2.1 Main Typological Features and Parts of Speech

Indo-European languages of the most archaic type (best represented by Ancient Greek and the two Old Indo-Iranian languages Old Indic—in particular the Vedic variety—and Avestan) have rich fusional morphologies with predominant use of suffixation and ablaut (i.e., alternation involving vowels or vowel-sonorant sequences) as formal devices in both inflection and derivation.

In the archaic IE morphological type lexical bases (“roots”) do not have to be associated with a particular part of speech (cf. also Alfieri, 2016). Therefore, two suffixless ‘base + ending’ structures containing the same base can belong to different word classes: cf. OI vac-mi ‘speak-1sg’, verb, vs. vāc-am ‘speech-acc’ (on the a/ā alternation—‘ablaut’—see section 2.3). According to the traditional description, such pairs are not derived synchronically the one from the other but represent co-derivatives of the same abstract lexical unit vac-.

In the Proto-Indo-European system the border between inflection and derivation must have been fairly blurred (Watkins, 1969, p. 19; cf. also Belardi, 1990; Kastovsky, 2005). Vedic verbal morphology still testifies such a condition. Thus, for example, the base bh(a)r- ‘carry’ appears in the two imperfective (‘present’) forms bhárati and bíbharti ‘carry.ipfv.prs.3sg’—with the latter indicating pluractionality—in the aorist abhārṣīt ‘carry.pfv.pst.3sg’, and in the perfect babhāra ‘carry.prf.3sg’. While bhárati and bíbharti seem to be interpretable as co-derivatives (i.e., as distinct lexemes), it is not obvious whether the aorist abhārṣīt and the perfect babhāra should be seen as belonging only to bhárati or also to bíbharti or to none of them. In fact, the Old Indic present/perfect/aorist opposition, though normally classified as an inflectional one (namely, differentiating aspects), displays features not typical of inflection (for the hypothesis—related to this issue—of a pre-PIE change from a former system based on actional categories to an aspect system, see Lazzeroni, 1977, 1984; Rix, 1986; Strunk, 1994). This is true, in particular, with respect to lack of bi-uniqueness and uniformity (cf. Dressler, 1989), since the different aspect categories have mutually independent systems of inflection classes. Thus, the type of present stem or stems found for a base does not permit inferring which types of aorist stems are formed from the same base (cf. Stump, 2006, p. 302). Significantly, in the lexicographic tradition of Sanskrit (i.e., classical Old Indic), bhárati, bíbharti, abhārṣīt, and babhāra are entered in a flat list under the root bh(a)r- used as a lemma.

On the other hand, in the archaic IE morphological type, a high productivity of derivation and composition processes must have allowed of a relatively non-marginal use of compounds and derivatives not stored in the memory of the speaker (cf. Belardi, 1990, pp. 167f.)—a condition which is relatively rarer in the derived lexicon of a modern European language such as English (cf. Dressler, 1989, p. 8). Thus, from an Old Indic verbal root speakers were able to automatically derive both a causative of desiderative (‘to cause to desire X’) and a desiderative of causative (‘to desire to cause X’). Such derivational structures correspond to syntactically complex expressions in morphologically poorer languages.

In addition to nouns, pronouns, and verbs, the archaic part-of-speech system comprises a class of adjectives, which—except for inflecting for gender—is weakly distinguished from that of nouns (adjectives largely agree with the noun inflectional patterns) and a class of adverbs.

In Vedic, the qualifying adverbial function is typically realized by (neuter singular) case forms of adjectives (especially the accusative), rather than by autonomous deadjectival adverbs. As for closed word classes, the ancient type lacks articles and it is thought by many that Proto-Indo-European also lacked true adpositions, a condition testified by Vedic, where true adpositions are marginal if not absent (cf. section 3.4). A syntactic organization characterized by a low degree of configurationality complements this part-of-speech system (Luraghi, 2010; cf. Meillet, 1937, pp. 440–441; for a different view, see Keydana, 2018).

2.2 Word Structure

The commonest non-compound word structure in the ancient type is

where E is a cumulative marker of more than one inflectional category, and R and, in some cases, S contain an ablaut alternant (see the Old Indic forms in (1)–(7)). Roots (i.e., non-analyzable lexical bases) are typically monosyllabic (allowing complex onsets and codas, and typically showing a rising-falling sonority contour), whereas other morphs usually contain zero or one syllabic nucleus.

Vedic Old Indic

(1)

(2)

(3)

(4)

(5)

(6)

(7)

2.3 Ablaut

The original (Proto-Indo-European) primary ablaut did not involve consonants and had three major alternants (or “grades”): e ~ o ~ zero: e.g., Greek leípō ‘leave.prs.1sg’, loipós ‘left.m.nom’ (adjective), élipon ‘leave.pfv.pst.1sg’. In the OI examples in (1) through (7), -tār- corresponds to an original o-grade with ā in the place of PIE *o; han- and bhar- are original e-grades with a < *e; ha- is a zero-grade with a < syllabic *n; and bhr̥- a zero-grade with syllabic r. In fact, phonological developments that took place in the single branches often involved a following sonorant in the alternation and split the alternant set into different, lexically selected, series. Thus, *ei̯ ~ oi̯ ~ i > Old English ī ~ ā ~ i in bītan/bāt/biton ‘to bite/he bit/they bit’; *en ~ on ~ n̩ > OE in ~ an ~ un in bindan/band/bundon ‘to bind/he bound/they bound’.

It is commonly hypothesized, following Saussure (1878), that an analogous split had already occurred in PIE, giving rise to the special ablaut of the so-called heavy bases, by which a long vowel appears in lieu of the e and o alternants, and a reduced vowel, PIE *ə (reflected in Indo-Iranian as i and elsewhere generally as a), in lieu of zero: e.g., Latin dōnum ‘gift’ < *dō-no-m, datus ‘given’ < *də-to-s. In such alternations, the long vowels are usually viewed as continuations of -VH- sequences—where H stands for the so-called laryngeal, a reconstructed segment difficult to define phonemically—whereas *ə is usually considered an original epenthetic vowel intervening between a laryngeal and an immediately following consonant: i.e., *dō-no-m < *doH-no-m, *də-to-s < *dHə-to-s. According to a widespread trilaryngealist theory, PIE had three distinct “laryngeal” protophonemes, differing from each other in the way in which they altered the quality of an adjecent e-vowel (i.e. the non-coloring H₁, the a-coloring H₂ and the o-coloring H₃).

While IE ablaut alternations are usually described in terms of morphophonological processes applying to a basic allomorph (e.g., OE bind- → bund-), the reconstructed PIE e-grade and o-grade alternants (e.g., *bhendh-, *bhondh-) can be analyzed as formed by a discontinous lexical morph (*bh.ndh-) plus a vocalic infix, since in PIE the phonemes e and o typically did not contribute to distinguishing between different lexical items—possibly except for their linear position in CeRC ~ CReC root pairs—and carried purely morphological information (Belardi, 1990). This fact points to a PIE typology quite similar to that of the conservative Semitic languages (on which, see Shimron, 2003).

In the archaic morphological type, the selection of the ablaut alternant was determined either by derivation processes—see (1)–(2) and (4)–(5), where agent noun derivation selects bhar-, han-, whereas adjective derivation selects br-, ha- –, or by morphological indexing referring to particular sets of cells within the inflectional paradigm—see the -tār- ~ -tr- and han- ~ ha- alternations in (1) and (3) and (6) and (7), respectively.

2.4 Affixation and Reduplication

The archaic morphological type uses suffixes and prefixing reduplicants; exceptions are a nasal infix characterizing some present stems (and probably originating as a metathesized allomorph of a nasal suffix)—e.g., OI yuj- ‘join’ → yu‹ná›k-ti ‘(he) joins’ (cf. Lat iungō ‘I join’ vs. iugum ‘yoke’)—a prefix e- serving as a past tense marker (the so-called augment, which in fact is found only in Greek, Armenian, Phrygian, and—appearing as a-—in Indo-Iranian), and a few adverbial prefixes forming adjectives (Grk a(n)-, OI a(n)-, Goth un-, Lat in- ‘not’; Grk eu-, OI su-, Avestan hu- ‘well, good’).

As shown in particular by OI, Greek, and Gothic, IE reduplicants typically exhibit a CV prosodic template and can have a partially fixed segmentism (e.g., Ce-, Ci-). They are attached to verbal bases and characterize either stems associated with tense-aspect categories or derivative verbs such as desideratives. Thus, in Greek and OI a CV- reduplicant regularly appears in perfect stems (e.g., Grk pé~poith-a ‘I trust’) and is also found with several lexemes in the imperfective (“present”) stem (e.g., Grk dí~dō-mi ‘I give’). With vowel-initial bases, reduplication is generally substituted by initial lengthening in both OI and Greek, but, in some cases, less common reduplication types are also found, such as VC- reduplication in some Greek perfects like ódōda ‘I smell’ (base od-—cf. Sihler, 1995, p. 489) or infixing reduplication, e.g., in OI á‹śi›ś-iṣ-a-ti ‘he wishes to eat’, desiderative derivative of aś- ‘eat’ (cf. McCarthy & Prince, 1996, pp. 14f.).

2.5 Thematicity and Accentual Mobility

A distinction cross-cutting the noun-verb division is traditionally posited between athematic and thematic inflection types, with the latter being characterized by the presence of a “thematic vowel” (Grk e alternating with o, OI a or ā) immediately before the ending, e.g.:

sg.nom: them. Grk lúk-o-s, OI vŕ̥k-a-ḥ ‘wolf’, vs. athem. Grk ophrû-s, OI bhrū́-ḥ ‘eyebrow’;

2pl: them. Grk. phér-e-te, OI bhár-a-tha ‘you carry’ vs. athem. Grk deík-nu-te ‘you show’, OI aś-nu-thá ‘you attain’.

In some paradigm cells, the thematic versus athematic opposition behaves as an inflection class distinction and determines the selection of ending allomorphs:

prs.1sg: them. Grk phér-ō ‘I carry’ vs. athem. Grk deíknū-mi ‘I show’;

pl.ins: them. OI vŕ̥k-ais ‘by the wolves’ vs. athem. pitŕ̥-bhiḥ ‘by the (fore)fathers’.

Zero endings mark the 2sg imperative in thematic verbs and, in most cases, the vocative singular of both thematic and athematic nouns.

In the diachrony of IE languages, athematic inflections with obstruent-final stems tend to be eliminated, thus preventing clusters prone to assimilate from appearing at the stem-ending boundary: e.g., Lat nepotibus, Lith nepuotims ‘grandson.pl.dat’ (from an original stem *nepōt-) with secondary -i- according to the i-stem inflection.

Relevant phenomena regarding accent distribution are shown by Vedic Indic, which, unlike Classical Sanskrit, preserves the PIE contrastive stress: both nominal and verbal athematic inflection can show accentual mobility—possibly accompanied by ablaut alternations—across the paradigm, with stressedness mostly alternating between predesinential and desinential syllable (cf. OI hán-ti: ha-tháḥ; han-tā́r-am : han-tr-é in (6), (7), (1), (3); see also Kiparsky, 2010).

2.6 Nominal Grammatical Categories

The richest inflectional system is found in Old Indic and Old Iranian, where nouns are inflected for three numbers (singular, plural, dual) and eight cases: nominative, vocative (typically lacking in pronouns), accusative (also indicating goal of motion and spatial and temporal extension), genitive, dative, instrumental, locative, and ablative. Nouns belong to one of the three gender classes: masculine, neuter, and feminine. Gender assignment is usually semantically determined for human nouns. In the ancient languages some other tendential semantic patterns are found: e.g., neuter for fruits, feminine for trees, masculine for winds. Moreover, neuter is the gender of adjectives converted into non-human nouns referring to abstract or generic entities (e.g., Lat bonum ‘good thing’) and is used for agreement with non-nominal controllers. Thus, neuter gender is selected by pronouns referring to states of affairs or propositions, whence the fact that neuter pronouns are recurrently diachronic sources for complementizers (e.g., OI yád, Lat quod, Eng that). Formally, all neuter nominals are characterized by the synchronic syncretism of nominative, vocative, and accusative in all numbers. Moreover, in Greek and Avestan neuter plural subjects take singular verbal agreement.

Anatolian shows a distinction between two genders, ‘common’ (including both animates and inanimates) versus neuter, which can be compared, respectively, to the masculine and the neuter of the other branches. Whether the absence of the feminine is an extremely archaic or innovative feature is debated (Melchert, 2017, p. 178). A peculiarity of Hittite neuter nouns is that their nominative-accusative form can figure as an object of a transitive verb or as a subject of an intransitive verb but not as a subject of a transitive verb: instead, nominative forms of derived common-gender nouns in -ant- are employed. According to some analyses (Legate, 2014), this pattern is an instance of split-ergativity based on arbitrary morphological classes (i.e., common vs. neuter) rather than on the animacy/definiteness scale (as is typologically commoner).

Unlike in Uralic and Semitic languages, no possessive affixes are found in IE languages.

Gradation of adjectives includes synthetic comparatives and superlatives: cf. Gr barús, barúteros, barútatos; Lat gravis, gravior, gravissimus, ‘heavy, heavier, heaviest’ (see Table 1). Insular Celtic has a set of four forms, including a synthetic equative: cf. OIr dub ‘black’, duibiu ‘blacker’, duibem ‘blackest’, duibithir ‘as black as’. High-frequency adjectives have often suppletive gradation forms: cf. Lat bonus, melior, optimus ‘good, better, best’.

Number marking and NP-internal gender/number/case agreement are typically obligatory. This morphosyntactic condition reduces or eliminates ambiguity in sentences with syntactically discontinuous NPs, which are common in the archaic poetic style well testified by Vedic Hymns (Old Indic) and Pindar’s Odes (Grk) (cf. Watkins, 2002):

(8) Vedic. RigVeda 4.41.4

(9) Ancient Greek. Pindar, Pythian10.55–56

2.7 Nominal Endings, Cumulative Exponence, Syncretism, and Stem Alternations of Nouns

Endings express number and case cumulatively, with the vocative being identical to the nominative except for the non-neuter singular (but in Vedic Indic vocative also shows peculiar proto-tonicity or unstressedness), and the ablative originally having autonomous non-syncretic (i.e., non-homonymous with any other case) forms only in the singular of the thematic class and of some pronouns.

In word classes serving as the target of agreement (pronouns, adjectives, participles), the feminine gender is distinguished from the masculine and the neuter by what was originally an athematic suffix (reconstructed according to the three-laryngeal theory as PIE *-H₂-, or—with bases ending with -i-, -u- or consonant—*-iH₂-; cf. Meier-Brügger, 2003, p. 218). This element fused early with the following number/case ending and, if present, with the preceding thematic vowel, thus creating cumulation of gender, number, and case: cf. Grk né-ās ‘new-f.sg.gen’ (< PIE *néw-e-H₂-es, where -e- is the thematic vowel and H₂ has an a-coloring effect) as opposed, for example, to né-os ‘new-m.sg.nom’ (< PIE *néw-o-s, where -o- is the thematic vowel).

As is shown by the OI -a- adjectives (Tables 2.1 and 2.2), in such three-dimensional paradigms, several groups of cells associated with combinations of relatively rare values of gender, number, and case (e.g., feminine, dual, ablative) share the same ending (cf., e.g., the cells exhibiting the morph -ābhyām) (Milizia, 2013). In IE nominals this kind of syncretism typically affects gender and case values in the context of the less frequent number values (dual and singular). A further development in the same direction, shown, for example, by Russian and German, consists of the total loss of gender distinctions in the non-singular (cf. Baerman, Brown, & Corbett, 2005, p. 28).

While case syncretism in the context of a number value is frequent (e.g., OI and Latin show a general dative-ablative syncretism in the plural, going back to PIE), number syncretism in the context of a case value is rare: an instance is provided by Hittite, where the ablative and instrumental cases do not distinguish singular and plural (Hoffner & Melchert, 2008, p. 76; for a Tocharian example, see Pinault, 2008, p. 468).

OI athematic nouns and adjectives can show alternations between two stems (so-called strong and weak), mostly distinguished by having different apophonic grades in the predesinential syllable. An OI pattern of stem indexing is represented in table 3, where the cells associated with the strong stem are gray-colored (more in Stump, 2001, p. 186).

Table 3. Declension of the Masculine and Neuter of OI bhágavant- ‘fortunate’

Open in new tab

Open in new tab

2.8 Pronoun Inflection

In Old Indic, Greek, and Latin, demonstrative and relative pronouns use ending sets partially identical to those of the thematic adjectives, but they have peculiar endings for specific groups of cells. The set of divergent cells partially varies across languages but typically includes the nominative-accusative singular neuter:

IE pronouns, especially the personal ones, are generally characterized by a high degree of stem allomorphy/suppletion: e.g., OI ahám/mā́m ‘I.nom/acc’ (cf. Eng. I/me). Nominal and pronominal inflections can also be distinguished by different syncretism patterns: for example, in Old Armenian the nominative and accusative are distinguished in the singular only in the personal pronoun. Specific inflections with specific syncretism patterns can also be associated with clitic pronouns; for example, in OI genitive-dative syncretism is peculiar to clitic personal pronouns (me ‘I.gen/dat’).

While IE clitic pronouns typically do not appear as subjects, Anatolian developed third person subject clitic pronouns apparently restricted to unaccusative intransitive verbs (i.e., appearing with, e.g., irmaliya- ‘get sick’ but apparently neither with, e.g., išḫamāi- ‘sing’ nor with a transitive verb).

2.9 Verbal Grammatical Categories, Stems, and Endings

In Old Indic and Greek verbal inflection is compartmentalized according to the following four aspect-tense categories:

The perfect is a resultative that may involve a change in valency, e.g., Grk kat=er~ḗrip-e ‘lies in ruins’, perfect of katereípō ‘destroy’. As a consequence, the proper imperfective correlate of a perfect like Grk pépoitha ‘I trust’ is the middle-voice peíthomai ‘I am persuaded’ rather than the active peíthō ‘I persuade’.

Each paradigm section has its own stem, though future stems are realized in etymologically non-identical ways in the two languages.

Different stem-formation processes are found for the imperfective and the perfective, the choice among them being determined by the verbal lexeme: cf., e.g., OI reduplicated and nasal-infixed presents such as dá-dhā-ti ‘puts’, yu‹ná›k-ti ‘joins’. If employed at all, the suffix-less and athematic B+E structure is able to serve, on a lexical basis, either as imperfective or as perfective: e.g., OI á-pā-t ‘protect.impfv.pst.3sg’, á-dhā-t ‘put.pfv.pst.3sg’ (on the related notion of metaconjugation, see Stump, 2016, pp. 202–217). See table 4 for the present active of two thematic OI and Greek verbs.

Endings cumulatively mark subject agreement (with nominative-accusative alignment) and, by means of the use of different ending series, several morphological oppositions:

Unlike in Uralic, IE verbs do not exhibit object agreement.

Instances of multiple exponence are not infrequent in this kind of system: thus, for example, in the Greek perfect léloipa ‘I have left’ the presence of the Ce- reduplicant le-, the appearance of the o-grade alternant loip- of the root lVip-, and the selection of the 1sg ending of the perfect-series -a are all associated with the category of perfect (compare the present leípō ‘I leave’).

In addition to indicative and imperative, the archaic system possesses two other moods, the subjunctive (also used in Vedic for future events, in competition with the future tense) and the optative (which indicates wish or possibility), both marked by pre-desinential suffixes. With athematic verbs, the subjunctive is marked by the presence of a thematic vowel (e.g., OI ás-ti/ás-a-ti ‘be.3sg.ind/sbjv’), whereas in the subjunctive of thematic verbs the thematic vowel is lengthened (e.g., OI váhate/váhāte ‘carry.mid.3sg.ind/sbjv’, Grk phéretai/phérētai ‘carry.mid.3sg.ind/sbjv’). The optative is signaled by the suffix OI -yā-/-ī-, Grk (postvocalic) -iē- < PIE *-yeH-/iH- (e.g., OI dadhyā́t ‘put.3sg.opt’, cf. Grk titheíē), or, for thematic verbs, by OI -e-, Grk -oi- (e.g., OI bhareta, Grk phéroito ‘carry.mid.3sg.opt’).

In both Old Indic (and Old Iranian) and Greek the categories temporally specified as past, that is, the imperfect (the imperfective past built on the ‘present’ stem) and the aorist (perfective), are characterized by the presence of the prefix OI a-, Grk e-, the ‘augment’, and by the selection of endings of the so-called secondary series.

In Vedic Old Indic augmentless counterparts of the imperfect and aorist, which are also found in Old Iranian and in Homeric Greek, function as an additional verbal category, the so-called present injunctive and aorist injunctive. The Vedic injunctive has been described as unspecified for mood and tense (Kiparsky, 2005). An innovation observable in Old Indic is the use of the original perfective past (aorist) as a recent past.

PIE had no passive and a passive diathesis distinguished from the middle is only partially developed in Indo-Iranian and in Greek. Old Indic and Old Iranian have a -yá-suffixed passive only in the present and a special 3sg passive aorist form in -i. In Greek, the suffixes -ē-, which was originally a derivational marker for inchoatives of states (cf. Grk emánēn ‘I became mad’), and -thē- serve to form aorist and future stems with passive meaning.

In both Greek and Old Indic the middle diathesis serves for a series of functions having in common the fact that the described situation affects the subject of the verb: e.g., auto-benefactive, indirect reflexive, reflexive—though this last is typically expressed in the active by means of a reflexive pronoun (Klaiman, 1991, pp. 82–104). Moreover, the middle diathesis can serve as a passive (with tenses lacking a dedicated passive form) and as the intransitive member of a causative/anticausative opposition: e.g., Grk egeírō/egeíromai ‘I awake (transitive/intransitive)’; OI vardhati/vardhate ‘makes/becomes bigger’. Verbs lacking the active or the middle inflection (activa or media tantum, respectively) are not uncommon: e.g., Grk baínō, active, ‘I go’; Grk hépomai, middle, ‘I follow’.

A set of active and middle participles formed from the temporal-aspectual stems is inherited from PIE by both Old Indic and Greek, while other kinds of nominal forms, including infinitives and gerundives, were mostly developed in the history of the different IE branches.

Despite the antiquity of its attestation, Hittite exhibits a much poorer verbal morphology, which distinguishes only two moods (indicative and imperative) and two tenses (present and preterite). As in the case of the lack of the feminine, this system is considered highly archaic or highly innovative depending on the scholar.

2.10 Word Formation

The archaic type possesses a large number of derivational patterns, with noun formation processes being typically gender-imposing. Thus, for the Old Indic base śro-/śru- ‘hear’ we have, e.g., the masculine noun OI śravá- ‘act of hearing’, the feminine śrúti- ‘act of hearing’, the neuters śrótra- ‘instrument of hearing, ear’, śrávaṇa- ‘act of hearing’, śrávas- (= Grk kléos) ‘sound, resonance, fame’, the agent noun śrotár- ‘one who hears’, the adjective śrutá- ‘heard, famous’. Specific motion processes are employed for creating feminine counterparts of masculine human or humanized nouns: e.g., OI vŕ̥ka- ‘wolf’ ~ vŕ̥kī- ‘she-wolf’, janitár- ‘genitor’ ~ jánitrī- ‘female genitor’ (cf. Latin pairs lupus, lupa and genitor, genetrix).

Old Indic verbal morphology is enriched by a series of productive derivational processes forming deverbal verbs (causatives, desideratives, and intensives), which, together with the realization of inflectional categories, can produce forms with a high degree of both synthesis and formal complexity, for example,

(10) Classical Sanskrit (Old Indic). Bhattikavya 2.40

In this form (i) ghr̥k- is the allomorph before (s/ṣ [s/ʂ]) of the ablaut-alternant gr̥h- of the root grah-; (ii) ni=grah- is a compound lexical unit comprising the preverb (ni-); (iii) reduplication (ji-) accompanied by -s/ṣ(a)-suffixation is the marker of the desiderative; (iv) -ay(a)- is the suffix of the causative; (v) -(i)ṣy(a)- is the suffix of the future.

Moreover, the formation of denominative verbs is very productive in Old Indic (e.g., amitrayáti ‘behaves like an enemy’ from amítra- ‘enemy’), and Greek (e.g., basileúō ‘I am king’ from basileús ‘king’, or paideúō ‘educate’ from paîs ‘child’, genitive paidós).

In this morphologically rich system, though non-derived lexemes do, of course, exist (cf. OI nás- ‘nose’ f.), derivationally complex words are significantly frequent. This has led scholars to consider the complex “(root+suffix)+ending” word structure as the default one and to view derivationally simple “root+ending” words as special instantiations of the general complex pattern, i.e., as ‘(root+0)+ending’. According to Meillet (1937, p. 147), “there is no bare root: there are only stems that are characterised by the absence of a suffix, or, stated differently, by a ‘zero suffix’.”

On the other hand, in a so conceived morphological system, the dichotomy between simple and complex words comes up again as a distinction between primary and secondary derivatives—a division which is indeed traditional in IE studies. Crucially, secondary derivation, whereby stems give rise to further stems, may entail multiple suffixation: e.g., OI śro-/śru- ‘hear’ → śrú-ti- ‘hear-nmlz(f), (the) hearing’ → śrú-ti-mant- ‘hear-nmlz-proprietive, who has the hearing, i.e., who has learned’.

A noteworthy Old Indo-Iranian device of probable Indo-European origin for forming secondary derivatives is the so-called vr̥ddhi-derivation, characterized by the selection of an ‘incremented’ (cf. OI vr̥ddhi- ‘increment’) ablaut alternant of the vowel of the first syllable (accompanied, in several cases, by the thematicization of non-thematic bases): śrúti- ‘hearing’ → śrauta- ‘relating to the hearing’; śrótra- ‘ear’ → śrautrá- ‘relating to the ear’. Vr̥ddhi derivation is freely applicable to compounds and prefixed adjectives, so that it can make it possible for ablaut alternations to also appear in otherwise non-ablauting morphs: su-śrávas- ‘who has a good (su) fame (śrávas-)’ → sauśravasá- (same meaning).

A characteristic feature of the archaic IE type is the productivity of compounding, a peculiarity of IE composition being that the first element normally appears in its stem form (i.e., without inflectional endings), even when it is a nominal or a verb: e.g., Grk arguró-toksos ‘he who has a silver (árguros) bow (tókson)’ (semantically exocentric); kako-ī́lios ‘the unhappy (kakós) city of Ilios’ (semantically endocentric). It is typologically remarkable that in this system exocentric (‘possessive’) compounds without verbal constituents (cf. arguró-toksos) are frequent. Ancient IE compounds, when they are not coordinative, are typically right-headed, but left-headedness is also found with nominal compounds with verbal head: Grk pheréoikos ‘the one who bears (phere-) his house (oîkos) = snail’.

Compound nominals played a non-minor role in the culture of several ancient Indo-European peoples (cf. also Schmitt, 1973), serving as a common, though not exclusive, way of forming anthroponyms—e.g., Grk Eu-klē̂s ≈ Middle Persian Hu-sraw ‘having good fame’, Runic (Germanic) Haþu-wulafr ‘wolf of the battle’—epithets of heroes or gods—e.g., OI Índra- Vr̥tra-hán- ‘Indra the killer of Vr̥trá-’—and, especially in Germanic and Celtic, poetic kennings—e.g., OE hilde-nædre ‘war-snake, i.e., arrow’.

In Old Indo-Aryan the use of compounding increases so that in Classical Sanskrit nonce, non-lexicalized compounds are commonly used in place of syntactic expressions (e.g., X-nāman- ‘whose name is X’, X-tīra- ‘bank of the river X’).

Compound verbs are generally less common than compound nouns in IE languages (cf. Cowgill, 1966, p. 129), except for preverbed verbs (see section 3.13).

3.1 Morphological Decay and Morphological Enrichment

In the historical development of several IE languages, recurrent phenomena are observed that lead to an increase in configurationality and a decrease in the degree of synthesis (use of adpositions at the expense of case forms, rise of auxiliaries, and increasing use of periphrastic morphology, creation of determiners). In this respect, English morphology represents a good example of extreme departure from the ancient type.

The development of IE languages has been often—in particular in the 19th century—represented as a history of “morphological decay” (cf. Bynon, 2004), with progressive simplification of inflectional categories and loss of morphologically meaningful phonological material in the diachronic evolution of words, as shown by the comparison by Schleicher (1860, p. 34) between Goth habaidēdeima ‘had.pst.sbjv.1pl’, Ger hätten and Eng. had. This view, however, can grasp only a part of the overall picture. Not only do many modern IE languages preserve relatively complex inflections, but in several cases secondary synthetic forms, resulting from univerbations involving auxiliaries or pronouns, coexist with other inherited ones: cf., e.g., in Marathi the resynthesized present ćāl^atos ‘go.prs.m.2sg’ (deriving from a present participle plus copula) and the inherited past habitual (< old present) ćāles ‘go.pst.hab.2sg’, in Lithuanian the resynthesized conditional (“optative”) dègčiau ‘burn.cond.1sg’ (ultimately deriving from a periphrasis containing the supine) and the inherited present degù ‘burn.prs.1sg’, in Italian the resynthesized conditional scriverebbero ‘write.cond.3pl’ (from infinitive plus a past form of ‘have’) and the inherited present scrivono ‘write.prs.3pl’. Let alone the fact that, according to most scholars, even Gothic habaidēdeima (and the Germanic “weak” preterite in general) is the result of a pre-documentary resynthesization involving the past tense of the Germanic verb for ‘do’.

As for the degree of fusion (i.e., lack of segmentability and invariance of morphs; cf. Comrie, 1989, pp. 44–52), which is the actual peculiarity of the “flexive morphology,” it often increases in the development of the single languages. Thus, for example, the Latin noun shows a greater number of inflectional classes than Proto-Indo-European, whereas verb fusionality seems to have sharply risen in the development from Latin to Romance—see, e.g., Matthews (1970, p. 180) on the low transparency and segmentability of the Italian verb, and Maiden (2018) on the complex patterns of verbal stem allomorphy exhibited by several Romance languages.

Thus, if the contemporary Baltic and Slavic languages typically exhibit particularly rich and complex morphologies, this is due to conservative as well as to innovatory features: cf., e.g., in Lithuanian the preservation of several case oppositions on the one hand, and the creation of a new complex system of verbal inflection classes (see Dressler, Kilani-Schoch, Gagarina, Pestal, & Pöchtrager, 2006) on the other.

3.2 Morphophonological Developments

In the history of several IE languages, regressive (contact or distance) assimilation phenomena triggered by segments belonging to inflectional endings or suffixes make inflectional signals appear to the left of the original morphological exponent. Instances of umlaut (regressive distance vowel-to-vowel assimilation), also called ‘metaphony’, are largely recurrent. In North and West Germanic, new alternations due to umlaut complement the inherited ablaut alternations: cf., e.g., Ger werdet ~ wurdet ‘become.2pl/became.2pl’, with the e(r)/u(r) alternation stemming from the ancient ablaut, and wurdet ~ würdet ‘became.2pl.ind/sbjv’, where the u/ü [u/y] alternation originated from an u > ü sound change (the regressive distance assimilation of frontness called “i-umlaut”) triggered by the marker -ī- (the inherited optative suffix) which originally appeared in the second syllable of the subjunctive. Instances of umlaut alternations with morphological function are also found in several Romance varieties: cf. Upper Southern Italo-Romance [ˈmesə]/[ˈmisə] ‘month.sg/pl’, with height assimilation [ˈmisə] < [*ˈmesi]. Eastern examples morphologized umlaut phenomena are not lacking: cf. Shughni (New Eastern Iranian) tux̌p/tax̌p ‘sour.m/f’ < *tr̥fšah/tr̥fšā.

Consonantal palatalization is also easily morphologized: cf., e.g., OIr macc/maicc [mak/mak^j] ‘son.nom.sg/gen.sg’, with maicc < *makk^wī. Significantly, the principles according to which the inflectional paradigm is organized can pass unaltered through such changes in the means of expression. Thus, in the Old Irish example above, the shift from ending-affixation to segmental alternation does not affect the inflectional cumulation of genitive and singular.

A peculiar feature of Insular Celtic is morphophonological alternations that involve the initial segment of a word and are the diachronic product of sandhi phenomena (i.e., phonological alternations occurring at word boundaries) triggered by the subsequently disappeared ending of the preceding word: for example, Old Irish a thech [a θ‎ex]/a tech [a tex] ‘his/her house’ < *esyo#tegos / esyās#tegos, where initial spirantization does not appear with the feminine possessive due to the originally post-consonantal position of the word boundary.

New patterns of paradigmatic accentual mobility were created in Balto-Slavic in both nominal and verbal inflection: for example, Lith galvà/gálva ‘head.sg.nom/sg.ins’; Russian mogú/móžete ‘be able.1sg/2pl’ (see Halle & Kiparsky, 1981; Jasanoff, 2017). In the Latin-Romance phylum, where contrastive stress was first lost and subsequently acquired again, newly arisen patterns of stress mobility in verbal inflection (e.g., Italian cantávo/cantavámo ‘I sang/we sang’) tend to undergo intra-paradigmatic leveling, according to the so-called stress column principle (cf. Spanish cantába, cantábamos) (Lüdtke, 1987).

3.3 Inflectional Classes

The PIE opposition between thematic and athematic inflection class is superseded in the different languages by new inflection class systems, in which secondary thematic vowels are created: thus Lat manus ‘hand’ and ignis ‘fire’, whose forerunners could inflect according to the same (“athematic”) declension and could be segmented as manu-s, igni-s (nom sg), synchronically belong to the u-class (fourth declension) and to the i-/consonantal-class (third declension), respectively, and can be analyzed as man-us, ign-is: cf. gen sg man-ūs, ign-ĭs. Analogous developments are found, mutatis mutandis, in verbal morphology, often exhibiting complex inflectional class systems which can be usefully described as hierarchically organized in macro classes and various levels of subclasses (cf. the systems of Lithuanian, Russian, German, Latin, and French described in Dressler et al., 2006).

Newly created noun inflectional classes can be associated with a particular gender value. Thus, in Russian—except for nouns with semantic gender assignment—nouns of declension II (e.g., škola ‘school’) and III (e.g., kostˊ ‘bone’) are always feminine, whereas masculine and neuter nouns belong to declensions I/IV. In some cases verb inflection classes are categorically or tendentially restricted to intransitive or to transitive verbs. Thus, for example, all the verbs of the IV weak conjugation of Gothic are intransitive (e.g., fullnan ‘become full’), whereas the largest number of Marathi verbs is distributed into two inflection classes according to their transitivity value (Masica, 1991, p. 261).

3.4 Cases and Adpositions

A recurrent development among IE languages (cf. Table 5) consists of the reduction of the number of case values (“diachronic case syncretism”) accompanied by an increasing resort to adpositional phrases (typically prepositional but postpositional in some OV languages like Hittite, Tocharian, and New Indo-Aryan). In most of the IE languages (a special case is that of New Indo-Arian, cf. Bloch, 1965, pp. 179–181; Reinhöl, 2016), the main diachronic source of prepositions is a set of local particles which could originally be employed as both adverbal and adnominal modifiers (Cuzzolin, Putzu, & Ramat, 2006; Hewson & Bubeník, 2006; Kuryłowicz, 1964, pp. 171–178), as can be seen in Vedic Old Indic. Consider the following three examples with directional accusative (taken from Hettrich, 2007), in which no proper adposition is found:

(11) Rigveda 6.18.9 (no particle is present)

(12) Rigveda 5.33.3 (the particle ádhi is used adnominally with rátham, without being a true adposition)

(13) Rigveda 5.63.1 (the particle ádhi modifies the verb)

Different kinds of relations may obtain between adposition and case. In Ancient Greek eight prepositions can govern three cases (like epí ‘upon’, which can govern genitive, dative, and accusative), three prepositions can govern two cases (e.g., hupér ‘over’ can govern the genitive or the accusative), and seven prepositions govern a single case (like ek ‘from’, which selects the genitive, or sun ‘with’, which selects the dative)—thus roughly two cases per preposition on average, which means that in this language case values bear a large amount of information even in prepositional phrases. In Gothic the corresponding ratio is less than 1.5.

IE languages often have a set of local prepositions that can govern two cases, with the selection between the two usually depending on whether the phrase indicates location or goal of motion (in which circumstance the selected value is accusative): cf. Lat in ‘in’, sub ‘under’, super ‘over’. In several languages (e.g., in Latin, Old Irish, and Modern German) these local prepositions are the only ones to govern more than one case. On the other hand, the Russian locative and the Late Old/Middle Irish do not typically occur outside prepositional phrases and, therefore, are also called “prepositional.” Besides local expressions, comitatives and passive agents (especially if [+human]) are frequently associated with prepositional structures (cf. Heine & Kuteva, 2006; Hettrich, 1990; Luraghi, 1986).

Some IE languages have acquired secondary sets of cases. Old Lithuanian developed a subsystem of local cases, partially analogous to that of Finnic, comprising an illative (motion into), an adessive (location by), and an allative (motion to/toward) in addition to the inherited locative. The new cases are obtained by univerbating a case form with a postposition. As the inherited ones, they trigger agreement (or “spreading”; cf. Bickel & Nichols, 2007, p. 235) within the NP: cf., e.g., the adessive NP támpi žmõgup ‘that.adess.sg man.adess.sg’, with the original postposition -p(i) appearing twice (cf. Stang, 1966, p. 228). Secondary cases created through the univerbation of postpositions with a case form are also found in Tocharian. In this IE branch an optional pattern of group inflection—i.e., a series of coordinated nouns exhibiting case marking only in the last element —presumably originating in NPs with secondary cases is extended to the non-secondary genitive (cf. Pinault, 2008, pp. 463–467).

The “secondary cases” of New Indo-Aryan have been also analyzed as noun + postposition sequences. In this view, a layered structure like Hindi laṛke ke sāth ‘with the boy’ is seen as the oblique case laṛke of laṛkā, ‘boy’, followed by two stacked postpositions (ke lit. ‘of’ and sāth ‘with’) rather than as a secondarily created genitive laṛke ke plus a single postposition (Masica, 1991, pp. 231ff.; Spencer, 2005).

3.5 Typological Shifts in Nominal Morphology

Some IE languages develop toward an agglutinative nominal morphology (cf. also Igartua, 2015). A particular system is that of Sogdian nouns (Middle Eastern Iranian), where “light stems,” which distinguish six cases, exhibit the expected number/case cumulative exponence, while “heavy stems,” which have only two case forms, use separate number and case markers. In Yaghnobi, a New Iranian dialect strictly close to Sogdian, agglutinative inflection is found with all nouns. In the Caucasian area, Ossetic (genealogically a New Eastern Iranian language) also expounds number and case separately. Moreover, its highly restructured case system comprises a comitative, a similative, and a subset of local cases. Separative exponence of number and case is also found, in most of the inflections, in New Armenian nouns (cf. Dum-Tragut, 2009, pp. 68ff.).

In Persian and other Western Iranian languages, an element called ezafe, which is classified as a phrasal affix or as a clitic and continues the PIE relative pronoun *yo-, signals the presence of a following adjective or possessor noun, thus creating a pattern of head-marking (cf. Nichols, 1986): e.g., kuh-e boland ‘mountain-ez high = the high mountain’, sāye-ye deraxtān ‘shade-ez trees = the shade of the trees’. In other Iranian varieties, the ezafe exhibits a complex morphology: thus, in Zaza, its form varies according to the morphosyntactic properties of the head noun and to the word class of the dependent element (Paul, 1998, pp. 30–31).

3.6 Developments Involving Number, Gender, and Noun Classes

The dual got lost in most of the IE languages. In Anatolian languages, Phrygian, Latin, Sabellian (Oscan and Umbrian) languages, Armenian, and Albanian it is absent already in the earliest records, while Indic, Iranian, and Greek preserved it only in their old period. Among the contemporary IE languages the dual is alive in some Slavic languages (Slovene, Sorbian, Slovincian) and in Lithuanian (in particular in Lithuanian pronouns). While in Vedic Indic the use of the dual is obligatory, other languages, e.g., Attic Greek (Meillet, 1930, p. 47) and contemporary Slovene (Corbett, 2000, p. 43), exhibit a different usage pattern, by which the dual is optional especially with natural pairs such as dual body parts. In Tocharian special dual endings are found with nouns whose referents can constitute natural pairs (e.g., ‘eye’), which has led some scholars to the unsound (cf. Pinault, 2008, p. 462) postulation of a peculiar ‘paral’ number value. Some Eastern Iranian languages—Sogdian (Middle period), Pashto—and Middle Welsh (Insular Celtic), along different paths, developed special numerative forms used with numerals (Pedersen, 1913, p. 132; Sims-Williams, 1979).

A series of languages loses gender as a category: Armenian (since the earliest records), several Iranian languages (among which Persian—since the Middle period—Ossetic, Balochi, Sulaimania Kurdish, Parachi, Wakhi) and New Indo-Aryan (NIA) languages (spoken Sinhala, Nepali, and the Eastern languages Bengali, Assamese, Oriya).

Assamese and other, mostly Eastern, NIA varieties developed a system of numeral classifiers, a feature typically found in mainland Southeast Asian languages (Aikhenvald, 2000, pp. 102–103; Barz & Diller, 1985). In Sinhala (a language showing heavy Dravidian influence) a numeral classifier is taken only by human nouns (Geiger, 1900, pp. 65–66; Hurford, 2003, p. 590).

‘Personal’ numeral series reserved for nouns denoting humans have been developed in Old Irish (McCone, 1994, § 35.7; cf. Hurford, 2003, on Scottish Gaelic) and Bulgarian (Scatton, 1993, p. 209).

A masculine:feminine system obtains—mostly via inclusion of neuter nouns in the masculine—in a large number of languages: Middle Irish and Middle Welsh, almost all Romance varieties, most of the Indo-Iranian languages. Among modern languages, the three-gender system is preserved in Marathi and Gujarati (NIA), in Greek, in Albanian, in German, Icelandic, and Faroese (Germanic), and generally in Slavonic languages. On the contrary, a common-gender:neuter opposition arises in some Germanic languages (Dutch, Danish, Swedish) through the merger of masculine and feminine.

The usage of the neuter (limited to the singular number) with non-nominal controllers or in impersonal predicatives—that is, the usage of the neuter as a ‘neutral’ in the sense of Corbett (1991, p. 159, 2006, p. 97)—may persist even in languages having lost neuter nouns. In Lithuanian controllers triggering neuter agreement can be pronouns or adjectives converted into nouns but not true nouns, and neuter gender is also used with adjectives appearing in impersonal predicates: e.g., pirkiojè tamsù ‘house.loc dark.n = it is dark in the house’ (Ambrazas, 1997).

Similar or partially similar situations are found in a dialect of Slovene (village of Sele; Corbett, 2006, p. 97) as well as in some languages of the Romance group (Loporcaro, 2018). In Old French and Old Occitan and still in modern Sursilvan, Romansh adjectives have neuter forms reserved for agreement with non-nominal controllers:

(14) Old French. Roman d’Eneas, 1629 (with an infinitive as a controller of agreement)

In many contemporary Romance languages this kind of neuter is found in pronouns and determiners: cf. Spanish el.m/ella.f/eso.n ‘that’.

In some Romance varieties, the neuter, in addition to preserving the inherited property of serving for agreement with non-nominal controllers, is also found in association with uncountable nouns, a usage which also requires only singular-number forms. Thus, in Asturian a non-prenominal target of agreement appears in the neuter gender when the controller noun (whether masculine or feminine) is uncountable: Asturian of Lena la tsiche frio/**friu/**fria ‘def.f milk.f cold.n/**m/**f, the cold milk’ (see Loporcaro, 2018, p. 172). In a part of the Central-Southern Italo-Romance area neuter forms of some agreement targets (usually pronouns and determiners) are required with nominal controllers belonging to an arbitrary subset of (formerly masculine or neuter) uncountable nouns.

A non-autonomous ‘alternating’ gender with masculine agreement in the singular and feminine agreement in the plural was created—mainly due to phonological mergers—in Tocharian and in some Romance languages (Romanian, but also Italo-Romance varieties) in a parallel way. An alternating gender is also found in Albanian (Pedersen, 1897, p. 290). In Slavic languages, subgender distinctions arose based on the animate/inanimate opposition (Corbett, 1991, pp. 161–167; Vaillant, 1958, pp. 17–18).

3.7 Definiteness

Ancient IE languages (e.g., Old Indic, the Homeric variety of Ancient Greek, Latin, and Hittite) had no definite article. In Europe, this situation is generally preserved in Baltic and Slavic (exceptions are Macedonian, Bulgarian, and spoken Slovene; see Priestly, 1993, p. 411; see also Trovesi, 2004). The other IE languages of Europe sooner or later—Greek already in the Ancient period—developed fully grammaticalized definite articles, typically from original demonstratives (cf. Heine & Kuteva, 2006, pp. 97–139). A definite article is also found in the languages of the Romani group (Matras, 2004, pp. 96–98), which belong to the Indo-Aryan branch but have been spoken in Europe since the Middle Ages. In the Germanic Scandinavian Languages and in some languages of the Balkan area (Albanian, the Romance language Romanian, and the Slavic languages Bulgarian and Macedonian) postnominal definite articles were developed, the status of which is debated with respect to the affix ~ clitic distinction (Börjars & Harries, 2008; Ortmann & Popescu, 2000).

In Baltic and Slavic, definiteness becomes a dimension of the inflectional paradigm of adjectives, since a distinction arises between an indefinite and a definite declension, the latter being created through the univerbation of adjectives with a postposed (originally relative) pronoun continuing PIE *yo-: e.g., Old Church Slavonic slěpa žena ‘a blind woman’ vs. slěpaja žena ‘the blind woman’ (Lunt, 2001, p. 64). Due to the lack of a definite article, in the archaic Balto-Slavic system an NP without a demonstrative element is typically ambiguous as to definiteness unless it contains an adjective. This situation is found in Old Church Slavonic, Lithuanian (Stolz, 2010), and Latvian (Lyons, 1999, p. 84), whereas in modern Slavonic languages the definite (or ‘long’) forms tend to be generalized, at least in attributive function (see Corbett, 2004, p. 206; Timberlake, 1993, p. 863).

An only partially analogous situation is found in Germanic (cf. Harbert, 2007, pp. 130–137), where adjectives have a ‘strong’ and a ‘weak’ declension, with the latter being formed by the addition of a nasal formative that continues a PIE individualizing suffix (cf. Grk Strábōn ‘the squinty-eyed one’ vs. strabós ‘squinty-eyed’). Germanic weak adjectives, though being typically employed in definite contexts, do not have a disambiguating function since, as occurs, for example, in Gothic, they are used in vocative phrases and in phrases where definiteness is signaled by the context—that is, in phrases which contain, in addition to their head noun, a demonstrative or a personal pronoun:

(15) Gothic. Ephesians 6.13

3.8 Developments Involving Stems and Endings of Verbs and Person Agreement

In several IE languages lexemes belonging to non-productive inflection classes often show sets of mutually unpredictable stems associated with different subsections of the paradigm (‘principal parts’). This lack of predictability can be due to word-internal sound changes as well as to purely morphological diachronic reasons. In Latin, for instance, the form of the present stem cannot be predicted from a perfectum stem ending in -x- [ks], since [ks] is the phonological outcome of several dorsal + s clusters: cf., e.g., dīcō/dīxī ‘I say/said’, coquō/coxī ‘I cook/cooked, vehō/vexī ‘I convey/conveyed’. On the other hand, because of its heterogeneous morphological origin, the stem of the perfectum can exhibit several alternative formal devices, among which reduplication (like the PIE perfect)—cf. mordeō/momordī ‘I bite/bit’—and s-suffixation (like the PIE s-aorist)—cf. suādeō/suāsī ‘I advise/advised’. Therefore, the present stem does not allow speakers to predict the shape of the perfectum.

Stem suppletion, expecially with high-frequency verbs, is a recurring phenomenon, but suppletive pairs are typically different in the different languages (cf. Grk phérō/ḗnegkon ‘carry.prs.1sg/pst.pfv.1sg’ ~ Lat ferō/tulī ‘carry.prs.1sg/ pst.pfv.1sg’) and are often newly created in the history of the single languages: thus the suppletive pairs English go/went, French vais/allons ‘go.1sg/1pl’, and Marathi jāṇẽ/gelā ‘go.inf/go.pst.3sg.m’ were unknown to the respective ancestor languages (Old English, Latin, and Old/Middle Indic).

A particular distinction between ‘absolute’, ‘conjunct’, and ‘relative’ inflection arose in Old Irish. Absolute inflection, which shows phonologically heavier personal endings, is restricted to sentence-initial verbs, whereas compound (i.e., preverbed) verbs and verbs preceded by a sentence-initial “conjunct particle” (a class comprising conjunctions as well as negative and question particles) take the reduced endings proper to the conjunct conjugation: e.g., bermai [b^jermi] ‘we carry’ ~ ní beram [b^jerəμ‎] ‘we do not carry’; beirid [b^jer^jəð^j] ‘carries’ ~ ní beir [b^jer^j] ‘does not carry’. The conjunct conjugation is also selected with compound verbs formed with the empty preverb no-, whose sole function is to allow the appearance of a pronoun of the so-called infixed class, when the selected verb is not compound and no conjunct particle is present. This occurs because ‘infixed’ pronouns can only be placed between a preverb or a conjunct particle and the verb: no-t-beir ‘carries you’. The different hypotheses on the origin of the conjunct inflection mostly rely on the fact that second-position clitics may have prevented the endings of a sentence-initial verb (Old Irish has VSO word order) from undergoing the phonological decay that otherwise regularly affected the end of prosodic words (McCone, 1994, § 24.1). The ‘relative’ endings, signaling a relative clause, arose in Old Irish from original sequences of verb plus clitic relative particle: e.g., bertae [b^jerde] ‘carry.3pl.relative’ = ‘who carry’ or ‘whom they carry’ < *beronti-yo vs. berait [b^jerəd^j] / -berat [b^jerəd] ‘carry.3pl.absolute/conjunct’.

A noteworthy development of some Eastern New Indo-Aryan languages such as Bengali, Maithili, and Assamese is the refunctionalization of original plural verbal endings as markers of honorificity (Masica, 1991, p. 226). Maithili, moreover, developed inflections encoding two-dimensional person agreement (subject/object and other types—cf. Yadav, 1996, pp. 173ff.)

3.9 Developments Involving Diathesis

The middle diathesis, originally used in both middle function proper and as a passive, tends to be lost. While in Modern Greek both the contrastive mediopassive and the class of media tantum verbs are preserved (e.g., dénō ‘I bind’, dénomai ‘I am bound’, érkhomai ‘I come’), in Indo-Aryan the middle diathesis has practically disappeared since the Middle period. A similar decay is shown by Iranian, where the middle is still preserved in Khotanese and Sogdian (Middle languages of the Eastern group). On the other hand, Middle Persian developed a new passive suffix -īh- (Rastorgueva & Molčanova, 1981, pp. 124–125), which is not continued in New Persian.

Among the Germanic languages, only Gothic exhibits a continuation of the IE middle diathesis, which has only a passive function and is restricted to the present tense.

An identical specialization also obtains in Latin, where the section of the verbal paradigm built on the present stem has a synthetic passive marked by a set of personal endings which, in part, continue special variants of the PIE middle endings characterized by final -r (e.g., laudātur ‘is praised’)—middle endings in -r or -ri are also found in Celtic, Phrygian, Tocharian, and Hittite but not in Greek and Indo-Iranian. The same inflection pattern applies to originally media tantum verbs (e.g., sequitur ‘follows’, to be compared to Grk hépetai), which synchronically appear as active verbs which idiosyncratically take passive endings, i.e., as “deponents” (on deponency as a theoretical concept, see Baerman, Corbett, Brown, & Hippisley, 2007).

The situation of Old Irish, which also exhibits continuations of r- endings, is similar to that of Latin, with, however, the crucial difference that the passive inflections of 3sg and 3pl were secondarily differentiated from the corresponding deponent forms through the generalization of different patterns of syncope. Therefore, even deponent verbs can be passivized: e.g., seichithir, active, ‘follows’ vs. seichthir, passive, ‘is followed’. Old Irish deponents thus constitute a strictly speaking non-deponent inflection class. As for non-third persons, Old Irish passives are formed by means of an impersonal construction.

Similarly to Modern Greek, Albanian still preserves a mediopassive diathesis that, in the present/imperfect, is morphologically expressed by means of endings that continue the original PIE middle series: e.g., lidh, active, ‘I bind’ vs. lidhem, middle, ‘I am bound, I bind myself’.

In Old Armenian, a newly formed distinction between active and mediopassive is found in the perfective (“aorist”) section of the verbal paradigm of all verbs (e.g., beri ‘I carried’ ~ beray ‘I was carried’), and in the subsection of the imperfective exclusively for verbs belonging to a specific inflection class (-e- stems). In the rest of the paradigm, active and mediopassive are syncretic: e.g., əntʿeṙnu “he reads/(something) is read” (Godel, 1975, pp. 47–49). In Cilician Armenian (Middle period) this situation had already ceased and a suffix -vi- served to form the passive (Karst, 1901, pp. 292–298).

A North-European development involving Old Nordic (Germanic) and some Balto-Slavic languages (Lithuanian, Russian) is the creation of an anticausative derivation or of an inflectional middle voice by the grammaticalization of a reflexive clitic into a suffix (cf. Holvoet, 2015, pp. 455–463). In Old Nordic, where the outcome is clearly inflectional, a new middle ending series arises from the fusion of the endings with the reflexive sik (cf. Ger sich): e.g., kallizk (= kallitsk) ‘call.mid.2pl’ < kallið + s(i)k (Faarlund, 2004, p. 54; Noreen, 1923, pp. 367–373).

3.10 Developments Involving Moods

In general, the tense-aspect-mood systems of the IE languages show deeper restructurations than case-number-gender systems (Cowgill, 1966, p. 134).

The IE opposition indicative ~ imperative is relatively stable. As for non-indicative, non-imperative moods, Middle Iranian languages still preserved both the optative and the subjunctive (Sogdian also has two further moods), and an analogous pair of moods is also found in Tocharian, where the subjunctive, however, is a newly created category functioning as a perfective non-past.

The competition between subjunctive and optative forms is often resolved in favor of the latter. In Indo-Aryan the IE subjunctive was lost already in Classical Sanskrit, where, however, non-indicative non-imperative functions are filled not only by the optative but also by the imperative itself, which inflects for all persons. The IE optative is also the source of the Germanic subjunctive. In Latin, at least as far as the verb for ‘be’ is considered, the IE optative is the ancestor of the subjunctive—which, as in Germanic, is the only non-indicative/non-imperative mood—(cf. Old Latin sied < *s-yē-t), while the IE subjunctive is the basis of the future tense (erit ‘will be’ < *es-e-ti). Modern Greek, on the contrary, lost the old optative, but preserved the subjunctive, which, in the aorist subsystem, still has an inflection different from that of the indicative. The origin of the Old Irish subjunctive, which is also not opposed to an optative, is debated.

A recurrent phenomenon is the creation of a conditional mood formed as a ‘past of the future’. This type is found in Old Indic (cond. a-bhariṣya-t ~ fut. bhariṣya-ti, cf. present bharati ‘bears’) and in Old Irish (cond. -léicfed ~ fut. léicfid, cf. present léicid ‘leaves’). A similar relationship is also observed between modern Germanic periphrastic conditionals and futures (Eng. would do ~ will do) and between Romance resynthesized conditionals and futures, which derive from Lat habēre ‘have’ preceded by an infinitive: e.g., French chanterait ‘would sing’ < cantāre habēbat lit. ‘he had to sing’ ~ chantera ‘will sing’ < cantāre habet lit. ‘he has to sing’).

In Slavic the old optative is continued by the jussive/imperative, which is the only inherited mood besides the indicative (a conditional mood is formed periphrastically). New modal formations are found in Baltic (Stang, 1966, pp. 421ff.), including, in East Baltic, a resynthesized conditional which derives from a periphrasis involving the supine.

Albanian also has a highly restructured mood system, which includes a subjunctive, a conditional, and an optative (which probably reflects an original s-suffixed aorist optative). Moreover, Albanian exhibits the category of admirativity, which encodes the speaker’s surprise or annoyance.

The single non-indicative/non-imperative mood of Old Armenian, “the subjunctive,” contains a -icʿ- formative, which might be connected to an IE actional suffix *-(i)sk-.

The creation of evidential categories—i.e., categories codifying the source of information (e.g., hearsay or inference as opposed to observation by the speaker)—is observed in a series of IE languages of the Balkan and Circumbaltic areas: cf., e.g., the Bulgarian “renarrated mood,” the Romanian “presumptive mood,” and the Lithuanian “oblique mood” (cf. Aikhenvald, 2004—on evidentials in Indo-Iranian, see Bashir in Hock & Bashir, 2016, pp. 584–589).

As for non-finite verbs, particularly complex participle systems are found, leaving Ancient Greek aside, in the Slavonic and in the Baltic languages (cf. Vaillant, 1966, p. 111). In Tables 6.1, 6.2, 6.3, and 6.4, the participle systems of Greek, Latin, Gothic, and Old Church Slavonic are placed side by side.

A recurrent drift is the growing usage of gerund/converb forms (cf. the gerunds of Modern Greek, Romance languages, Classical Sanskrit) at the expense of the use of participial forms in converb-like function (i.e., as adverbial modifiers), the latter being a syntactic pattern characteristic of ancient IE languges, in particular of Ancient Greek and Latin (cf. Haspelmath, 1995).

The particularly large set of Ancient Greek infinitive forms (one for each combination of tense-aspect and diathesis)—a conspicuous innovation with respect to PIE, which apparently had no infinitival verb form—was completely dropped in Greek in the modern period, so that Modern Greek has no infinitive at all. The secondary loss—or the reduced employment—of the infinitive is also observed in other languages of the Balkan area (Macedonian, Bulgarian, Romanian, Albanian).

3.11 Developments Involving Tenses

The original threefold distinction between present/imperfect ~ aorist ~ perfect is simplified in several languages (cf. table 7). A recurrent development is, moreover, the shift from the PIE predominantly aspect-oriented system to a predominantly tense-oriented one (Latin-Romance, Germanic, Insular Celtic), with aspect distinctions being sometimes confined to the indicative of the past (as in Latin-Romance).

Tables 8.1, 8.2, 8.3, and 8.4 give a partial comparison between the rich system of synthetic forms of the Greek verb and those of Latin, Gothic, and Old Church Slavonic.

In Latin the perfectum, whose inherited forms are, on a lexical basis, continuations of either aorists (dīxī ‘I said’) or perfects (cecinī ‘I sang’), mostly indicates a completed past event. Its stem, moreover, is associated with a newly created verbal subsystem (the system of the perfectum) that is structured in a way largely parallel to that of the present (or infectum), in that both also comprise a ‘past’, a ‘future’, a subjunctive, and a ‘past’ subjunctive, these four categories being expressed by new or refunctionalized formations (cf. above on the shift subjunctive > future). ‘Pasts’ and ‘futures’ of the perfectum (i.e., pluperfect and future perfect, respectively) are actually absolute-relative tenses that indicate events preceding a reference point situated in the past or in the future. The verbal system of Sabellian languages is organized in a similar way.

Like Latin, Old Irish has a past tense (the preterite) that comprises continuations of IE s-aorists (-léicset ‘they left’) and perfects (-cúalae ‘he heard’ < *ku-klow-e), and a newly created future, while the Old Irish past habitual, called “imperfect,” seems to continue the middle-voice inflection of the PIE imperfect (cf. McCone, 1994, § 28.1—on the aspectual marker ro-, se section 3.13).

In Old Germanic languages, a binary tense distinction present ~ preterite is found, with the preterite forms—which are either new formations (for the ‘weak’ verbs) or continuations of the ancient perfect (for the ‘strong’ verbs)—not distinguishing between imperfective and perfective aspect.

In Baltic the tense-system comprises a present, a preterite with an -ā- or -ē-suffixed stem, and a future which contains a continuation of the IE future suffix *-sye/o- also testified by Indo-Iranian (Lith dúosiu ‘I will give’). Lithuanian also shows a newly created past habitual.

In the original Slavonic system (e.g., in Old Church Slavonic) the present is flanked by two past tenses: the aorist (perfective), whose stem often contains the continuation -x-/-š- of the original aorist suffix -s-, and the imperfect (imperfective), which shows a newly formed suffix -ěax/ěaš- or -aax/aaš-. The aspectual opposition between these two formations pertains to a different layer from that of the distinction between ‘perfective’ and ‘imperfective’ verbs (see section 3.13). Future time is expressed periphrastically or—for ‘perfective’ (telic) verbs—by the present (on the periphrastic perfect, see section 3.12).

In Tocharian, in addition to the present, two aspectually opposite past forms are found: the preterite and the imperfect, the latter probably continuing an original optative. Indeed, the use of the optative for expressing an imperfective (habitual) past is also a characteristic feature of several Iranian languages (already since the Old period).

The Old Armenian system is organized around an aspectual opposition between an imperfective and a perfective subsystem. The former comprises a present and a past (imperfect) indicative, the latter only an indicative with past time reference (aorist). Both subsystems have imperative and subjunctive forms.

The core system of the Albanian verb comprises a present, an imperfect, and an aorist (perfective past). A future is formed analytically by combining a preposed particle do with the forms of the subjunctive, which in turn is characterized by the particle të (e.g., do të punojë ‘he will work’). Homonymy between subjunctive and future excepting the different preposed particles is also found in Modern Greek (na désoume ‘bind.pfv.sbjv.1pl’ vs. tha désoume ‘bind.pfv.fut.1pl’).

Modal categories can be used as means of referring to future time. Thus, in both Tocharian and Old Armenian, future time reference is expressed by means of subjunctive forms (aorist-subjunctive in Old Armenian).

3.12 Periphrases With Past Participles and Split-Ergativity

Among the several periphrastic constructions developed in the various IE languages for expressing diathesis, tense-aspect or mood, the following patterns can be mentioned, which develop along the path resultative present > current relevance perfect > (perfective or recent) past.

(i) A periphrasis formed by a patient-oriented participle, typically a continuation of the PIE adjective in *-to-, either accompanied by ‘be’ or without copula (see Drinka, 2009, for a hypothesis about periphrases with -to-adjectives in PIE):

(a) the periphrasis belongs to the default voice, with shifting to ergative alignment:

(16)

The constructions in (16) are considered here as perpiphrastic because they are (copulaless) predicates with participles. In (16a) gahio (which continues the nominative of the OI verbal adjective/ past participle gṛhīta- ‘taken’) shows patient-agreement, whereas in (16b) the intransitive participle gao (cf. OI gata- ‘gone’) agrees with the subject.

(b) the periphrasis is restricted to transitive verbs and functions as a passive voice opposed to a synthetic active voice: e.g., Latin amātus sum ‘I was loved’, cf. the synthetic active transitive amāvī ‘I loved’ and intransitive appropinquāvī ‘I came near’ (not **appropinquātus sum). Consider also the following Gothic example, where qam is a synthetic active and dáupiþs was is a periphrastic passive:

(17) Gothic. Mark 1.9

(ii) A periphrasis formed by the same kind of patient-oriented participle as in the previous examples, but accompanied by ‘have’ (possibly in complementary distribution with ‘be’, with the latter restricted to intransitive or to unaccusative predicates):

(18) Old French. Chastelaine de Vergi 919–20 (cf. the ‘passé composé’ of Modern French)

(19) Old English. Beowulf 939–40 (cf. the Modern English perfect)

This kind of ‘have’ auxiliary is typical of the European area (cf. Drinka, 2017).

(iii) A periphrasis, typical of Slavonic, formed by an agent-oriented past participle (from a PIE adjective in *-lo-) either accompanied by ‘be’ or without copula: Old Church Slavonic perfect—e.g., bilŭ jesmĭ ‘strike.prf.ptcp.m.sg   be.prs.1sg = I have struck’; Russian past—e.g., ja bil ‘I strike.pst.m.sg = I struck’.

The pattern (i)(a) has a major consequence in the development of morphosyntax in that it determines the rising of a “split ergativity” system—with ergative-alignment restricted to the past—observed in the development of several Iranian and Indo-Aryan languages (Butt & Deo, 2017; Dahl & Stroński, 2016). Some languages show a further development of this type of system characterized by a secondary return to full accusative alignment. Thus, for example, New Persian bordam ‘I carried’ contains the continuation of Middle Persian burd < Proto-Indo-Iranian *bhr̥-ta- ‘carried’.

3.13 Preverbation and Prefixation

The same class of local particles/adverbs that served as the typical source of IE adpositions is also prone to combine with verbs to form multiword lexical units, and hence compound preverb-verb lexemes (cf. Kuryłowicz, 1964, pp. 171–178; Pinault, 1995; Watkins, 1964): e.g., Lat supplicō ‘beg humbly’, from sub ‘under’ + placo ‘placate’; Grk apodídōmi ‘give back’, from apó ‘away’ + dídōmi ‘give’.

In Vedic texts—which are archaic but also metrically constrained—‘compound verbs’ still behave as multiword items, allowing several syntactic positions. In the older hymns, the most frequent patterns for finite forms in non-negative independent clauses are:

(i) [_S pcl … v (…)], i.e., sentence-initial particle and tendentially sentence-final verb;

(ii) [_S…pcl v (…)], i.e., tendentially sentence-final sequence of particle plus verb.

Moreover, when the complex particle + verb appears sentence-initially, possible second-position clitics are positioned between the two parts of the multiword lexeme:

(iii) [_S pcl (clitic₁…) v (…)].

Semantic non-compositionality does not seem to entail syntactic contiguity (Pinault, 1995, pp. 47–48); compare the following example, where the discontinuous prati-pad- ‘respond, utter’ (or, according to a different interpretation, ‘undertake’) is formed from prati ‘toward, against’ and pad- ‘go, step’:

(20) Old Indic. RigVeda 10.114.9

Unbound preverbs are also found in Greek, in Homeric exameters (the so-called tmesis; cf. Chantraine, 1953, pp. 82ff.) and in other forms of archaic poetry (on Pindar, see Watkins, 2002):

(21) Homeric Greek. Iliad 1.98

Early Latin also shows the same pattern: compare the formula of prayer sub vos placo ‘I beg you.pl’ (see Cuzzolin, 1995).

The typical development of preverbed verbs might be described as follows: unbound component of a multiword item > first element of a compound verb > prefix (> unanalyzable phonological material). The term preverb is generally used both for items also appearing as autonomous words—as adverbs or adpositions—and for pure verb prefixes (cf. Booij & van Kemenade, 2003). However, positional properties and degree of grammaticalization do not necessarily go hand in hand. Thus, interposition of “infixed” pronouns between a preverb (possibly a highly grammaticalized one, such as ro-) and a verb in Old Irish reflects the original sentence-initial pattern particle-clitic-verb.

Preverbed verbs are often telic semantically (Haverling, 2010): cf. Lat suādeō ‘try to persuade’ ~ persuādeō ‘persuade’; Lat taceō ‘I am silent’ ~ conticeō ‘I stop talking’ (with per- and com- originally meaning ‘through’ and ‘with’, respectively); Goth þahan ~gaþahan ‘be silent’ ~ ‘stop talking’ (with ga- corresponding to Lat com-).

In Slavic the possibility of indicating the ‘completed’ versus ‘not-completed’ distinction became systematic. Thus, in Russian most verbs are [+telic] and prefixed and have a (further) suffixed counterpart with ‘not-completed’ value: e.g., za-pisat′/za-pis-yv-at′ ‘note down’. The two members of such pairs are traditionally considered to belong to a single lexeme and to be associated with the two values (perfective and imperfective, respectively) of the inflectional aspect feature (cf. Spencer, 2013, p. 89), but they are viewed by other scholars as conveying a derivational actional distinction (Bertinetto & Delfitto, 2000).

Inflectional prefixes can originate from original preverbs. In Old Irish, the preverb ro- (< PIE *pro- ‘forwards’) developed into an inflectional marker having resultative as well as prospective value (McCone, 1997, pp. 93–126). In German, the preverb ge- (cf. Goth ga-, probably from PIE *kom ‘with’) has been grammaticalized as a formative of the past participle (e.g., binden ‘bind’ → gebunden ‘bound’). In the evolution from Middle to New Persian, a local directional marker meaning ‘away’ developed into an inflectional mood marker, New Persian be-, which appears in imperative and subjunctive forms. In the same system, another inflectional prefix, mi-, deriving from the adverb hamē ‘always’, signals the tense-aspect categories of present and imperfect (Windfuhr, 1979, pp. 83–97).

Preverbs are a major source of derivational prefixes in the history of IE languages (Cowgill, 1966, p. 138). In addition to appearing in nominal and adjectival derivatives of preverbed verbs, prefixes of preverbal origin are, in some cases, directly extended to other word classes than verbs: e.g., OHG ur-alt ‘very old’ (with ur- originally meaning ‘out of’, cf. Goth us-), Lat per-bonus ‘very good’.

3.14 Some Other Developments Involving Word Formation

A common development in the history of word formation in the IE family is the decline of the root-based organization of morphology and the average increase of the categoriality of lexical bases—that is, their being or tending to be uniquely associated with a given word class (cf. Lehmann, 2008).

Besides a combination of old suffixes and phenomena of reanalysis, loss of syntactic autonomy by second members of compounds is a relevant source of new suffixes in several IE languages, among which the Germanic (e.g., Ger offenbar ‘apparent’, with -bar < West Germanic *-bāri- ‘carrying’, Krahe & Meid, 1967, pp. 218ff.) and Persian (e.g., Pers amuzgar ‘teacher’, with -gar < Iranian *-kara- ‘doer’).

In the modern IE languages the stratified nature of lexical systems, which often comprise a considerable amount of loanwords and of learned words of Latin or Greek origin, interacts with derivational morphology. Thus, as with lexemes, Romance languages may have doublets of derivational affixes of Latin origin comprising an item having an uninterrupted tradition and a learned one (e.g., French natur-el ‘natural’ ~ front-al ‘frontal’, cf. Lat -alis). In English the ‘latinate’ lexicon of Norman French or Latin origin exhibits specific properties in word formation (cf. Aronoff, 1976).

As for the East, in the Persian lexicon, imbued with Arabic loanwords, a typologically non-trivial formation such as the arabic intensive noun pattern CaCCāC seems to have had some sporadic productivity (e.g., kaffāš ‘shoemaker’ from Persian kafš ‘shoe’—Ṣādeqī, 1986), wheareas in Hindi-Urdu it is possible to find affixes of Persian (cf. the privative prefix be-) or Perso-Arabic (cf. the negative prefix lā-) origin.

As for compounding, modern Germanic languages such as German and English exhibit a remarkable productivity of this word-formation device, with recursion being also allowed. A noteworthy characteristic of Modern Greek consists of the use of verbal compounds with right-headed Adverb-Verb or coordinative Verb-Verb structure (e.g., kalo-gnorízi ‘knows well’, trogo-píni ‘eats and drinks’).

A noticeable drift, found in New Persian and Hindi-Urdu, lies in the formation of lexical systems in which a significant amount of verbal entries consists of a lexical (often nominal) element plus a grammatical support verb (or ‘light verb’): e.g., Pers tarǰome kard ‘translation made’ → ‘translated’, tarǰome šod ‘translation became’ → ‘was translated’ (Windfuhr, 1979, pp. 113ff.); Urdu ijāzat dēnā ‘permission give’ → ‘permit’. Given that verbs are generally lesser prone than nouns to be borrowed (cf. Wohlgemuth, 2009), such constructions favor the introduction of loanwords into the verbal lexicon (see the periphrases quoted immediately above, where the elements tarǰome ‘translation’ and ijāzat ‘permission’ are of Arabic and Perso-Arabic origin, respectively).