Unbounded repetition, habituality, and aspect from a comparative perspective

2.1 Definition of habituality and unbounded repetition

The term ‘habitual’ or ‘habituality’ is used by various linguists (see Boneh and Jędrzejowski [2019] for an overview). From a theoretical perspective, the term ‘habituality’ can be used in three different ways: (i) to refer to a language-specific category; (ii) for a comparative concept in the sense of Haspelmath (2010); and (iii) for a universal category (the universal status may then be due to biological factors or universal communicative principles). In this paper, I will treat habituality as a comparative concept. Comrie (1976: 27–28) uses the following crosslinguistically applicable general definition of habituals; that is, forms/constructions that express a habitual meaning:

[T]hey [habituals] describe a situation which is characteristic of an extended period of time, so extended in fact that the situation referred to is viewed not as an accidental property of the moment but, precisely, as a characteristic feature of a whole period.

In Comrie’s analysis, habituals can be dedicated habitual forms or constructions which do not explicitly refer to repetition or constructions that contain an ipfv verb form usually combined with expressions of unbounded repetition, such as always, often, every day. The term ‘unbounded repetition’ is used to indicate that there is no limit to the number of repetitions (within a specific timeframe), in contrast to bounded repetition, as in twice, three times, a couple of times, etc. where an event is repeated a (specific) number of times (for pluractionality of situations, see also Shluinskij [2005]). Consequently, the unbounded repetition readily suggests that the repeated situation is characteristic of an extended period of time, and the world’s structure (cf. Langacker 2000: 252).^[1] Comrie also points out that a habitual meaning is sometimes expressed by an ipfv verb without habitual marking. Comrie’s approach to habituality and aspect is similar to Dahl’s (1985: 79, 95–98), although Dahl makes a further subcategorization of habituals into (i) habitual, (ii) habitual-generic, (iii) habitual past.

Comrie’s definition provides a good starting point for the analysis of habituals, with the additional comment that his notion of “extended period of time” is a relative notion, and that this requirement also depends on the type of habitual construction. Compare French (1) with the expression for ‘every night’ and an ipfv past with English (2) with used to:

One could argue that this shows that (1) is not a habitual sentence in the strict sense. Sentence (1) can be compared to the sentence in the present tense: Pierre goes to the movies every night. This clearly indicates something that is characteristic of an extended period of time, without any temporal restriction. Such sentences have a clear habitual character since they indicate something of the current world’s structure. When we talk about past events, it is possible to indicate that the unbounded repetition took place during a longer or shorter stretch of time (when I was young, last week). The shorter the stretch of time, the less clear is the idea that something that is characteristic of the world’s structure or that something that is someone’s habit is expressed. Nevertheless, because it is difficult to draw a clear line between an extended period of time and a non-extended period of time, I think it is preferrable to include all sentences with unbounded repetition into the category of habitual sentences.

In general, the definitions of habituality given within descriptive linguistic frameworks are similar to those given in formal semantic frameworks even though formal semantic approaches to habituality often seem to suggest that the category of habituality is universal. Krifka et al. (1995: 17–19, 25, 32, 36) argue that habituals are a subcategory of ‘characterizing’ sentences (which describe some ‘essential property’), i.e. sentences that contain an episodic verb – which reports on a specific event or occasion – and are neither stative nor generic. According to Krifka et al. (1995: 17) “[h]abitual sentences express generalizations over situations that are specified by the corresponding episodic verbal predicate.” In their formal semantic description of habituality a so-called generic operator (GEN) plays an important part. Krifka et al. (1995: 25) also remark that “the genericity of characterizing sentences takes sentential scope, and […] should be treated as similar to adverbs such as always, often, seldom, and the like. Adverbs such as usually, typically, and in general are closest in meaning to the generic operator, which often is not realized phonologically”. As such, in their view the more prototypical instances of habituality express a law-like pattern, which occurs often and regularly, and which do not explicitly indicate repetition. Carlson (2009: 376) argues that habituality can be expressed by dedicated markers, such as English used to, tend to but it is also a component of the meaning of most frequency adverbs, including often.

Bertinetto and Lenci (2012: 860) provide an analysis of habituality within a formal semantic framework that also explicitly takes verbal aspect into account. They argue that iterative sentences are mostly pfv,^[2] whereas habitual sentences (in languages with explicit aspectual marking) are ipfv by default (cf. Mari et al. [2013: 47] for a similar claim). Put differently, in my terminology, the property ‘ipfv’ is part of their comparative concept of habituality. According to Bertinetto and Lenci, habitual sentences, unlike iterative ones, are intrinsically characterizing: they attribute a defining property to the intended referent(s). In Bertinetto and Lenci’s analysis, iterative sentences, for example sentences with twice, seven times, differ from habitual sentences in the following ways: (i) the iteration is specifiable, (ii) iterativity is, unlike habituality, impossible in the present domain, (iii) the timeframe in which the iteration occurs is strictly delimiting, and (iv), which is an extension of (i), the exact number of occurrences is determinable or at least potentially determinable. It should be noted that the distinction between habitual sentences and iterative sentences proposed by Bertinetto and Lenci is not identical to the distinction between unbounded and bounded repetition because some sentences that contain expressions of unbounded repetition are still classified as iterative by Bertinetto and Lenci (2012). Take the French example (3), which contains a passé composé, defined by them as a pfv marker, and an expression of unbounded repetition:

Bertinetto and Lenci classify this sentence as iterative and not habitual for two reasons: (i) the sentence contains a pfv marker, and (ii) the iteration has characteristics of iterative sentences: for example, the event takes place within a limited timeframe (‘last year’) and hence, they seem to argue, the exact number of occurrences is potentially determinable. Another claim by Bertinetto and Lenci (2019: 868–871) and Lenci and Bertinetto (2000: 254) is that “the pfv and habitual ipfv aspects also differ because the former is typically intensional whereas the latter is typically extensional”. In their view, this is true for all languages with pfv and ipfv aspect, such as Romance and Slavic languages (Lenci and Bertinetto 2000: 215, 216, 224). They also suggest a way to model this within a formal semantic-syntactic framework, i.e. by interpreting “the generic operator as a modal quantifier” (Lenci and Bertinetto 2000: 256). In their approach, the difference between intensional (habitual) and extensional, for example iterative, sentences boils down to the difference between accidental generalizations (in their view expressed by the pfv) and nomic generalizations (in their view expressed by the ipfv) (Lenci and Bertinetto 2000: 254–255). The intuition behind this, as I understand it, is that nomic generalizations do not just refer to specific or concrete individual situations (episodic fact; cf. Lenci and Bertinetto 2000: 265), but to any situation to which the law applies (normative fact; cf. Lenci and Bertinetto 2000: 265). However, it is often not clear how one can objectively determine (apart from the presence of aspect itself) whether a sentence is iterative or not. This can be illustrated with the following Greek example, which seems to have a characterizing function that is typical of habituals but without any characteristics of iterative sentences even though we find a pfv verb form:

In my view, the main issue with Bertinetto and Lenci’s definition of habituality is that it requires the property ipfv (non-pfv), even though other properties that are required (intrinsically characterizing, and non-potentially determinable) can be found with pfv verb forms as well. Of course, one can argue that an ipfv verb form is very suitable to indicate nomic generalizations. The meaning of non-totality (envisaging the event within its limits or not envisaging the event in the attainment of its relevant limit), as expressed by the ipfv, accords well with the generic or unbounded character of the nomic generalization. But, as I already showed with respect to the Greek example (4), sentences with a pfv may also have a law-like character. Furthermore, in some languages constructions that refer to concrete ‘episodic’ events can get a habitual interpretation because the event is presented as typical for how things are in general, suggesting a nomic meaning. An example is the so-called exemplary meaning of the pfv present (prototypically used as a pfv future) in Russian, to indicate a past habitual event. This construction occurs with pfv present verb forms with or without the past habitual marker byvalo:

This type requires the idea of picking an individual instance, to illustrate something general, to some extent similar to English would as a marker of habituality, and Carlson and Spejewski’s (1997: 143) analysis of constructions with whenever. They argue that in sentences with whenever the main clause is episodic, but the sentence as a whole is generic (habitual).^[3] As such, even though it is possible to exclude pfv habituals by definition, I think that this would limit the definition of habituality in an undesirable way.

To conclude this section, based on the data given in the literature, it is possible to distinguish the following types of habituals, which all adhere to the semantic definition of habituals as defined by Comrie (1976), which I treat as a comparative concept:

1. Sentences with dedicated habitual constructions or habitual aspectual markers (e.g. English used to, Russian byvalo, Ossetic -iu, South Conchucos Quechua past habitual marker -q).

2. Sentences with adverbs (or similar parts of speech which can modify a predicate) that explicitly indicate the idea of a habit or something which occurs on a regular basis (e.g. English usually, Russian obyčno).

3. Sentences with expressions of unbounded repetition (every day, often, always, sometimes), which express their own meaning of frequency but at the same time adhere to the definition of habituality.

4. Sentences with ipfv verb forms (e.g. Russian ipfv, Georgian imperfect) without any of these markers, which adhere to the definition of habituality. As Russian (5) shows, perfectives can also indicate a habitual meaning without an expression of unbounded repetition.

I do not think that it is possible from a crosslinguistic perspective to clearly delineate (i) and (ii), but the main difference can be illustrated with English, where used to + infinitive is a grammaticalized construction which functions as a predicate, whereas usually adheres to the general rules of other adverbs and modifies a whole proposition or meaning as expressed by a predicate. At the same time, English usually also shows formal and semantic similarities to the adverbs given under (iii). In my view (i) and (ii) must be seen as habituals in the strictest sense since they are specialized in expressing a habitual meaning, whereas (iii) and (iv) must be seen as forms and constructions which indicate a habitual meaning, similar to (i) and (ii), in addition to having their own semantics. Within category (iii) it could be argued that meanings like ‘sometimes’ are not habitual since they do not indicate what took place in the majority of occasions, something which according to some scholars is an inherent part of the meaning of habituality (for example Dahl 1985: 97).^[4] However, in my view, it is better to treat the category of unbounded repetition as a single category.

In order to test the relationship between pfv or ipfv aspect and habituality, the habituals mentioned under (i) are not suitable because they often do not combine with an ipfv or pfv verb form, or because they express an aspect of their own. Because of this, I will focus on sentences that contain an expression of unbounded repetition, such as every day, often, or always, as a more objective way to test the relation between habituals and aspect. The choice to test the relationship between aspect and habituality by looking at contexts of unbounded repetition, is also motivated by the fact that in the literature (Comrie 1976; Dahl 1985), many of the examples of pfv habituals contain expressions of unbounded repetition.^[5] As such, the inclusion of expressions of unbounded repetition (irrespective of verbal aspect) may provide interesting crosslinguistic insights. For some languages, I will, however, also consider dedicated habitual expressions that adhere to Comrie’s definition, in order to see whether they can be combined with ipfv or pfv markers. I will reserve the term ‘iterativity’ for expressions of bounded repetition (twice, seven times, many times, etc.). Such expressions typically occur with the pfv aspect (e.g. Dahl 1985: 78) even though there also language-specific differences. To give an example, Barentsen et al. (2015), Barentsen (2018) show that in Russian the ipfv is the most common aspect in contexts of bounded repetition, but in Czech the pfv is the most common, whereas Bulgarian, Polish, and BSC taken an intermediate position.

2.2 Definition of pfv and ipfv

The term (verbal) aspect can be used for constructions or forms that provide an internal temporal perspective on an event (i.e., state of affairs), that is, how an event extends over time: for example, whether it is seen as a whole (something complete or total), often called ‘pfv’, or not. Aspect deals with the specific temporal way in which the speaker confines a claim with respect to a situation that extends over time (Klein 1994: 4, 99). Different authors describe terms like pfv and ipfv in different ways, using different theoretical concepts and basing their descriptions on different languages (see Sasse [2002] for a comprehensive overview). The terms pfv and ipfv may refer to meanings within one specific language (e.g., the pfv form, which expresses a pfv meaning) or to comparable meanings across languages (e.g., the pfv in Slavic). In the latter case, it must be seen as a comparative concept in the sense of Haspelmath (2010). Dahl (1985: 78), who provides a comprehensive comparative overview of aspect, gives the following comparative definition for the pfv:

A PFV verb will typically denote a single event, seen as an unanalysed whole, with a well-defined result or end-state, located in the past. More often than not, the event will be punctual, or at least, it will be seen as a single transition from one state to its opposite, the duration of which can be disregarded.

Dahl does not give a definition of ipfv, although one might gather from his discussion that it must be defined in opposition to the pfv. Whether or not a verb form, or rather verbal category, in a specific language is classified by Dahl as an instance of the pfv or the ipfv depends on the type of context used, which Dahl determined by means of a questionnaire. Because different verb forms (categories) from different languages do not behave identically, the pfv and ipfv forms converge on prototypical contexts serving as the basis for their identification. Other authors define the pfv and ipfv differently, using different terminology. For example, Klein (1994: 108) argues that the ipfv indicates that the time for which an assertion is made (topic time) falls entirely within the time of the situation, giving the impression that the situation is seen from within, whereas in the case of the pfv, the time of the situation is interpreted as partly including the time for which the assertion is made, giving rise to the idea of change (Klein 1994: 110). These different definitions lead Klein and Dahl to classify categories from individual languages differently. An example is the simple present in English, which is seen as a pfv form by Klein (1994: 109; cf. Smith’s [1991] analysis); in Dahl’s approach, however, the English simple present does not show enough pfv properties as a category to be classified as a pfv verb form.

Another difficult issue is that languages with a pfv or ipfv marker do not necessarily have a binary pfv∼ipfv structure. In my approach to the study of aspect, this is an important issue since the question what form-meaning elements are available to the language user, and how meanings are correlated with forms depends on the division of labor within the linguistic structure (see also Johanson 2000; Fortuin 2019). This means that languages where a pfv form-meaning element occurs side by side with an ipfv form meaning element, are different from languages where there the pfv is not opposed to an ipfv. Dahl (1985: 69, 71–72) states that in his sample the pfv∼ipfv distinction occurs “in various disguises” in 44 languages out of his total sample of 64. Indo-European languages are clearly overrepresented here (19 out of the 44 languages), while the two other largest groups are Afro-Asiatic languages (6) and Niger-Congo languages (6). In his sample, there is still a clear correlation between Indo-European languages and a binary pfv∼ipfv aspectual system.^[6] Languages with a simple binary pfv∼ipfv system are not common among the languages of the world. The WALS (Dahl and Velupillai 2013) provides data about various languages with a pfv∼ipfv structure (including the ones described by Dahl [1985]), but closer inspection reveals that many of these languages do not have a simple binary structure with a pfv opposed to a general (single) ipfv marker. In many languages, also in Dahl’s sample, the pfv verb is in fact opposed to a habitual verb form or a progressive verb form, or to another verb form, such as a the simple past, which has no clear aspectual pfv or ipfv character. For such languages, one could actually question whether it is useful to speak about the category ipfv at all. An example is Korean, which is classified as a pfv∼ipfv language in WALS (based on Lee 1991: 261–269). However, Korean does not have a simple binary structure since the general ipfv form -nûn is basically a non-past marker (Lee 1991). In the past, we find -ôss-, which is described either as an anterior or past tense form (Lee 1991) or as a pfv marker (Sohn 1995: 26–30). There is no single ipfv past tense, but only a dedicated habitual (kon ha-) and a dedicated progressive marker (-ko is’), which are attached to the past tense (pfv).^[7] In my view, this clearly shows that terms like ipfv or even pfv are not really useful in analyzing Korean. Yet even languages with a general ipfv marker often have more complex aspectual structures. This can be illustrated with Evenki, which is given in WALS as a pfv∼ipfv aspectual language (based on Nedjalkov 1997). When we look more closely at Evenki, however, we observe that it has various habitual aspectual markers, which means that some of the functions that are part of the ipfv (or pfv) in some languages, are part of the habitual forms in Evenki. Additionally, in Evenki the pfv marker (the non-future -rA) only has a pfv function with non-stative verbs and can also be combined with the ipfv marker (and other aspectual markers). Furthermore, the ipfv marker (-d’A) has a rather specialized meaning, is often used to indicate the present tense, and exists alongside a separate durative (continuous) marker (-t). Evenki is not exceptional in this respect; many of the languages classified as pfv∼ipfv in WALS in fact have more complex structures of this kind, or structures without a clear binary character.

In addition to this, there are also important language-specific differences between categories such as pfv or ipfv. Even within Indo-European languages, we can already observe a striking difference between aspect as found in Greek, both Ancient and Modern, where we see a pfv past (sometimes called preterite, aorist, or simple past) as opposed to an ipfv past (also called imperfect), and aspect as found in the Slavic languages. Whereas Slavic aspect in the case of the pfv centers around the attainment of an inherent or imposed boundary (telos, (ad)terminus) of the event, Greek-style aspect centers around the concept of a temporally bounded event, irrespective of whether the event is telic or not. This can be illustrated with the difference between the aorist in Ancient Greek and the past tense forms in Russian. Consider the following Greek example of an aorist provided by Comrie (1976:19):

As Comrie (1976: 19) points out, in Ancient Greek the aorist (prototypically the pfv, which refers to a past event) of the verb reign either refers to a complete reign or has an ingressive interpretation (‘become king’). It differs from the imperfect (ipfv past) in that it does not say anything about events that occurred during this reign. In terms of the theory of Johanson (2000), the aorist signals past and –intra (i.e. opposed to intraterminality, that is, envisaging the event within its limits, intra terminos). This is different from Russian, where sentences similar to (6) could contain an ipfv or pfv past; this depends on whether the events are presented as having a boundary ((ad)terminus) that has been reached and actualized or not (in terms of Johanson [2000], +ad), determined by whether the context is narrative, as in (7) where the (delimitative) pfv prefix po- is added to the verb, or retrospective, as in (8), a consideration that is irrelevant for the Greek aorist:

Thus, even though both the Russian and Ancient Greek instances of the pfv past are examples of the pfv defined as an unanalysable whole, they nevertheless show different behavior, and should be defined differently. This becomes even clearer if we look at more typologically divergent languages. In many languages that do not have a grammaticalized way to express tense (e.g., Berber languages; Mayan languages, such as Tzotzil), the pfv is the standard way to refer to a past event. This type of ‘pfv’ is very different from the pfv in Slavic or Greek, for example. Dahl’s (1985) solution to this is to treat aspectual categories such as pfv as a prototype-like category, defined in terms of a set of typical properties. In this paper, I will build further on Dahl’s insight, although – as explained later – I will also point out the difficulty of this classification.

As such, a comparative approach to verbal aspect may give insight into aspect from a crosslinguistic perspective, but the differences between individual languages should also be taken into account.

2.3 Discussion of the language sample

The association between the pfv or ipfv aspect and habituality (or unbounded repetition) can be determined in a straightforward manner for languages that have a binary pfv∼ipfv structure; however, as I discussed above, this is typologically not very common. As a result, it is not possible to work with a large and genetically diverse sample. This research is therefore confined to a smaller set of languages (36 in total), most of which do indeed have a binary (pfv∼ipfv) aspectual structure.

For the sample, I collected data from languages that have the following properties:

1. Aspectual languages with a clear binary (pfv∼ipfv) system, where prefixation or suffixation associated with telicity (or limitation, (ad)terminativity) is involved in the expression of aspect. These are Slavic languages (Russian, Czech, Slovene, Bulgarian) and two non-Slavic languages (Georgian, Ossetic).

2. Aspectual languages with a clear binary system (pfv∼ipfv), where aspect is only expressed in the past, with two past tense forms (often called ‘aorist’ and ‘imperfect’). These are the Indo-European languages Italian, French, Spanish (Romance), Albanian, Greek, Armenian, and Persian. Bulgarian and Georgian from the first group also belong to this group.

3. Aspectual languages with a clear binary system (pfv∼ipfv) without tense. These are two Semitic languages (MS Arabic, Tarifiyt Berber), two languages from Middle America (Тzotzil, Mixtec), and one language from Africa (Luwo).

4. Non-Indo-European aspectual languages with a clear binary system (pfv and ipfv). In my sample these are two Bantu languages with complex aspectual systems (Xhosa, Zulu) and Totonac from Middle America.

5. Languages without a clear binary pfv∼ipfv system:

   1. Languages with a pfv but without one general ipfv marker (Turkish, Korean, Ewe, Seneca, Mohawk, Hindi);

   2. Languages where aspectual markers can be combined with a habitual marker (Mian, Evenki, Quechua, Totela);

   3. Languages where there is a simple past as opposed to a habitual past (English, Lithuanian);

   4. Languages where aspectual marking is not obligatory (Burmese, Chinese).

I collected the data from the languages in three ways:

1. using existing literature and grammars;

2. consultation of native speakers or language experts, in some cases using a questionnaire as in Dahl (1985);

3. using a corpus of the language or data from the Internet.

See Appendix A for a classification of the languages used in my study, and Appendix B for an overview of how the data were collected per language.

3.1 Languages where perfectivity (and imperfectivity) is related to telicity

Some Slavic languages only allow for the ipfv in the case of unbounded repetition in the past (e.g., Russian), while others allow for both aspects (e.g., Czech, Slovene).^[8] In the latter type of languages, the ipfv by itself can trigger a habitual interpretation, whereas the pfv always needs an expression of unbounded repetition, such as every day, for a habitual reading to occur. The use of both aspects in Czech or Slovene is explained in the literature as the result of the inherent nature of unbounded repetition (see, e.g., Benacchio and Pila 2015; Dickey 2000; Dübbers 2015; Mønnesland 1984; Stunová 1986). In the case of an event that is repeated unboundedly, the event is fully completed a non-definite number of times. This full completion of each individual event, called the ‘microlevel’ of the complex of repeated events, makes it possible to use the pfv, which in Slavic signals full completion of a telic event. Using the ipfv, however, brings the focus to the ‘macrolevel’, where the whole of repeated situations forms a series of events without clear boundaries. This can be illustrated with Slovene (9), where the ipfv portrays the action of opening as something ongoing during a specific timeframe, whereas in (10) the subordinate clause with the pfv describes a single, fully completed instance as characteristic of something the person does every day:

Among the factors that trigger a pfv are (i) the frequency of the iteration as expressed by the adverb (the higher the frequency, the more likely the use of the ipfv), (ii) the extent to which the context focuses on an internal set of subsituations that are all repeated, and (iii) the avoidance of a progressive or conative meaning, which would be implied by the use of the ipfv. Slovene and Czech differ from Russian, where the ipfv is obligatory in past tense contexts of unbounded repetition:

In Russian, the unbounded character of the repetition makes it hard to say where each individual event begins and ends, which results in an ipfv conceptualization (for an analysis, see Dickey [2000], Fortuin and Kamphuis [2015], discussing the East-West theory of Slavic aspect).

In Bulgarian and Macedonian we find a different situation than in the other Slavic languages due to their different aspectual structure, where there are, in addition to ipfv and pfv verb stems, also separate ipfv and pfv past tenses that combine in various degrees with the ipfv and pfv verb stems. The aorist refers to a completed past event and differs from the imperfect (ipfv past) in that it does not say anything about events that occurred during the event (+past, –intr). Put differently, the imperfect signals to the hearer to look for other events that occurred simultaneously, or signals that the event was repeated in an unbounded way. The added meaning of the pfv is that it signals that the boundary ((ad)terminus, telos) of the event is reached or actualized (+ad). In the case of unbounded repetition, the pfv aorist is never used in Bulgarian and Macedonian (Dickey 2000: 74). In Macedonian, the ipfv aorist is lost, but in Bulgarian, the ipfv aorist can be used in contexts of unbounded repetition (with telic and atelic verbs), even though its use is restricted and not accepted by all speakers. Because the aorist indicates a closed event, such sentences imply or require the idea of a timeframe in which the repetition took place. It occurs, primarily in the spoken language of Sofia, with često ‘often’ as in (12b), with the non-prefixed (simplex) verbs igraja ‘play’; compare:

In (12a) with an ipfv imperfect the focus is on the fact that the event regularly occurred, whereas in (12b) with an ipfv aorist the idea of regularity is absent and there may be an implication that George does not play football anymore. Sentences like these are very infrequent and probably only possible (for some speakers) with non-(ad)terminative simplex verbs such as ‘play football’, and with verbs that express actions that people usually do as a habit. Similar sentences are considered less acceptable with ponjakoga ‘sometimes’ and vinagi ‘always’ and fully unacceptable with ‘every day’ (vseki den), even if an explicit timeframe is added, due to the highly cyclic meaning of ‘every day’:

As I will show in Section 3.2, this restriction is in fact reminiscent of two other languages with an aorist∼imperfect opposition in my sample (Greek or Armenian), but different from Romance languages, where there seem to be less restrictions on sentences like (13). In some contexts, it is also possible to use the pfv imperfect in Bulgarian and Macedonian. This form is only used in the case of so-called conditional-temporal constructions. In such cases the pfv is triggered because the construction singles out an individual, fully completed instance that is linked to another individual, completed instance: ‘each time that x, y occurred’, as in the following Bulgarian example:

Both Georgian and Ossetic have an aspectual system to some extent similar to Slavic, where the pfv is inherently linked to telicity. The addition of the prefix to a verb stem means that the verb expresses that the (ad)terminus of the event has been reached (cf. Tomelleri [2010]). In Georgian there is a difference between the imperfect (ipfv past) and the aorist (pfv past). As in Bulgarian and Macedonian, we find a non-prefixed (‘ipfv’) imperfect and a prefixed (‘pfv’) imperfect (the aorist occurs almost exclusively with verbal prefixes). In the case of unbounded repetition, Georgian typically uses a non-prefixed (i.e., ipfv) imperfect. It is also possible, however, to use the prefixed (i.e., pfv) imperfect to indicate that the event (or events) was fully completed (cf. Comrie 1976: 32; Hewitt 1995: 174, 195). This use normally occurs with xolme (‘as a rule’), with or without an adverb of unbounded repetition, as in the example given by Comrie, where the use of the pfv is triggered by the sequential nature of the repeated events, emphasizing that each single subevent is fully completed, before the next one starts:^[12]

The pfv imperfect is also used as a past of the future, and therefore often called ‘conditional’. As such, the habitual use of the pfv imperfect is reminiscent of English constructions with would, or similar uses in Bulgarian, Macedonian and old Croatian (see Fortuin [to appear] for a discussion). The most prototypical pfv form, the (prefixed) aorist, is not used in past contexts of unbounded repetition.^[13]

In Ossetic a non-prefixed past tense can be seen as an ipfv past, whereas a prefixed past tense can be seen as a pfv past. Abaev (1964: 45) notes that to express recurrence or customariness the particle -iu is often used, and provides examples with pfv and ipfv past tense verbs. The particle or clitic -iu (literally ‘once’) can be regarded as a habitual marker. In the case of sentences without an adverb of unbounded repetition (for example aræx ‘often’, xatt ‘sometimes’) and with the clitic -iu, we often find the pfv past tense. This is because the idea of a habit easily presupposes the idea that a telic event was fully completed a non-specific number of times. It is also in full accordance with the original meaning of -iu, namely ‘once’. Cf. the following examples:

In (16), the pfv of the verb ‘listen’ focuses on the idea of completion and success (for example that someone has taken to heart what is said), whereas the use of the non-prefixed (ipfv) verb in (17) presents the action as an ongoing process, not necessarily with the idea of success. In the case of sentences with an expression of unbounded repetition and the past tense, we find various combinations (pfv or ipfv with or without habitual marker), although the pfv with habitual marker and the ipfv without habitual marker seem to be the most common.

3.2 Languages with a pfv∼ipfv system in the past (Spanish, Albanian, Armenian, Greek, Persian)

In this section I discuss Indo-European languages that have a system with a pfv past (often called aorist) and an ipfv past (often called imperfect).^[14] Even though in Romance, habituality is strongly associated with the ipfv aspect, we also find instances with a context of unbounded repetition and the pfv aspect. This can be illustrated with the following sentences from Spanish. Both of them have a habitual character since they report on a characteristic situation that was typical of an extended period of time in the past:

In (18) we find a narrative with only pfv’s, except for one ipfv past. The use of the pfv with ‘each day’ singles out a series of completed events that are presented as typical of how things proceeded during a specific time in the past, although there is no linguistic reference to this (e.g., during his stay there, last summer, etc.). Both the specificity of the events and the timeframe, and the fact that the events are part of a chain, make it possible to present them as closed or total. In (19), with only ipfv’s, the reoccurring, habitual, event is presented as an ongoing general habit that is characteristic of a person during the given, extended period of time. The first two ipfv’s serve as a background for the story and provide general information about a person.

In Armenian the pfv past can also occur in contexts of unbounded repetition, but such uses are more limited than in Spanish. They do not occur with “every day” as in Spanish (18), but are exclusive to the adverb hačax ‘often’ (Donabédian-Demopoulos 2016), for example with the verb ‘remember’:^[15]

In this particular example, the use of the pfv is triggered by the specific context, expressing that on various occasions a given event occurred, followed by another event. The Armenian data are partly reminiscent of Greek and Albanian. In Greek the pfv can occur with expressions of unbounded repetition, such as sychná ‘often’, as illustrated with the earlier example (4), and merikés forés ‘sometimes’, as in the following example:

Both expressions of unbounded repetition presuppose the idea of an occasion or time that occurs either frequently (often; i.e., on various occasions) or with some regularity (sometimes; i.e., on some occasions). The same is in fact true for Albanian, where the pfv (aorist) can be found in contexts with shpesh ‘often’ and ndonjëherë ‘sometimes’. In the languages in my sample, the meaning which is expressed in English by often seems to occur more frequently with the pfv than the meaning expressed in by English sometimes. The explanation for this is that in many languages ‘often’ can be used more easily with respect to a bounded and shorter timeframe than ‘sometimes’, which requires the idea of a stretch of time consisting of various intervals. This explains why Dutch (22a) with vaak ‘often’ is acceptable, whereas (22b) with soms ‘sometimes’ is not; in both sentences the present perfect is used:^[16]

Nevertheless, expressions with ‘sometimes’ as in Greek (21), given that the iteration of the event is relatively low, may also suggest the idea of a specific occasion when the event occurred. In such sentences, the speaker presents one completed (pfv) occurrence of the repeated events as typical of how things proceeded. Such sentences show properties of bounded repetition, specifically the idea of a concrete occurrence of an event, which in the case of bounded repetition is often indicated by expressions referring to times (as in English three times, a couple of times, etc.). The same individuating character may also occur with ‘often’. In these sentences there are additional factors that trigger the pfv. To give an example, in (4) the pfv is triggered by the context where the repeated event (“face antisemitism”) occurs against the background of some other action (“returning to their homelands”), which makes it easier to present it as something complete. The use of the pfv is much more restricted or even absent with ‘every day’ (see, e.g., Newton [1979: 141, 165] and Lindvall [1997] for Greek káthe méra ‘every day’). This expression has a cyclic meaning and therefore easily suggests the idea of something ongoing. Because of this, it triggers the idea of several points in time, which is typical of the ipfv. There are exceptions, however, as in the following Albanian example, where the pfv occurs with çdo ditë ‘every day’:

In Greek the pfv would be chosen here as well. The reason is that the repetition refers to a limited period of time in a narration, and on each day one specific event occurs. If the time span is shorter, and if the events that are repeated are more specific, it is generally easier to name or guess the exact number of iterations (cf. Bertinetto and Lenci 2012). As such, examples such as these conform less to the definition of habituals as describing a situation which is characteristic of an extended period of time, than as something characteristic of a whole period. Interestingly, in Greek the pfv is also possible in some contexts with pánta ‘always’, even though this adverb refers to a continuous time span:

In this case the pfv can be explained by the retrospective character, where the speaker takes stock at a particular moment in the present of a specific period in the past. By using the pfv, the speaker emphasizes that the actions of this period must be seen as closed, as something belonging to the past.^[18] In (25) the ipfv focuses on the ongoing (re-occurring) past action of supporting, which in this example also continues into the present:

It is interesting to compare Greek, Armenian, and Albanian with Persian, because Persian has a similar pfv∼ipfv structure but does not allow the use of the pfv with har ruz ‘every day’, hamiše ‘always’, aqlab ‘often’, ba’zi-vaqth ‘sometimes’, barxi auqat ‘sometimes’. In all these contexts, the ipfv past with mi- is used and the pfv past is excluded (see, e.g., Mahootian 1997: 241; Taleghani 2008: 86, 116).^[20] Only in very specific contexts, the pfv is not fully excluded, as can be illustrated with the following Persian translation from The Old Man and the Sea by Ernest Hemingway (note that this example was judged to be perfectly acceptable by native speakers of Persian):

The pfv in (26) is facilitated by the use of se hafteh ‘for three weeks’. This expression indicates that the episode of catching fish every day occurred during a limited and relatively short timeframe in a narrative, which is part of a chain of events (‘first without fish, and then caught fish’). Such a limited (bounded) and relatively brief ‘for’-timeframe can facilitate the use of the pfv because it indicates the limits for the realization of the re-occurring event as part of the narration. Nevertheless, the use of the pfv in contexts of unbounded repetition is much more restricted than in languages such as Spanish or Greek and does for example not occur with expressions such as ‘last week/year, etc.’:

As I have shown, the presence of a limited (bounded) timeframe is not a prerequisite for using the pfv in my sample of languages. Besides the presence or presupposition of a limited timeframe, there are other factors that may trigger the pfv, such as the presence of a series of repeated events, the type of repetition, and the possibility of conceptualizing the unbounded repetition in terms of a specific instance that was typical of the general repetition in the past.

3.3 Languages with a pfv∼ipfv system without tense

In my sample there are five tenseless languages with a pfv∼ipfv system (Arabic, Tarifiyt Berber, Tzotzil, Luwo and Mixtec). In all these languages, the pfv is closely associated with a past reference. Structures of this kind entail potential competition between the use of the pfv to signal a past reference, and the use of the ipfv to signify that there are no boundaries to the repetition (‘non-totality’).^[21] Table 1 gives an overview of the aspect in contexts of unbounded repetition.

In this group we find different strategies, probably due to the association between past reference and the pfv on the one hand, and unboundedness and the ipfv on the other. In Tarifiyt Berber, habituals are always ipfv, which means that the past reference has to be inferred from the context.^[22] In Arabic, we find a construction consisting of a (perfect(ive)) auxiliary verb kāna ‘was/were’, which explicitly signals past reference, and an ipfv verb (O’Brien 2003: 76), which signals the habitual character of the event. In Luwo, both a pfv construction and an ipfv verb are possible. In Tzotzil, the ipfv is typically used with expressions of unbounded repetition, such as ju-jun k’ak’al ‘every day’. However, with regard to Tzotzil, Vinogradov (ms) also mentions that the pfv can be used in past tense habituals to indicate that the habit only refers to the past; an example is given in (28) with li junabie ‘last year’, although the timeframe can also be a longer period as in (29):

In these sentences, the event is situated in a specific timeframe, which emphasizes the idea that at the moment of speech the series of events has been completed, presenting a fact about the past. This retrospective-like conceptualization differs from instances with an ipfv, where the focus is on the past where the events are repeated unboundedly. The Tzotzil data are partly comparable to the Nuyóo Mixtec data. In this language, habitual situations (Bickford and Marlett 1988: 8) can be indicated by both the pfv and ipfv:

Precisely because of the temporal subordinate clause in (30a), the event of living is presented as something that occurred in a limited timeframe in the past and no longer in the present, which makes it possible to use the pfv. In the case of the ipfv as in (30b) some other time span (such as the moment of speech) coincides with part of the habitual macrosituation, thus “imposing an ‘inside out’ perspective on it” (Bickford and Marlett 1988: 8).

In the non-tensed pfv∼ipfv languages, the pfv can also be chosen if the repetition shares properties of both unbounded and bounded repetition. This is the case in Arabic sentences with tāratan ….wa-ṭura(nan) ‘sometimes X, sometimes Y’, in constructions with kullamaa ‘whenever’ (both in the main clause and subordinate clause; Bahloul [2008: 59]), which all share properties with bounded repetition. Similarly in Tzotzil, depending on the construction, the pfv or ipfv is used in sentences that express ‘whenever’.^[24]

3.4 Non-Indo-European pfv∼ipfv languages (Xhosa, Zulu, Totonac)

In my sample, there are three non-Indo-European languages with a (partly) tensed binary ipfv∼pfv structure, namely the related Bantu languages Xhosa and Zulu (see also WALS and Dahl [1985] for Zulu), and Upper Necaxa Totonac from Mexico.

In Upper Necaxa Totonac, past habituals (with or without expressions of unbounded repetition) always occur with the past tense plus the ipfv aspect marker, and not with the pfv aspect marker, which is used to indicate complete events.^[25] As such, there is a clear relation between past habituality and the ipfv. In Xhosa and Zulu, past habituals are expressed by the remote past plus ipfv aspectual marking (see, e.g., Dahl [1985] for Zulu, and Savić [2017] and [2020] for Xhosa). Savić [2017] suggests that in Xhosa there is a sliding scale between the use of the pfv and ipfv in the case of iteration. In his opinion, the more unspecified the number of iterations, the higher the probability that the verb will take the ipfv aspect. In fact, examples of pfv habituals with either a remote past or a recent past can be found, as illustrated by the following example with yonke imihla ‘every day’ and a perfective remote past:

The use of the pfv is perfectly acceptable, because the speaker has in mind a specific period of time in which the iteration occurs. Xhosa thus shows similarities to other languages with a pfv∼ipfv system.^[27]

3.5 ipfv∼pfv languages without a (clear) binary system