Obligatory features of Lithuanian verbal inflection classes

Lithuanian verbal inflection classes are defined by tense suffixes, stem‐forming affixes, vowel and consonant alternations, and accentuation pat‐ terns. I make a distinction between obligatory features that are relevant for every verb and non‐obligatory features that characterize only part of the verbs. I argue that the obligatory features are the present and the past tense suffixes combined with mobile and immobile accentuation patterns, while the rest of the features are optional. When only the obligatory features are taken into ac‐ count, three types of the present tense (‐a‐, ‐i‐, ‐o‐) and two types of the past tense (‐ė‐, ‐o‐) suffixes are found in five combinations (‐a‐/‐ė‐, ‐a‐/‐o‐, ‐i‐/‐o‐, ‐o‐/‐ė‐, ‐o‐/‐o‐) with further variants defined by two types of mobile and one type of immobile accentuation, resulting in eighteen suffixal‐accentual combi‐ nations in standard Lithuanian. The combinations of features characterizing the present and the past stems support the view of inflection classes as classes of stems rather than of lexemes (Stump 2016).


Introduction
Lithuanian verbs express values of morphosemantic and morphosyntactic features 1 by a range of prosodic, morphophonological, morphological, and syntactic devices. The expression of some of the features varies from lexeme to lexeme, providing a basis for distinguishing verbal inflection classes (ICs), 269 traditionally also termed conjugations 2 . As will be shown below, Lithuanian verbal ICs are defined by tense suffixes, stem-forming affixes, vowel and consonant alternations, and accentuation patterns. These markings will be referred to as IC features (Corbett & Baerman 2006, 235-239, Corbett 2012.
I propose that obligatory and non-obligatory features of ICs can be distinguished, and discuss them in the following order. Section 2 provides a general overview of IC features and briefly presents the traditional classification of Lithuanian verbal ICs. Section 3 focuses on present and past tense suffixes as obligatory features and treats their combinations as the obligatory suffixal basis of ICs. Section 4 discusses mobile and immobile accentuation patterns as another obligatory feature, and describes combinations of present and past tense suffixes with accentuation patterns resulting in suffixal-accentual ICs. Section 5 provides a brief recapitulation of the main proposals.
I would like to note that the present article would have been impossible had I not met Axel Holvoet many years ago. I have always been fascinated by his own ability and continuous encouragement to colleagues to take a fresh look at the Baltic data and to search for novel solutions. I would like to offer this text as a thank-you note to Axel for his inspiration.

Features of verbal inflection classes
As previously mentioned, Lithuanian verbal ICs are defined by tense suffixes, stem-forming affixes, vowel and consonant alternations, and accentuation patterns. I will focus on tense suffixes and accentuation patterns because these will be argued to be obligatory features of ICs, while other phenomena will be discussed briefly due to their non-obligatory character. With respect to the formal nature of the features, we will be dealing with the following three types: affixal (tense suffixes and stem-forming affixes), morphophonological (vowel and consonant alternations), and prosodic (accentuation patterns).

Tense suffixes
Lithuanian has synthetic present, past, past habitual, and future tense forms. Suffixes of past habitual and future are uniform for all verbs, -dav-o-3 and -s(i)-/-sia-4 respectively. The suffixes of present and past, however, vary across verbs. Present tense suffixes are -a-, -i-, -o-and past tense suffixes are -ė-, -o-, exemplified in Table 1  I will refer further to these paradigms as inflection types of the present tense (PRS) and the past tense (PST), for example, "PRS a-type", "PST o-type" or simply "a-type", "o-type", etc. I treat PRS and PST types as obligatory affixal IC features since there are no verbs for which the choice of these types would be irrelevant: each and every non-defective Lithuanian verb needs to be inflected according to one of the three PRS types and according to one of the two PST types. I limit myself to finite forms in this paper, but I should mention that the formation of PRS active (inflecting and non-inflecting) and passive participles also employs these suffixes; PST active participles, however, do not reflect the suffixal difference between ė-and o-types, while PST passive participles are interpreted as based on the infinitive stem or their own stem is distinguished. The fact that PRS and PST paradigms may be expressed by the same markings (PRS o-type and PST o-type) will be addressed in Section 3.
A few explanatory remarks with respect to segmentation of PRS and PST tense suffixes are in order. Subject person suffixes in Lithuanian are 1SG -u, 2SG -i, 1PL -me, 2PL -te, while 3SG/PL bears no dedicated marker 9 . PRS suffixes -a-and -i-cannot be segmented in PRS.1SG/2SG, and as a result, suffixes PRS.1SG -u and PRS.2SG -i in these paradigms cumulatively express person and tense. PRS and PST suffix -o-can be seen as represented by allomorph -a-in 1SG/2SG if segmentation of diphthongal -au and -ai as -a-u and -a-i is accepted; alternatively, -au and -ai can be recognized as cumulative exponents of tense and person just as in PRS a-and i-types. In a similar fashion, PST suffix -ė-may be seen as having allomorph -a-in 1SG (-a-u) 10 and -e-in 2SG (-e-i); alternatively, -au and -ei can be interpreted as cumulatively expressing person and tense. A recurring pattern is evident: PRS and PST suffixes are easily segmented in 1PL, 2PL, and 3SG/PL, while 1SG and 2SG are problematic: there is no segmentable suffix (a-and i-type) and a potentially segmentable suffix (ė-and o-types). For the sake of simplicity, I usually refer to suffixes representing types in Table 1 as tense suffixes, but in 1/2SG forms they actually are or can be interpreted as cumulative tense-person suffixes. When 1/2SG 10 In Ambrazas (1997,311) this allomorph is presented as -e-(< -ė-), but I attempt to treat it as -abecause vowel [a] explains affrication in PST.1SG (I assume that the cell of 1SG of PST ė-type bears a feature of palatalization(-affrication) and non-front vowel is a condition for it to be realized), e.g. PST.1SG meči-au 'I threw', vedži-au 'I led' alongside PST.2SG met-ei 'you threw ', ved-ei 'you led' (roots met-, ved-). If -e-is assumed, stops [t] and [d] would not alternate with affricates [ʧ j ], [ʤ j ] (spelled as <či>, <dži>) before the front vowels, see also Section 2.2. So when -e-(< -ė-) is assumed, an additional morphophonological rule is needed to convert it to -a-with palatalization(-affrication) of the preceding consonant (Andronov 2000, 39); in Ambrazas (1997, 311) this step is not explicitly stated and only a comment with regard to spelling of [ɛʊ] as <iau> after palatalized consonants is made.
are in the focus of the discussion I refer to the corresponding morphemes as tense-person suffixes. Now let us address some properties of PRS and PST types within the framework of canonical typology (Corbett 2009). As shown below (Section 3), PRS and PST types are building blocks of verbal ICs and support the view that ICs can be seen as classes of stems rather than of lexemes (Stump 2016, 92-95). I will first discuss canonicity of PRS and PST types as separate ICs and later interpret them as sets of ICs in Section 3.
According to Principle I, "[c]anonical inflectional classes are fully comparable and are distinguished as clearly as is possible" (Corbett 2009, 3). According to Criterion 1 of this principle 11 , "[i]n the canonical situation, forms differ as consistently as possible across inflectional classes, cell by cell" (Corbett 2009, 4). Among finite indicative forms, a non-contrasting cell is found in PRS.2SG of a-and i-type (suffix -i in both types). PRS.1SG of a-and i-types has the same suffix -u, but this cell is contrasted by non-palatalized last consonant of the stem (a-type) vs. palatalized(-affricated) last consonant of the stem (i-type); the same holds for PST.1SG of ė-and o-types. However, we should bear in mind that the PRS a-type also has a palatalized(-affricated) variety (see Section 2.2 below) where the expression of PRS.1SG is the same as that of i-type, e.g. palatalized a-type PRS.1SG verki-u 'I cry' = i-type PRS.1SG tiki-u 'I believe'. The fact that all cells (of finite indicative forms) of the PRS o-type equal the PST o-type will be addressed in Section 3.
According to Criterion 2, "[c]anonical inflectional classes realize the same morphosyntactic or morphosemantic distinctions (they are of the same structure)" (Corbett 2009, 4). In this respect all cells, except for 3SG/PL, will usually be lacking for impersonal verbs, but this property is not bound up with any particular IC (i.e. type of PRS/PST). According to Criterion 3, "[w]ithin a canonical inflectional class each member behaves identically" and "this implies that there are no stem differences, alternants or other subclasses" (Corbett 2009, 4). The majority of PRS/PST types would be non-canonical in this respect: PRS a-type, PST ė-and o-types have subtypes defined by additional affixation and alternations (see some notes in Section 2.2), rare 11 All criteria discussed here belong to Principle I. See also a discussion of Criteria 1 to 4 as simple deviations from the canonical ideal in Stump (2015, 116-119); complex deviations are discussed in Stump (2015, 119-123) and addressed in Section 3 of the present paper.
subtypes can be also found in PRS i-type (see footnote 13 below), so only PRS o-type would be strictly canonical according to Criterion 3. According to Criterion 4, "[w]ithin a canonical inflectional class each paradigm cell is of equal status"; as a result, "[i]n the canonical situation, where all forms are distinct between classes, the form for each cell predicts all the others within a class" (Corbett 2009, 5). This criterion is tightly knit with the first one and non-contrasting cells mentioned above will be unable to predict IC membership unambiguously. Further deviations from the canonical ideal will be addressed in Section 3 where combinations of PRS and PST types are discussed.

Stem-forming affixes and alternations
Other affixal features of Lithuanian verbal ICs pertain to the formation of particular stems. These affixes are of non-obligatory nature and are relevant only for part of the verbal lexicon: Lithuanian verbal stems may, but need not contain any of these affixes.
The inflectional system of the Lithuanian verb is traditionally described as based on three stems referred to as infinitive (INF) stem, PRS stem, and PST stem 12 . In addition to that, a past passive participle (PPP) stem can be distinguished due to its prosodic differences (accent realization and mobility) from the INF stem (Arkadiev 2012, 19-21); stem-forming affixes of the PPP stem are the same as the ones of the INF stem, and for the sake of simplicity I will speak only of the INF stem below. Now let us review the optional affixes found in PRS, PST, and INF stems. The PRS stem may contain the infix -n-and the suffixes -n-, -st-which are found only in combination with the PRS a-type and constitute subtypes of it 13 ,

e.g. PRS.3 ti-n-k-a 'fit(s)' (root tik-), gau-n-a 'get(s)', tirp-st-a 'melt(s)'. The PST stem may contain suffixes -ė-([eː]) and -o-([oː]) which co-occur with the
tek-a 'flow(s)', žin-o 'know(s)' (both suffixes are represented by -ėj-, -oj-in antevocalic position and no corresponding suffixes are found in the PRS stem). If stem-forming suffixes -ė-, -o-are used in the PST stem, the same suffixes will always be present also in the INF stem (represented by anteconsonantal -ė-, In addition to that the INF stem may contain the suffix -y-([iː]) which is found only in the INF stem and absent both in the PRS and PST stems, e.g.
Lithuanian verbal stems can also be optionally differentiated by morphophonological vowel and consonant alternations 14 . Vowel alternations have various types and for the sake of brevity I will present just some of them; see more data in Ambrazas (1997, 287-290). For example, the root vowel of INF and PST stems can be contrasted with that of PRS stem, e.g.
). Of the consonant alternations, relevant for the present study is the contrast between non-palatalized and palatalized consonants and between non-palatalized consonants and palatalized affricates (only stops [t] and [d] alternate with affricates [tʃ j ] and [dʒ j ]). Palatalization (but not affrication) before front vowels is automatic and only the position before non-front vowels is morphologically relevant. The said alternation creates two subtypes of the PRS a-type, e.g. PRS.1SG verki-u 'I cry' (<ki> = [k j ]), PRS.1SG lieči-u 'I touch' (<či> = [tʃ j ]) (palatalized(-affricated) variety of PRS a-type) vs. PRS.1SG suk-u 'I turn', met-u 'I throw' (non-palatalized variety of PRS a-type). It should be noted that the 1SG cell of PRS i-type and PST ė-type is always affected by the said alternation and there are no non-palatalized variants as there are in the case of PRS a-type, e.g.

Accentuation patterns
Lithuanian has a system of free stress and similarly to affixal and morphophonological features, one needs to recognize the role of stems which, as will be seen below and in Section 4, are assigned certain accentuation patterns. In this respect Lithuanian shows that prosodic features of ICs may have scope not only on whole lexemes, but also on lower-level morphological features (Corbett & Baerman 2006, 237, Corbett 2012, such as stems.
There are two main accentual patterns found in inflectional paradigms of Lithuanian finite verbal forms: immobile and mobile. In the immobile pattern, the stress largely remains on the same syllable throughout the paradigm of a given stem, whereas in the mobile pattern, the stress lands on different syllables. In what follows, I limit myself to the discussion of productive finite forms for the sake of brevity 16 .
Finite forms based on the INF stem always assume the immobile pattern, and the stress remains on the same lexically determined syllable 17 , e.g. FUT.1SG bėǵsiu 'I will run', kèpsiu 'I will bake', FUT.2SG bėǵsi 'you will run', kèpsi 'you will bake', etc.
This demonstrates that the mobile pattern of PRS/PST actually has two subtypes: the one where stress moves from the lexically determined syllable to the prefix or to the 1/2SG suffix (when the prefix is absent) and the one where the stress moves only to the 1/2SG suffix. The assignment of stress patterns is obligatory, and each and every verb needs to be accentuated according to a certain set of PRS/PST patterns. Each PRS/PST stem type in theory can be mobile (with two subtypes) and immobile, and all attested combinations of stem types and accentuation patterns will be discussed in Section 4.
In addition to that, verbal stems may differ in respect to which syllable (morpheme) the stress is lexically assigned to. That syllable may be the same (as in INF bėǵ-ti 'run', PRS.3 bėǵ-a, PST.3 bėǵ-o, the stress is always on the root) or different (as in INF myl-ė-ti 'love', PRS.3 mýl-i, PST.3 myl-ėj-o; the stress is on the suffix -ė-in INF/PST and on the root in PRS). As I focus on interaction of PRS and PST stems with mobile/immobile patterns, I do not discuss the stress placement variation seen across stems in this article.
It should be also noted that I do not make reference to tonal properties of syllables that carry lexically assigned stress (and appear directly before tense-person suffixes) as it is customary to do in the traditional approach, see Section 2.4. Due to the common neutralization of tonal oppositions in syllables with long vowels and diphthongs [iɛ] and [uɔ] 18 , I prefer to speak of simply mobile and immobile patterns without reference to tonal properties of the relevant syllables. I acknowledge, however, that the (im)mobility of stress can be seen as motivated in part of the stems by tonal properties of syllables that have short vowels, diphthongs other than [iɛ] and [uɔ], and tautosyllabic VR sequences, where tonal properties are mostly well distinguished.

Traditional classification
According to the traditional classification, three conjugations are distinguished on the basis of PRS suffixes -a-, -i-, and -o-, termed thematic vowels (Ambrazas 1997, 298). The three conjugations are further subdivided into four groups in the first conjugation and into two groups in the third conjugation, while the second conjugation has no groups, see Table 3 below. The groups are mostly distinguished on the basis of PRS and PST types and certain structural types (see below); Group 3 and 4 of the first conjugation differ from Group 1 and 2 in consonant alternations of PRS stem (palatalized(-affricated) vs. non-palatalized). Groups 1 to 4 of the first conjugation have further subgroups based on other alternations, stem-forming affixation, the use of irregular stems, and derivational features not reflected in Table 3.
The traditional classification also makes reference to the following structural types of the verbs: primary, mixed, and suffixal, see the fourth row of Table 3. These types are defined in Ambrazas (1997, 285); in what follows, I partly reformulate the definitions. The primary verbs have no stem-forming suffixes in their INF stem, but they may contain stem-forming affixes in their PRS stem and can display root vowel alternations. The mixed verbs have stem-forming suffixes -ė-, -y-, -o-in their INF stem, of which the suffixes -ė-, -o-are also present in the PST stem; these suffixes are always absent in the PRS stem which structurally equals that of the primary verbs (without any additional affixes), hence the term "mixed", i.e. the verb is structurally suffixal and primary at the same time. The suffixal verbs contain suffixes in all three stems; these suffixes synchronically are or diachronically were derivational, e.g.  In addition to Ambrazas (1997), I would like to mention a few other approaches to Lithuanian verbal ICs. The classification of Mathiassen (1996, 95-96) is based on the PRS stem (-a-, -i-, -o-). It is noted that a-type has palatalized and non-palatalized varieties and predictability of the IC on the basis of INF is discussed. Further subtypes of PRS stem are found in the section on PRS forms and the relation between conjugations (i.e. PRS types) and PST types are discussed in the section on PST forms (Mathiassen 1996, 97-100, 103-105). In the classification of Andronov (2000, 43) structural types are taken as a prima-ry classificatory criterion and the mixed type has two varieties where INF stem differs from PRS/PST by having a suffix -y-or INF/PST stem differs from PRS by having a suffix -ė-or -o-. Further varieties are defined by combination of PRS types (-a-, -i-, -o-) with the PST types (-ė-, -o-) yielding 7 classes in total (combination PRS/PST -a-/-o-appears three times) with further potential distinctions defined by alternations and additional affixes ("submorphs") of PRS stem. The idea to take obligatory features as a point of departure (Andronov 2000, 42-43) and the combination of PRS/PST types coincides with the approach advocated in the present paper. I would like to note, however, that I consider only obligatory features properties of specific stems (PRS/PST types and accentuation patterns) and see the structural types as being of different nature and characterizing not the specific stems (INF, PRS, PST), but their relations 19 . In this respect my approach is closer to that of Arkadiev (2017, 9) where first level parameters of classification are PRS types (-i-, -o-, -a-with certain subtypes) and PST types (the latter ones, however, are defined differently 20 ).
The accentuation patterns of PRS and PST tense finite paradigms are discussed in, e.g., Ambrazas (1997, 309, 311). The traditional description makes reference to prosodic features of the syllable to which the stress is lexically assigned and the relation of these features to stress mobility. Possible combinations of accentuation patterns and conjugations are not reviewed. The description of Ambrazas (1997) lacks the discussion of accentuation of prefixal forms in contrast to Mathiassen (1996, 97, 101-103, 105-106) where accentuation of forms with and without prefixes is well integrated into the description of formation of PRS/PST. Andronov (2000) and Arkadiev (2017) focus on segmental features and morphophonological alternations and do not discuss accentuation patterns.

Combinations of present and past tense suffixes
As mentioned earlier, I propose that tense suffixes are obligatory features and constitute the basis of the IC system together with accentuation patterns. The traditional approach favors the PRS suffixes, but there is no reason why these suffixes should be given priority. I suggest that the obligatory affixal basis of ICs of Lithuanian verbs can be seen as combinations of PRS and PST types, cf. implementation of this idea in earlier treatments in Section 2.4.
There are three types of PRS (-a-, -i-, -o-) and two types of PST (-ė-, -o-), which yield six theoretical combinations. Out of them, five combinations are attested in standard Lithuanian, as illustrated in Table 4

Combinations of PRS and PST types in standard Lithuanian
According to this interpretation, instead of three conjugations that (on the first level of classification) are distinguished on the basis of PRS types (-a-, -i-, -o-), one could recognize five combinations of PRS and PST types (-a-/-ė-, -a-/-o-, -i-/-o-, -o-/-o-). It is evident that Lithuanian verbal ICs support the idea that ICs can be interpreted as classes of stems and not of lexemes (Stump 2016, 92-95); in this case an inflectional profile of a given lexeme is seen as a combination of ICs of its stems.
We should bear in mind that the combinations mentioned in Table 4 include only the obligatory suffixal bases of ICs and are not yet fully specified ICs without the values of another obligatory feature-the accentuation pattern, see Section 4. I should also mention that further subtypes of ICs would be distinguished according to non-obligatory alternations and stem-forming affixes, as is done in, e.g., Ambrazas (1997, 298-307), Arkadiev (2017), and would be mostly pertinent to the combinations -a-/-ė and -a-/-o-, because taking into account additional features-the stem-forming suffixes-would not yield additional subtypes of the -i- /-ė-, -o-/-ė-, -o-/-o-combinations 22 .
In Section 2.1 we began discussing Lithuanian verbal ICs within the framework of canonical typology and reviewed simple deviations from the canonical ideal. One of the complex deviations 23 seen in Lithuanian is that its verb exhibits modularity comparable to that of Classical Sanskrit and similar systems. The present system of forms in Classical Sanskrit follows one of the ten conjugations, while the aorist system can be formed according to one of the seven conjugations 24 , and such organization is referred to as segregated inflection classes while their sets (combinations) are termed segregated sets (Stump 2015, 119-120), see also Finkel & Stump (2007), Stump (2016, 90, 202-203). The segregated sets presented in Table 4 above in many cases show deviations from the canonical ideal according to Criterion 1 (see Section 2. 1 and Stump (2015, 120)); for example, a lexeme having PRS/PST set -a-/-ė will differ from another lexeme characterized by set -a-/-o-only in the cells of PST. Here I consider only the case when there are no further differences found due to additional (non-obligatory) stem-forming affixes and alternations 25 , for example, cf. kepti and bėgti in Table 4. In contrast, the inflection according to set -i-/-o-in theory could fully coincide with that of -o-/-o-set in PST, but this is not what we find: the lexemes with PRS i-type always have stem-forming affix -ėj-in PST while the lexemes with PRS o-type (and PST o-type) always feature stem-forming affix -oj-in PST, cf. PST.3 myl-ėj-o 'loved' vs. saug-oj-o 'protected' in Table 4. The cells of PRS of the sets -o-/-ė-and -o-/-o-, however, will always have exactly the same exponence. Criterion 4 (see Section 2. 1 and Stump (2015, 120)) is also violated in a majority of cases: only PRS i-type cells (except for 1/2SG mentioned in Section 2.1) predict PST o-type (combined with stem-forming affix -ėj-).
Another deviation from the canonical ideal is when the same exponence is found in distinct sectors, such as overlapping morphology of Sanskrit imperfect forms in present-system sixth conjugation and aorist forms in thematic aorist conjugation; this case is termed metaconjugation (Stump 2015, 120-121, Stump 2016. In Table 4 metaconjugation is exemplified by Lithuanian o-type tense suffixes found both in PRS and PST stems 26 . Metaconjugation violates not only Criterion 1 and Criterion 4 (as a case of segregated sets discussed above), but also Criterion 3 as observed by Stump (2015, 120-121): o-type membership defines PRS exponence for some lexemes and PST exponence for others, i.e. the members of this class do not behave the same way.

Combinations of tense suffixes and accentuation patterns
Each set of PRS and PST types discussed in Section 3 theoretically may appear in four combinations with accentual patterns, for example, PRS may be mobile, but PST may be immobile, both PRS and PST may be mobile, etc. In Section 2.3 we saw that there are two varieties of mobility, prefixal-suffixal mobility (the stress moves to the suffix of 1/2SG, but if a prefix is present, the stress moves to the prefix) and suffixal mobility (the stress moves only to the suffix of 1/2SG). For the sake of simplicity, let us first take a look at possible combinations of just mobile (mob) and immobile (immob) patterns in (1). The two varieties of mobility will be taken into account a bit later.
(1) PRS mob PRS immob PRS mob PRS immob PST mob PST immob PST immob PST mob The four combinations presented in (1) theoretically may co-occur with five PRS/PST sets discussed in Section 3, yielding 20 sets in total. Out of them, 14 are attested in standard Lithuanian and will be discussed below.
Both -a-/-ė-and -a-/-o-tense suffix sets are found in all four combinations with accentuation patterns, see examples in Table 5 and 6 (note that some examples here and below contain additional features of stems, such as stem-form-ing affixes and vowel alternations; these features enable mixed models, i.e. immob combined with mob or vice versa, see notes below Table 9).
PRS.1SG a-mob: kep-ù a-immob: ėd-u a-mob: keli-ù a-immob: gìmst-u PST.1SG ė-mob: kepi-aũė-immob: ėdži-au ė-immob: kėĺi-au ė-mob: gimi-auĨ NF kèpti 'bake' ėśti 'eat, devour' kélti 'raise' gìmti 'be born'   Non-attested patterns in Table 7 and 8 can be explained as follows. Mobile pattern of PST o-type is not found due to the stem-forming suffix -ėj-which is always present when PRS i-type is paired with PST o-type. The same is true of the stem-forming suffix -oj-which is always found when PRS o-type is paired with PST o-type (PRS o-type can be also paired with PST ė-type, see below). The PST stems with stem-forming suffixes -ėj-and -oj-are always immobile 27 , e.g.  with the stress on the root, etc.
The set of tense suffixes -o-/-ė-is found in two combinations with accentuation patterns, see Table 9  Compared to combinations discussed above, the accentuation of the set -o-/-ė-is always uniform: either it is mobile, or immobile; mixed patterns are not attested. This can be explained by taking a look at the types which have mixed patterns enabled by certain differences between PRS and PST stems: such differences are never found in the -o-/-ė-set. For example (see Table 5 and 6 above), PRS.1SG gìm-st-u 'I am being born', sén-st-u 'I grow old' (immobile) contain the stem-forming suffix -st-which is absent in the PST stem: PST.1SG gimi-aũ 'I was born', sen-aũ 'I was growing old' (mobile); another possibility is root vowel apophony, as in PRS.1SG keli-ù 'I raise' (mobile) vs. PST.1SG kėĺi-au 'I raised' (immobile). In both cases short vowel of the first syllable (PST.1SG gi.miaũ 'I was born', PRS.1SG ke.liù 'I raise') precludes the possibility of the immobile paradigm (long vowel or syllabic VR sequence is required). Mixed accentual patterns are also possible if stem-forming suffixes -ėj-or -oj-are found in PST (immobile pattern) alongside PRS which lacks these suffixes and may have mobile pattern, see notes on sets -i-/-o-and -o-/-oabove; consider also some verbs of the -a-/-o-set: PRS.1SG dreb-ù 'I tremble' (mobile) vs. dreb-ėj-au 'I trembled' (immobile, contains stem-forming suffix -ėj-), PRS.1SG mieg-ù 'I sleep' (mobile) vs. mieg-ój-au 'I slept' (immobile, contains stem-forming suffix -oj-), etc. It should be noted that in order to explain unattested (or some attested) combinations here and earlier I had to venture into the zone of non-obligatory features such as root vowel alternations and stem-forming suffixes.
Let us sum-up the results of attested PRS/PST sets with mobile and immobile accentuation patterns in Table 10: Total 14 Now it is time to review additional variation when two subtypes of mobility are taken into account: prefixal-suffixal (mob prf-suff ) and suffixal (mob suff ), see their discussion in Section 2.3. With two varieties of mobility (and one pattern of immobility) we get nine theoretical combinations, listed in (2). When combined with five PRS/PST sets they could yield 45 possible variations, but we will soon see that only 18 are attested.
(2) PRS mob prf We only need to review the cells where the mobile pattern is found, and to save space abbreviated versions of the tables used above can be reproduced. Now the classification will be more detailed, because instead of simply noting that the accentuation pattern is mobile we have information on the particular version of mobility in each combination: prefixal-suffixal or just suffixal. For example, in Table 5 the forms PRS.1SG kep-ù 'I bake', PST.1SG kepi-aũ 'I baked' illustrated combinations a-mob and ė-mob correspondingly; these combinations in Table 11 are now specified as a-mob prf-suff (kep-ù 'I bake', nè-kepu 'I do not bake') and ė-mob prf-suff (kepi-aũ 'I baked', nè-kepi-au 'I did not bake'). In addition to that, a new combination was discovered due to distinction of two types of mobility: a-mob suff (PRS.1SG šauki-ù 'I cry', ne-šauki-ù 'I do not cry' 28 ) bound up with ė-mob prf-suff (PRS.1SG šauki-aũ 'I cried', nè-šauki-au 'I did not cry') alongside a-mob prf-suff with ė-mob prf-suff (previous example). PRS PST.1SG ė-mob prf-suff kepi-aũ nè-kepi-au šauki-aũ nè-šauki-au ė-mob suff n/a ė-mob prf-suff gimi-aũ nè-gimi-au gìmti 'be born' kélti 'raise'  Table 13). Variability of mobility (i.e. both options: mob prf-suff and mob suff ) is never found in PST. PST ė-type can be mob prf-suff or mob suff , but never in the same combination with a given PRS type: mob prf-suff is found with PRS a-type, mob suff occurs with PRS o-type. PRS/PST o-type, when it is mobile, is always mob suff . See further com-PRS.1SG a-mob prf-suff suk-ù nè-suk-u a-mob suff vargst-ù ne-vargst-ù (a-immob) a-mob prf-suff dreb-ù nè-dreb-u a-mob suff ried-ù ne-ried-ù  Andronovas (1995) on the relation of mobility to PRS/PST types and other crucial parameters such as root-syllable length. I would like to remind readers at this point that the present study aims at a classification based only on the obligatory features. As mentioned earlier, for a full classification, all non-obligatory features (stem-forming affixes and alternations) need to be taken into account, cf. Ambrazas (1997), Arkadiev (2017). The combinability with accentuation patterns, in turn, would have to be listed and discussed partially anew, because the current approach is blind to additional properties of the stems. As an exception, only some remarks on combinations with accentuation patterns were made when non-obligatory features such as root-vowel alternation and some stem-forming suffixes were involved.

Conclusions
Lithuanian verbal inflection classes are defined by tense suffixes, stem-forming affixes, vowel and consonant alternations, and accentuation patterns. Only tense suffixes and accentuation patterns are obligatory features relevant for each and every verb; the rest are of non-obligatory nature and pertinent only to part of the verbal lexicon.
There are three types of the present tense (-a-, -i-, -o-) and two types of the past tense (-ė-, -o-) suffixes. These types are attested in five combinations in standard Lithuanian (-a-/-ė-, -a- /-o-, -i-/-o-, -o-/-ė-, -o-/-o-) and further variation is defined by one immobile pattern and two mobile patterns of accentuation, resulting in eighteen suffixal-accentual combinations. A full picture of all possible combinations of IC features would be arrived at by taking into account non-obligatory features as well.
The combinations of features characterizing present and past stems in Lithuanian support the view that inflection classes can be understood as classes of stems rather than of lexemes (Stump 2016). The inflectional profile of a given lexeme is, as a result, a combination of inflectional classes of its stems.