doing so, I identified potential phonological constraints on habitual and continuative reduplication, but not on iterative reduplication, where the movement cycles are separated by means of pauses. Additional evidence for the distinct status of iteratives comes from another striking feature: the sign again can sometimes intervene between different instances of the verb, as in (21).

Here, the additional material appears in between single instances of the verb, i.e., instances with one movement cycle. Two potential explanations for the insertion of again can be offered. It is possible that the pauses in iterative reduplication allow for the insertion of (phonologically light) material in between reduplication cycles. Or, alternatively, the insertion of again in between several instances of the verb actually came first, and is in the process of grammaticalization into the iterative reduplication with pauses between reduplicants. The data are uninformative as to which explanation is the correct one, leaving this question open for future analyses. For the other two reduplication types, this type of inter-cycle insertion is never observed in the data.

The fact that iterative reduplication is optional in both data sets, despite not being phonologically constrained, suggests that it is not (yet) completely grammaticalized (pointing towards the second explanation offered in the previous paragraph). Indeed, more generally, there is quite some variation, since reduplication alternates with sentences without reduplication for all three aspects. This is reminiscent of Johnston et al.’s (2015) study on what they call finish-type signs in Auslan, which express perfective aspectual meaning: they suggest that the grammaticalization of such signs “is not well advanced” (p. 152) (see Johnston et al. [2015] for more elaborate discussion). The present data suggest the same for NGT aspectual reduplication.

Other studies, again in line with the present one, also describe optionality and variation in sign language aspectual marking. Recall from Footnote 6 that aspectual reduplication (denoting repeated or continuing events) was found to be optional in Auslan (see Gray 2013),^[16] and Palfreyman (2019) describes variation in the expression of completion in the urban sign language varieties of Solo and Makassar (Indonesia). Such variation and optionality is not entirely unexpected, as it has also been described for other grammatical domains in NGT (e.g., Oomen and Pfau [2017] for variation in negation; van Boven et al. [2023] for the optionality of negative concord), as well as in other sign languages (e.g., Fenlon et al. [2018] for optionality of verb modification in BSL, and the factors conditioning it; Palfreyman [2019] for variation in negation in the Solo and Makassar varieties).

When there is reduplication, however, iteratives are clearly distinct from habituals and continuatives in the data. Recall from Section 2.2.1 that earlier research on other sign languages already showed that aspect types can be distinguished by movement modulations such as adding pauses in between reduplication cycles (e.g., Klima and Bellugi 1979). The fact that NGT iterative reduplication is distinguished from other aspects by means of this modulation is thus not surprising from an intra-modal perspective. Interestingly, several spoken languages have also been described to employ different types of reduplication for different aspect types. For instance, in Coos (a now extinct isolate from Oregon), reduplication of the first syllable expresses “intensity of action, repetition, duration, and customary action” (Frachtenberg 1922: 377), while reduplication of the final syllable expresses “distribution, mutuality, and, in intransitive verbs, an action that is performed now and then” (Frachtenberg 1922: 380). The reduplication types thus have several different functions, some of them aspectual. Mithun (1999) presents similar findings on the Salish language family, where most languages are described to have three reduplication types that have different (some aspectual) functions. Similarly, in Hiaki, different reduplicant shapes are semantically contrastive, but only for specific verbs (e.g., noka ‘speak’ and vahume ‘swim’). A light-syllable reduplicant is used to derive habitual meaning for these verbs, while a reduplication + gemination form expresses emphatic, idiosyncratic, or iterative meaning (Harley and Leyva 2009). Although this is not described as a general pattern across all verbs in Hiaki, the parallel with the NGT data (and data from other sign languages) in using different types of reduplication to express iterative or habitual meaning is notable.

While the iteratives I analyzed thus appear to be distinct from the habituals and continuatives in terms of their reduplicative form, a formal distinction between the latter two could not be established in this study. In fact, in the present data, they are expressed by the same type of reduplication, without pauses in between, which appears phonologically constrained. I take this to suggest that the semantic distinction between habitual and continuative may not be grammaticalized in NGT in the form of verbal inflection. Recall from Section 2.1 that the habitual and continuous aspects have been proposed as a subdivision of imperfective aspect more generally (Comrie 1976), while iterative aspect is perfective. It thus might be that, when there is reduplication, NGT distinguishes perfective and imperfective, but does not make a further formal distinction between habitual and continuous. If this is indeed the case, then the phonological constraints in (20) can be revised again, as in (22).

From a cross-modal perspective, this finding is not surprising. As described in Section 2.1, Bybee (1985) reports that most spoken languages in her sample distinguish only the imperfective and the perfective inflectionally (see also Dahl and Velupillai 2013b). This pattern was actually more common than further distinguishing habitual and continuous. Compare (23), from Spanish, where the imperfect can express habitual or continuative meaning,^[17] to (24ab), where NGT also uses one and the same form to express habitual (24a) and continuative (24b) meaning.

From an intra-modal perspective, however, the findings presented here are more striking. In the literature on aspect marking in sign languages, specific verbal modulations for habituals or continuatives are often mentioned, as we also saw in Section 2.2.1 (e.g., Cabeza Pereiro and Fernández Soneira [2004] for Spanish Sign Language; Rathmann [2005] for ASL; Sutton-Spence and Woll [1999] for BSL). The fact that the NGT data exhibit one form for both habituals and continuatives, without modulating the movement, situates NGT differently from other sign languages in the landscape of aspectual inflection, thus adding to our understanding of intra-modal variation in this domain.

So far, it has been assumed that habitual and continuative aspect are both instances of the imperfective viewpoint. Note, however, that Rathmann (2005) analyzes these as instances of situation aspect in ASL.^[18] While our data are not informative as to whether the continuative/habitual morpheme(s) could be analyzed as an instance of situation-type aspect, the NGT data are different from ASL in at least one way: we cannot formally distinguish habituals from continuatives. To put it differently: the semantic distinction between habitual and continuous does not appear to be grammaticalized based on the NGT data – irrespective of whether they are situation- or viewpoint-type. For now, I assume that they are instances of imperfective viewpoint, and leave this matter for future studies. Such studies could focus on the question whether the reduplication that is used to encode both continuatives and habituals can combine with a perfective marker in NGT. If we are in fact dealing with imperfective viewpoint, the expectation would be that this combination is ruled out.

This outcome stands in sharp contrast with findings reported in previous studies on NGT: while Hoiting and Slobin (2001) and Oomen (2016) did not agree on the specific form of the marking, they both identified a formal distinction between habituals and continuatives. A likely explanation lies in the different methods of the studies: the previous studies on NGT only considered elicited data from a limited number of participants. The same is true for several previous studies on other sign languages (e.g., Cabeza Pereiro and Fernández Soneira [2004] consider only two participants for LSE). In contrast, this study draws on corpus and elicited data and includes more signers, and consequently, the data more closely represent the language as it is actually used. Further methodological considerations, as well as potential sociolinguistic variation, will be discussed in the next section.

5.3 Methodological and sociolinguistic considerations

The combination of corpus data and elicited data has several important implications that should be taken into account. First, there are some, mostly quantitative, differences between the data sets. For instance, while iteratives are relatively rare in the corpus (n = 28), they are the most frequent aspect type in the elicited data (n = 62). Footnote 13 already indicated that, while there is a significant relation between the data set (corpus or elicited) and aspect type (habitual, continuative, or iterative), this result might, in fact, be due to several dependencies in the data, most likely the fact that I targeted an equal amount of sentences for each aspect type in the elicited data, which may have resulted in an unusually high number of iteratives as compared to the naturalistic language use in the corpus data. In addition to this, the fewest searches were performed for iteratives on the translation tier (see Table 1), indicating that Dutch might have fewer particles expressing iteratives than particles expressing the other two aspect types. Iteratives are thus difficult to find in the corpus because they might be less likely to be overtly marked in Dutch.^[19]

Another difference between the two data sets lies in the frequency of predicate reduplication, as the statistical analysis showed that corpus participants were two times more likely to reduplicate the predicate than participants in the data elicitation (see Section 4.1). Again, there are a few possible explanations. In the elicited data, phonological features of the predicates were controlled for, and the stimuli thus included a fair number of predicates that cannot be reduplicated. Obviously, in the corpus data, it was impossible to control for such phonological features. Another potential explanation is that in the corpus data, different functions of reduplication were sometimes difficult to disentangle. This challenge has already been illustrated in Section 4.1.1 with (16), repeated here as (25), where it is ambiguous whether the reduplication expresses plurality of arguments, aspect, or both. It is thus quite possible that I sometimes annotated reduplication of the predicate as encoding aspectual meaning, when it actually was not used as such, resulting in an overestimation of how often the predicate was reduplicated for aspect. In the elicited data, this likely did not happen, as the potential meanings of reduplication could be controlled for (e.g., it was ensured that none of the sentences had a distributive meaning).

Finally, the difference in frequency of reduplication between the two data sets might also be due to how the data sets were compiled. I collected the corpus sample for the largest part by searching on the translation tier, selecting sentences with aspectual meaning. This way of collecting the sample did in principle not influence the type of aspect marking I would find: aspectual meaning could be expressed in several ways – e.g., by particles, reduplication, or both. This is different for the elicited data, given that there were aspectual particles in the stimuli in order to make clear which aspectual meaning was targeted. Recall, for example, the stimulus in Figure 2, which translates as “What has he been doing for hours?”. Participants, answering the stimulus questions in full sentences, often repeated this overt aspectual marker, in this case, “for hours”. It could be the case that such lexical indicators removed the need to also reduplicate the predicate. At the same time, it is also clear that lexical indicators and reduplication are not completely in complementary distribution: for instance, in the corpus data, reduplication and adverbs co-occur in 26.4 % of the habituals (see van Boven and Oomen 2021), in 27.4 % of the continuatives (these numbers might be higher if only unconstrained predicates are included in the data), and in 21.4 % of the iteratives. Also, if they were completely complementary, one would expect no reduplication at all in the elicited data – in fact, double marking is not uncommon in sign languages (e.g., in several sign languages, there is double marking of agreement, by inflection of the main verb and an auxiliary; see Steinbach and Pfau [2007]). Still, it is expected that the elicited data do not reflect the frequency of aspectual reduplication in natural language use, while the corpus data likely do.

Second, there is some variation between signers in our data. For instance, in the elicited data, one signer is overrepresented in reduplicating body-anchored signs for continuative/habitual aspect, namely the predicates melt, which involves contact with the non-dominant hand (see Figure 4a), and hug, a body-anchored sign (see Figure 4b), for which he repeats the body movement rather than the path movement: four, i.e., half, of the instances come from this signer. On the other hand, there is one signer in the elicited data who never reduplicates body-anchored and internal-movement predicates for habitual or continuative aspect. This suggests that it may, to some extent, be signer-dependent how strictly the phonological constraints are interpreted. Potentially, sociolinguistic factors play a role here. Interestingly, the signer who strictly maintains the phonological constraints is from Groningen, and Hoiting and Slobin (2001), who identified these constraints, based their observations on informants from the same region. The participant who is overrepresented in the reduplication of melt and hug, on the other hand, is from Amsterdam, and Oomen (2016), who did not find any restrictions on reduplication, tested a participant from this region. This is further evidence for regional variation in aspectual marking – something that Oomen already suggested as an explanation for the diverging findings. This should be researched further, by more systematically including signers from different regions.^[20]

Because of the nature of the data, I cannot answer all open questions. A first drawback is that the data do not provide negative evidence. In order to further test phonological constraints on aspectual reduplication, grammaticality judgment tasks should be conducted. Moreover, the data suggest that reduplication is not obligatory, even for non-restricted predicates: the elicited data included 29 non-restricted predicates that were not reduplicated for habitual or continuative aspect (see Section 4.1.1). Iteratives appear not to be subject to phonological constraints, and still 29 % of the elicited iteratives does not involve reduplication of the predicate. The corpus data also show inter-signer variation in the use of reduplication: some signers occasionally use reduplication to mark aspect on unconstrained predicates, but not always. We could thus conclude that aspectual reduplication is not obligatory, but it still remains unclear whether it is truly optional, or whether other (potentially syntactic or prosodic) factors are at play here. I consider this an avenue for further research.

Finally, as mentioned in Section 3.3.2, the statistical analyses reported here have some drawbacks. I reported the results of a logistic regression model, but the estimates should be taken with due caution, given the small sample size (even though the corpus and elicited data were taken together) (see Moineddin et al. 2007). Further, given that the elicited data constitutes an even smaller sample, for this data set, I reported chi-squared analyses, which assume that the observations are independent of each other, while this is not the case in the current data set (e.g., multiple observations per participant). It is hoped that in future research, the outcomes reported here can be investigated further based on a larger sample, in order to conduct logistic regression models and report the (more reliable) estimates. However, given the (relatively small) population under study, this may be a challenging undertaking.