1. Introduction

The distinction between lexical and functional categories is a well-known one in Western linguistics. It is both long-standing (possibly dating back as early as Aristotle)^[1] and widely adopted; although the details may vary, most modern syntactic theories include both notions. The basic idea behind this distinction is that, within a sentence, some morphemes^[2] play more formal or grammatical roles than others. For example, in (1)

'and' has the formal function of conjoining the two nominal phrases 'a bag' and 'a book'. As such, its role is different from both that of 'bag' and of 'book'. In formal semantic terms, the extension of 'bag' and 'book' is a set of entities (type <e, t>), whereas that of 'and' is a more complex function operating on sets (hence, a higher type than <e, t>).

One way in which this distinction is often characterized is in terms of 'meaningfulness', with lexical categories argued to be meaningful and functional categories meaningless or, at least, less meaningful than lexical categories. As often, this is partly a question of terminology; whether or not we think functional categories to be meaningless necessarily depends on how we define 'meaning'. It remains a valid point, however, that, even if we think of functional categories as meaningful, the meaning of 'and' in (1) will still be a different type of meaning from that of 'bag' (we can label it in different ways, e.g. grammatical vs. contentive meaning).

This basic intuition that lies at the heart of the lexical-functional distinction is well-established in Western linguistics. Issues arise when we move beyond theory and attempt to systematically apply the lexical-functional classification to real languages, in terms of how competing criteria should be applied to yield a consistent and meaningful classification. The literature on the topic suggests a range of classification criteria, including the types of morphological and syntactic processes that a morpheme can undergo (Rizzi and Cinque 2016) or the number of elements belonging to a given class of morphemes, that is whether the class is closed (i.e. consists of a quantifiable number of elements) or open (i.e. its members cannot be finitely enumerated). For some linguists, there is a complete overlap between lexical and open classes, functional and closed ones, whereas for others, other criteria bear more weight.

One of the greatest challenges comes from those categories of words that appear to share characteristics with both lexical and functional categories as traditionally defined and do not fall neatly into either. In English, this is the case with prepositions (Carlson 1983: 94 n.2). In the literature, prepositions are variously classified as lexical (Chomsky 1970) or functional (Carnie 2013); Abney (1987), instead, argues that they are unspecified for lexical-functional status. In the light of this and similar problems, perhaps a better approach is to think of the lexical-functional distinction as a gradient rather than a strict dichotomy. If lexical and functional are two extremes of a continuum, prepositions may lie somewhere in between the two.^[3]

2. The Sanskrit Grammatical Tradition

The Western concepts of lexical and functional categories are a question of both syntax and semantics. Although they are often thought of primarily in relation to meaning, most of the criteria that allow us to identify and define them are syntactic. In the Sanskrit grammatical tradition, there is no exact equivalent of the Western lexical-functional distinction. Most, if not all, the words that in Western terms would be classified as functional, in the Aṣṭādhyāyī belong to the group of avyayas, 'indeclinables' (see below).^[4] Because of how the Aṣṭādhyāyī works, certain notions cut across avyayas and other categories that in Western terms would count as lexical. Crucially, for the purposes of derivation, avyayas classify as nominal stems.^[5]

In practical terms, this means that an adposition like anu in (2) is derived in the same way as the noun nadīm.

To derive anu, we follow Aṣṭ. 4.1.2 and add an inflectional ending. Unlike for nadīm, however, we then need to delete the ending on anu, because it is an avyaya (Aṣṭ. 2.4.82). Although this adding and then deleting of inflectional endings complicates the derivation of avyayas, it simplifies the grammar at a higher level by, for example, streamlining the definition of the important category pada 'word'. The parallelism between avyayas and other categories thus serves the overall aims of concision and economy of the Aṣṭādhyāyī.

Besides their derivation, the Aṣṭādhyāyī discusses the functions of various subgroups of avyayas (e.g. the grammatical operations that they can trigger, such as retroflexion or particular accentual patterns; their relation to the verb in a sentence etc.), as well as the meanings of individual avyayas. These are essential to the complex classification of indeclinables that the Aṣṭādhyāyī develops. Within vyākaraṇa, however, it is the later grammarians (at least as early as Bhaṭṭojidīkṣita; see section 3) who add yet another dimension to the discussion of the semantics of indeclinables, distinguishing between two ways in which a word can bring out meaning (denotation and manifestation; see below) and debating how avyayas as a group fit into this picture by comparison with other types of words.

The question at issue for these grammarians is whether avyayas have a 'meaning' of their own and how they interact with the meaning of the words to which they are associated in the sentence. What grammarians mean when they speak of 'meaning' is quite specific to the particular semantic theorizing of vyākaraṇa. In Western linguistics (1 above), the key distinction underlying the lexical-functional divide lies in the type of information that a word (or morpheme) provides, lexical for lexical categories and grammatical for functional ones. Depending on one's notion of meaning, one may consider those words that convey grammatical information to be less 'meaningful', but there is generally no doubt that every word directly conveys at least some type of information.

Within the discussion of avyayas, 'meaning' is analyzed both in terms of what a word signifies and how that signification is expressed. In grammatical texts, the most general term for 'meaning' is artha, a highly polysemous word, whose uses range from a generic, unqualified sense of 'meaning' (covering more than one type of semantic concept, see below) to more specific types of meaning. Sometimes artha even stands for the referent of a word, that is, the object corresponding to that word in the extralinguistic world.^[6]

In the discussion of the semantics of indeclinables, the use of artha is closely related to two fundamental aspects of the vyākaraṇa theory of meaning: the notions of śakti and lakṣaṇā (the powers conveying the primary and secondary meanings of a word), and the distinction between vācaka ('denotative') and dyotaka ('manifesting') words.

In the tradition, śakti and lakṣaṇā are two powers of signification whereby a word expresses different types of meanings.^[7] śakti is the 'denotative power' determining the essential or primary meaning (śakya meaning) of a word. lakṣaṇā, instead, is the power whereby a word expresses any 'inferential' meanings (lakṣya meanings) that it may additionally have; these secondary meanings directly depend on the essential meaning and are based on convention. The standard example is gaṅgā. The primary meaning of this word is 'Ganges'; hence, gaṅgā has a śakti or denotative power whereby it expresses the meaning 'Ganges'. However, we can conventionally use gaṅgā to mean 'the bank of the Ganges', as in the phrase gaṅgāyāṃ ghoṣaḥ 'the village is on the bank of the Ganges'. This is a secondary, lakṣya, meaning and gaṅgā expresses it by lakṣaṇā. Despite the relative clarity of this example, the distribution of the meanings of words between śakti and lakṣaṇā is not always easy to determine and it appears to have been a point of divergence among grammarians themselves.

This notion of powers whereby a word expresses its own meanings interrelates with another fundamental distinction: that between denotation and manifestation. These are two modes in which a word can bring out meaning. Denotation is the mode by which a word brings out its own primary and secondary meaning(s). Thus, gaṅgā is said to be vācaka 'denotative' of both its primary meaning 'Ganges' and of its secondary meaning 'the banks of the Ganges'. These two meanings are in turn said to be vācya (literally 'to be denoted') meanings of gaṅgā.

However, there are words that have neither a primary nor, consequently, a secondary meaning, and which therefore have neither śakti nor lakṣaṇā. Such words are, therefore, never vācaka 'denotative'. But besides denotation, there is another way in which a word can bring out a meaning; this is manifestation. When a word indirectly expresses a meaning that is directly expressed (denoted) by another word, it is said to be dyotaka, 'manifesting', of that meaning. In the case of manifestation, the meaning being manifested must necessarily be a vācya meaning of some other word and this meaning is somehow 'activated' by the dyotaka word. The meaning that is being expressed is accordingly termed dyotya, 'what is to be manifested' in relation to the manifesting word.

This whole discussion plays into the grammarians' conceptualization of the meaning of avyayas, the main questions being whether avyayas are vācaka, because they have a meaning of their own, or whether they have no meaning but are still dyotaka of the meaning of other words in the sentence. In (5), for instance,

the problem is how to analyze the semantics of the preverb anu in relation to that of the verb bhavati and of the compound verb^[8] anubhavati as a whole. On its own, the root bhū- means 'become' and it is only when it is combined with anu that we get the meaning of 'experience'. Grammarians thus consider three options; whether the meaning of 'experience' is fundamentally expressed by: 1. the simple root bhū-, 2. the preverb anu or 3. a combination of the two. Kauṇḍabhaṭṭa, whose Vaiyākaraṇabhūṣaṇasāra contains one of the most detailed extant discussions of the semantics of avyayas, argues strongly in favour of the first option. According to him, the meaning of 'experience' is a secondary, lakṣya, meaning, of the verbal root bhū- and bhū- expresses this meaning by lakṣaṇā.^[9] Thus bhū- is 'denotative' (vācaka) of its own secondary meaning. On the contrary, anu does not have a meaning of its own but is said to 'manifest' the meaning of 'experience' when it combines with bhū-; that is, it is dyotaka of it.

anu and bhū- thus play different roles in relation to the meaning of 'experience' and, likewise, 'experience' counts as a different type of meaning for the preverb and the verbal root. From the perspective of bhū- itself, 'experience' is a vācya meaning; from that of anu, it is a dyotya meaning.^[10]

To go back to the concept of artha and its uses, grammarians quite ambiguously employ artha both generically in reference to any type of meaning, whether vācya or dyotya, or more specifically in relation to vācya meanings only. When they use artha in the laxer, more general sense, a preverb like anu can be said to be arthavat, 'having artha', because it expresses a dyotya meaning.

As seen (section 1 above), in the Western lexical-functional distinction, functional categories are sometimes thought of as being less meaningful than lexical ones. In this case, the 'meaningfulness' of a word depends on the type of information that this conveys. In the Sanskrit tradition, what matters is the relation that a word has to a specific meaning. The same type of information can count as a vācya or a dyotya meaning depending on the word from whose perspective it is analyzed. The 'meaningfulness' of a word then depends on whether it expresses information by denotation or manifestation. The distinction thus lies not so much in the type of information expressed by a word but in the way that that word can express meaning, whether its own or that of another word. Of the two relations to meaning, that of a dyotaka word is always less direct than that of a vācaka word because it depends on the latter; that is to say, a dyotaka element can only be dyotaka in relation to a meaning that is vācya by some other element.

3. The prādi list

In the Aṣṭādhyāyī, avyayas are organized according to a complex classification system based on such criteria as the grammatical operations that they can trigger. Table (6) illustrates the different types of avyayas identified and the main Pāṇinian sūtras where they are defined.

(6)

When grammarians discuss the semantics of avyayas, they mostly focus on the group of the nipātas, which further divide into the prādi and cādi lists (as well as some non-prādi gatis). Besides ca 'and', the cādi list includes other extremely common indeclinables, such as vā 'or', iva 'like', cet 'if' etc.^[11] In the Aṣṭādhyāyī, members of the prādi list receive different designations depending on their function within the sentence, thus dividing into four groups in total: preverbs (upasargas), such as anu in (5), those indeclinables that are morphologically identical to preverbs but function as independent adverbs or adpositions (karmapravacanīyas), such as anu in (2), and prādis when they function neither as preverbs nor as adpositions. All prādis that are upasargas also get the name gati, which is responsible for different operations.

The distinction between the prādi and cādi lists is central to the discussion of the semantics of nipātas. A question which grammarians consider is whether all nipātas should be analyzed like anu in (5) or whether there is a difference between the prādi and cādi lists. In relation to this question, as elsewhere in his commentary, Kauṇḍabhaṭṭa sets the view of the logicians, Naiyāyikas, against that of the grammarians. The logicians' stance on the matter is clear; they consider all indeclinables belonging to the prādi list to be dyotaka and all those belonging to the cādi list to be vācaka. The grammarians, instead, and Kauṇḍabhaṭṭa with them, argue that all nipātas, cādis included, are dyotaka.

In his refutation of the logicians, Kauṇḍabhaṭṭa quotes the grammarian Bhaṭṭojidīkṣita (c. 1600 AD) as an authority for the view that both prādis and cādis are dyotaka. This view, if we trust Kauṇḍabhaṭṭa's reading of Bhaṭṭojidīkṣita, is likely to have been the standard opinion among grammarians since at least the latter's time. The views of grammarians earlier than Bhaṭṭojidīkṣita are more difficult to reconstruct and, as always, one needs to be careful not to read back notions that quite clearly belong to the later tradition but for whose beginnings we have less evidence.

Patañjali (c. 150 BC) is a case in point. In general, he does not seem to be particularly interested in the question of denotation vs. manifestation. In the whole of the Mahābhāṣya, the terminology associated with the two concepts only occurs very limitedly; there are just a handful of instances of dyotya and none of dyotaka. More importantly, Patañjali seems never to openly engage with the issue.^[12] Although later grammarians occasionally quote passages from the Mahābhāṣya when arguing that nipātas are dyotaka, one may suspect that for Patañjali, the distinction between denotation and manifestation was not as strongly defined or as central as in later grammarians.

Bhartṛhari (c. 450 AD), instead, seems to have considered some nipātas to be vācaka (Vākyapadīya 2.192).^[13] Among the prādis, Bhartṛhari's view of preverbs seems to have been that they are dyotaka, and thus to be in line with that of the later tradition.^[14] While, then, we can know with some certainty that from Bhaṭṭojidīkṣita onwards, grammarians considered all prādis and cādis to be dyotaka, it is harder to know how the discussion had developed earlier on. After Bhaṭṭojidīkṣita and Kauṇḍabhaṭṭa, Nāgeśabhaṭṭa (c. 1700 AD) offers an extensive and elaborate discussion expanding on those of his predecessors in support of the grammarians' position.

4. The cādi list

If we trust Kauṇḍabhaṭṭa, while both grammarians and logicians agreed that prādis are dyotaka, in his day the semantics of cādis were a matter of controversy. Grammarians argued that they are dyotaka, whereas logicians that they are vācaka. As regards earlier grammarians, it is again difficult to know what Patañjali's position was. He discusses the meanings of ca at length (see 'Coordination') and from the perspective of later grammarians, we should think of these meanings as meanings by manifestation (dyotya). Patañjali himself, however, does not openly touch upon the question of whether ca is vācaka or dyotaka. Instead, as regards Bhartṛhari, although the evidence is rather limited, it seems that he may have agreed with later grammarians, considering cādis to be dyotaka.^[15]

As mentioned (see section 3), besides ca, cādis include well-known indeclinables such as vā 'or', iva 'like' and cet 'if'. The list is quite disparate and also contains words that look quite different from ca or other conjunctions, such as śaśvat 'perpetually' and yugapad 'at the same time'.

When discussing the semantics of cādis, the grammarians' argument parallels that for prādis. Consider the following example.

According to Kauṇḍabhaṭṭa's argument, the indeclinable iva has no meaning of its own. In (7), sādṛśya 'likeness' is a part of the latent (lakṣya) meaning of caitra and iva is simply dyotaka of this secondary meaning; i.e. it brings it out by manifestation. This is to say that caitra, whose primary meaning is (the PN) 'Caitra', can also secondarily mean 'like Caitra'. Unlike for prādis, Kauṇḍabhaṭṭa does not discuss the semantics of cādis in detail, but it can be plausibly supposed that the same argument always holds true, i.e. that every noun has a secondary meaning that is characterized by sādṛśya. The word for 'man' thus also secondarily means 'like a man', that for 'horse' 'like a horse' etc. The indeclinable iva selects these secondary meanings, exactly as preverbs select the secondary meanings of the verbal roots that they are joined with. The exact same reasoning applies to the other cādis.

5. Why manifestation?

This idea that such meanings as coordination or 'likeness' (7) are in a way inherent to every word and that ca and iva merely bring them out may appear rather surprising. Why not just suppose that iva means 'like' rather than thinking that 'likeness' is a property of the secondary meaning of all nouns? Kauṇḍabhaṭṭa offers specific reasons why cādis are dyotaka, but the main argument seems to be analogy with the prādis. It is an important point in the Vaiyākaraṇabhūṣaṇasāra that all nipātas should receive the same treatment and, given that Kauṇḍabhaṭṭa believes prādis to be dyotaka, this conclusion should be extended to cādis. In fact, prādis always remain the focus of Kauṇḍabhaṭṭa's analysis.

When discussing anubhavati in example (5), we saw that Kauṇḍabhaṭṭa listed three possible ways of analyzing this compound verb before opting for the view that preverbs are dyotaka and that the meaning of a compound verb is, properly speaking, a meaning of the root only. The two other options were to take 'experience' as a meaning of the preverb anu (i.e. to make the preverb vācaka) or to take it as the meaning of anubhavati as a complex whole.

Kauṇḍabhaṭṭa refutes the former option at length and the latter with the brief comment gauravāt, 'because it would be cumbersome'. In fact, thinking of a compound verb as a complex may not necessarily seem like an unreasonable approach. Outside of the Indian tradition, there are examples of 'compositional' approaches to the semantics of compound verbs, whereby the meaning of a compound verb is interpreted as the combination of the meaning of the preverb and that of the root. This is the approach adopted, for instance, by Hettrich and his team in their studies on preverbs and local particles in Sanskrit.^[16] The preverb itself is analyzed as carrying some type of basic meaning(s) contributing to the meaning of the compound verb.

On some level, Kauṇḍabhaṭṭa too thinks of a compound verb as a unit of preverb and verbal root, but for him, its meaning underlyingly belongs to the verbal root only. The reason for this choice is tightly connected to the general architecture of the grammar, as well as to specific technical issues. Both types of problems are encapsulated by the comment gauravāt. When grammarians speak of gaurava, they generally refer to a lack of economy of the relevant argument. In the present case, as commentators explain, the problems are that the Dhātupāṭha, the list of Sanskrit verbal roots, would need to be modified (because the combinations of preverbs and roots would now need separate entries from those of the simple roots and the number of entries would consequently be expanded by countless new roots) and that the processes of augmentation and reduplication would yield incorrect results.

In vyākaraṇa, derivation is thought of as starting from meaning; that is, the derivation starts in order to express something that the speaker wants to say. The necessary grammatical forms and operations are chosen on the basis of the meanings that need to be expressed. Let us suppose that the meaning that needs to be expressed is that of 'experience'. If 'experience' is the meaning of the compound verb as a whole, rather than just of one of its parts, we are forced to take anubhū- as a root separate from bhū- and, consequently, as the starting point for our derivation. But this would be highly uneconomical for the grammarians because it would require the creation and addition of countless new roots to the Dhātupāṭha.

Starting from anubhū- would also produce ungrammatical results. The main concern is the position of the augment (āgama) and reduplication in compound verbs. Because of how derivation works in Pāṇinian grammar, both augmentation and reduplication are operations that should take place before the root is joined with the preverb. An augment is always prefixed to a root (dhātu), e.g. abhavat '(s)he was' from the root bhū-. In a compound verb, the augment needs to come in between the preverb and the stem, as in the imperfect anv-a-bhavat '(s)he experienced'. To correctly derive this form, anu- can only be added once we have already obtained abhavat. If we started from the compound form, anubhū-, we would derive the ungrammatical form *ānubhavat. Similarly, the root first needs to be reduplicated before the preverb can be attached or the preverb itself would end up being reduplicated.

In the light of these two fundamental problems of economy and correct grammatical derivation, a better solution is inevitably one that does not force us to analyze anubhū- as a root. By ascribing the meaning of 'experience' to bhū- alone, grammarians do not need to take the compound verb as their starting point and can thus derive the correct forms.

Compound verbs are not the only case where grammarians avoid a compositional approach to semantics; one common approach to the semantics of compounds (samāsas) is to regard a compound as an independent unit whose meaning is underived, rather than resulting from the simple combination of the meanings of its constituent members (see nityapakṣa under 'Compounding'). Although the semantic analysis of samāsas is distinct from that of compound verbs, here too we have a case in which the whole is not merely the combination of its parts.

6. Other dyotakas

nipātas are not the only words to be defined as dyotaka by the vyākaraṇa tradition. Although discussions on manifestation vs. denotation tend to focus on nipātas, there are also e.g. pratyayas 'suffixes' that, in certain contexts, count as dyotaka. For instance, this is the case with the absolutive-forming suffix -Ktvā that we find in e.g. gatvā 'having gone'. In general, the agent of an absolutive needs to be coreferential with that of the governing verb (see 'Control'). This condition is called samānakartṛkatvam 'the state of having the same kartṛ'. In the Ktvādyarthanirṇaya of Kauṇḍabhaṭṭa's Vaiyākaraṇabhūṣaṇasāra, -Ktvā is said to be dyotaka of samānakartṛkatvam.

^[1] Carlson (1983).

^[2] It is better to speak of 'morphemes' rather than 'words' because this generally allows for a more comprehensive description (Rizzi and Cinque 2016: 140). That said, both 'word' and 'morpheme' remain somewhat problematic notions in linguistics.

^[3] Sometimes there may even be differences between the members of the same category. Cf. the difference between semantic and grammatical prepositions, as theorized within Lexical-Functional Grammar (Börjars 2019: 55-8).

^[4] If we speak of functional morphemes rather than full independent words, then there is also a great number of functional morphemes that belong to another group, that of pratyayas 'suffixes'.

^[5] Aṣṭ. 1.2.45 arthavad adhātur apratyayaḥ prātipadikam 'A nominal stem is any meaningful part of speech that is neither a verbal root nor a suffix nor a form that ends with a suffix (except those listed in the subsequent rule)'.

^[6] This 'realistic' (Deshpande 1972: 6ff.) sense of artha is occasionally used by Patañjali (c. 150 BC). In the later grammarians, however, artha generally refers to conceptual entities only.

^[7] The tradition also recognizes two other important concepts that play a role in discussions about 'meaning' (vyañjanā and tātparya), for which see 'Linguistic Fields: Pragmatics'.

^[8] This is how Western linguistics traditionally terms verbs when they are combined with preverbs, as opposed to 'simple' or 'uncompounded' verbs. In the Indian tradition, instead, these would not technically count as 'compounds'.

^[9] The use that Kauṇḍabhaṭṭa makes of the term lakṣaṇā in this context is not shared by all grammarians. Most grammarians acknowledge that the example of gaṅgā above involves lakṣaṇā. But some, unlike Kauṇḍabhaṭṭa, would talk of 'experience' in (5) as another primary meaning of bhū-, that is, another meaning expressed by śakti. Even so, they would still agree with the rest of Kauṇḍabhaṭṭa's analysis and take anu to be dyotaka (see immediately below).

^[10] Vaiyākaraṇabhūṣaṇasāra, Nipātārthanirṇaya 323-4.

^[11] See Dyen (1939) for the complete list.

^[12] Bandyopadhyay (1981: 50-1) argues that Patañjali regarded preverbs as dyotaka but, unlike later grammarians, also accepted that some nipātas could be vācaka. Bandyopadhyay's evidence, however, is at best slim. His interpretation of Mahābhāṣya 1.30.1-32.11, for instance, is based on an unsupported interpretation of artha as vācya artha.

^[13] nipātā dyotakāḥ kecit pṛthagarthābhidhāyinaḥ | āgamā iva ke 'pi syuḥ sambhūyārthasya vācakāḥ ||

'Some nipātas are dyotaka; others are independently meaningful (arthābhidhāyinaḥ); some, exactly like grammatical augments, express a meaning by denotation (vācakāḥ) when they are combined with another word'.

^[14] The discussion is at Vākyapadīya 2.187-91. On the notion of anumāna(s) that emerges in this passage, see Akamatsu (1999).

^[15] cādayo na prayujyante padatve sati kevalāḥ | pratyayo vācakatve 'pi kevalo na prayujyate || (Vākyapadīya 2.194) 'ca etc. are not employed on their own, even if they count as words (padatve), (exactly as) suffixes (pratyayas) are not employed on their own, even if they are denotative (vācakatve)'. See Panchal and Kulkarni (2019: 65). In this passage, denotation (vācakatve) seems to be referring to pratyayas only, especially given that the two halves of this verse are apparently meant to be parallel (with padatve referring to cādis and vācakatve to pratyayas). If cādis are not vācaka, then it follows that they are dyotaka.

^[16] For instance, in Casaretto and Schneider (2015: 225), a preverb is defined as a morpheme whose function is 'to modify the verb semantically. Both constituents may retain their original meanings (cf. Lat. ex-ire 'go out'), or they may be subject to lexicalization'. Generally, such an approach at least implicitly underlies diachronic studies on Sanskrit preverbs. The standard account of the development of preverbs in Sanskrit (as in other Indo-European languages) is that they started out as independent words signalling location or changes of location (local particles), before becoming tied to the verb. Under such premises, a diachronic study of preverbs will most probably involve some notion of compositionality in order to describe the progressive lexicalization of two separate words as a single one. Indian grammarians, however, are thinking of the language synchronically and do not discuss anything like the development from local particles to preverbs that we find in IE studies.