To Anna Wierzbicka, a closest friend for 55 years

Przyjaźń jest rzeczą diabelnie trudną ‘Friendship is a devilishly difficult thing’ [Wierzbicka 1971: 83].

Yes, Anna, generally speaking, this is so; but with you friendship is the easiest thing!

1. The problem stated

This paper constitutes a natural continuation of [Mel’čuk 2018], where six surface-syntactic relations necessary for the description of N→N_GEN phrases in Russian were proposed. Here we will consider the linear ordering of genitive nouns N_GEN-1, N_GEN-2, N_GEN-3, … cosubordinated to the same noun N in Russian. (In fact what is being ordered are the whole phrases headed by these N_GENs.)

Example (1) shows that this order is not free:

The subsequent discussion is framed in terms of the Meaning-Text approach: it is strictly synthetic (from meaning to text) and uses a dependency representation of the syntactic structures of sentences and phrases. The following three points must be taken into account in order to avoid misunderstandings.

• Our examples illustrate the surface implementation of the given surface-syntactic relations [SSyntRels] (i.e., “with these SSyntRels the linear arrangement of the given cosubordinated N_GENs is so and so”); if the resulting expression is ambiguous — in that it can express something else as well, this should not be paid attention to. The only thing that is important is the correct expression of the starting meaning.
• Our examples are based on our own linguistic intuition, which on several occasions can diverge from other speakers’ judgments. However, our main thrust is not establishing facts of Contemporary Russian, but formulating implications of the form “If the given linguistic expression X is correct / incorrect, then the order of N_GENs is so and so.”
• Once again, there can be disagreement between speakers concerning our evaluation of a given phrase: good, bad (*), hardly acceptable (^??), or questionable (^?). But what really matters is the difference between two arrangements of N_GENs: one is worse than the other, and this is sufficient for our statements.

NB: The relevant notions and formalisms cannot be explained here, and the reader is invited to consult [Mel’čuk 2012–2015]. All glosses in the examples are literal; two English words that correspond to one Russian form are united by a dot: Rus. boli ‘of.pain’.

Schematically, we are interested in correspondence (2):

In prose, we propose some surface-syntactic [SSynt-]rules that establish the correspondence between an SSynt-subtree and the deep-morphological [DMorph-]string implementing it. The SSynt-subtree under consideration has three properties:

1) It is headed by a noun N on which several genitive nouns N_GEN-1, N_GEN-2, N_GEN-3, … syntactically depend (each with its own dependents, if any).

2) The SSynt-relations r_i that subordinate these N_GENs to N impose the genitive case on them (in the DMorph-string). These SSynt-relations are six in number [Mel’čuk 2018]:

N–subjective-adnominal→N_GEN-subj

priezd–subj-adnom→otca ‘coming of.father’, stakan–subj-adnom→vody ‘glass of.water’

N_GEN-subj expresses N’s deep-syntactic [DSynt-]actant I.

N–objective-adnominal→NGEN-obj

osvoboždenie–obj-adnom→otca ‘liberation of.father’

portret–obj-adnom→rebënka ‘portrait of.child’

N_GEN-obj expresses N’s DSynt-actant II.

N–qualificative-adnominal→N_GEN-qual

balka–[nedostatočnoj]–qual-adnom→dliny ‘beam [of.insufficient] length’

N_GEN-qual denotes a predicate whose Sem-actant 1 or 2 is expressed by N (‘dlina / length–1→balka / beam’: balka dvuxmetrovoj dliny ‘beam of.two.meter length’; ‘mečta / dream–2→devuška / girl’: devuška moej mečty ‘girl of.my dream’). In Russian, an N_GEN-qual must necessarily have a syntactic dependent, normally an adjective, which expresses the other Semactant.

N–attributive-adnominal→N_GEN-attr

krik–attr-adnom→boli ‘scream of.pain’

živopisʹ–attr-adnom→Vozroždenija ‘painting of.Renaissance’

N_GEN-attr and N are semantically related not as a predicate and its argument, but by means of an “additional” predicate ‘σ’, which is not explicitly expressed in the sentence: ‘krik, vyražajuščij bolʹ / scream expressing pain’ or ‘živopisʹ vo.vremja Vozroždenija / painting during Renaissance’.

N–genitive-possessive→N_GEN-poss

stadion–gen-poss→universiteta ‘stadium of.University’

N_GEN-poss and N are semantically related by means of the predicate ‘σ’ = ‘belong’: ‘stadion,

prinadležaščij universitetu / stadium belonging to.University’.

N–metaphorical→NGEN-metaph

lenta–metaph→dorogi ‘ribbon of.road’

N_GEN-metaph is the lexeme whose metaphor is N: lenta_N ‘ribbon’ is the metaphor of doro ga_NGEN-metaph ‘road’.

3) These SSynt-relations require the postposition of their dependent N_GENs with respect to the modified N, with one exception: the qual-adnom SSyntRel allows the anteposition of its N_GEN-qual, if this N_GEN-qual 1) corresponds to Sem-actant 1 of N, 2) has itself no depending noun phrase and 3) is lexically marked as allowing for anteposition;¹ for instance, pojas golubogo cveta ‘belt of.light.blue color’ ~ golubogo cveta pojas. The anteposition of N_GEN-qual is left out of consideration in this paper.

The basic order of postposed cosubordinated N_GENs is determined syntactically — by the above SSyntRels: for each pair of these SSyntRels we indicate the mutual order of the corresponding N_GENs. As a result, we obtain a general six-position template (Figure 1 in Section 2) that specifies the correct position for each type of N_GEN.

Suchatemplateispossiblebecauseofthefollowingessentialfact:generallyspeaking,adependentN_GEN canoccupydifferent linearpositionswithrespecttoitsgoverningNasafunction of the SSyntRel r_i in the N–r_i→N_GEN phrase.Thus:

There are 15 logically possible pairs of N_GENs (the number of combinations from 6 by 2 without repetitions): N_GEN-subj – N_GEN-obj, N_GEN-subj – N_GEN-qual, etc. Three of these pairs are semantically impossible: N_GEN-metaph does not combine with N_GEN-obj, N_GEN-subj and N_GEN-poss (it is difficult to imagine a metaphorically used noun that has a subject / object actant or a possessor). As a result, we have 12 SSyntRel pairs. On the other hand, the qual-adnom and attr-adnom SSyntRels are repeatable, so that we end up with 14 SSyntRel pairs to consider.

However, the use of SSyntRels alone for the linear ordering of cosubordinated N_GENs is not sufficient: for some SSyntRel pairs, the order of N_GENs depends also on the meaning of N and/ or on that of N_GENs. Thus, in the phrase krik boli_NGEN-attr Peti_NGEN-subj ‘scream of.pain of.Pete’ the N_GEN-attr can only precede the N_GEN-subj (*krik Peti boli), but in proizvedenija vosʹmidesjatyx godov_NGEN-attr Lʹva Tolstogo_NGEN-subj ‘works of.1880s of.Leo Tolstoy’ ~ proizvedenija Lʹva Tolstogo vosʹmidesjatyx godov the N_GEN-attr can both precede or follow the N_GEN-subj — if the N_GEN-attr denotes the temporal coordinate of the fact denoted by the governor N. As a consequence, our rules have to account for semantic factors as well.

Before we proceed to the formulation of NGEN-ordering rules, the following principle has to be stated:

Each of our rules is valid only everything else being equal.

This means that the two cosubordinated N_GENs being compared and ordered are of the same “weight” (the corresponding phrases contain the same number of syllables and are of the same syntactic complexity) and there are no discourse factors intervening.

The expression “discourse factors” should be understood very broadly. It covers communicative and referential phenomena that can lead to violations of the word order observed in discourse-neutral contexts. In the following discussion, we ignore:

• The impact of the Communicative Structure. For instance, contrastive emphasis on one of cosubordinated N_GENs can change their habitual linear order. Thus, the neutral order is N + N_{GEN«MATERIAL»} + N_GEN«COLOR»:² stol krasnogo dereva bolʹšogo razmera ‘table of.mahogany of.big size’ ~ ^?stol bolʹšogo razmera krasnogo dereva; however, under emphasis, the dispreferred order is quite normal:

• The impact of the Referential Structure.

— A modifier either specifies a subclass of possible referents of the modified lexeme L (a restrictive modifier), or characterizes L’s referents without specifying a subclass of these (a qualifying modifier). In what follows we consider only restrictive modifiers. Thus, we exclude from our consideration the situation where one of the cosubordinated N_GENs is used as a qualifying modifier (in dashes):

— A restrictive modifier specifies a subclass of possible referents of the modified lexeme L; cosubordinated modifiers specify subsequent subclasses of L’s referent. In a discourse-neutral context, the order of isolating these subclasses is irrelevant for the Speaker; different characteris- tics of the L’s referents are, so to speak, informationally equal for him. In this case, the linear or- der of cosubordinated modifiers is determined by their own properties — syntactic and/or semantic. This is the situation studied in the present paper. Therefore, we exclude the situation where the Speaker first selects a particular subclass of L’s referents and then introduces a subclass of this subclass. For instance, the dispreferred order ^?stol bolʹšogo razmera krasnogo dereva is quite OK if one speaks about tables of big size and specifies a subclass of these in terms of their material; sentence (6) is absolutely correct because of the referential and communicative effects:

2. Rules for ordering cosubordinated N_GENs

The linear order of cosubordinated N_GENs postposed with respect to their common governor N is described by the rules of three types:

1) Rule for the standard linear order of different-type NGENs, represented by their maximal template (Figure 1 below).

By “standard linear order” we mean the order conditioned exclusively by surface-syntactic relations that subordinate N_GENs to their governor N, without recourse to semantic properties of the nouns involved. These properties are taken into account by Rules 2.

2) Rules specifying semantic factors that affect standard linear order of different-type N_GENs.

Rules 2 are, in a sense, stronger than Rule 1: they impose deviations from the standard order of N_GENs determined by Rule 1.

3) Rule for the linear order of same-type N_GENs, represented by their semantically-conditioned hierarchy (Figure 2).

Rules 1–3 are part of word order, or linearization, rules for Russian [Mel’čuk 2011]; more precisely, they are a subset of the quasi-local word order rules.

1) Standard linear order of different-type N_GENs

Figure 1. Linear order of different-type postposed cosubordinated NGENs in Russian

2) Semantic factors of the linear ordering of different-type cosubordinated N_GENs

Semantic properties of N_GEN

1. If N_GEN denotes a kind of N (rather than characterizing an individual N),

then this N_GEN precedes all other cosubordinated N_GENs.³

2. If N_GEN denotes the material of which N is made,

then this N_GEN precedes all other cosubordinated N_GENs except for N_GEN denoting kind.

3. If N_GEN-attr denotes the time of N,

then N_GEN-attr precedes or follows another N_GEN-attr, N_GEN-subj, N_GEN-obj and N_GEN-poss.

4. If N_GEN-attr denotes the causer of N,

then this N_GEN-attr precedes or follows N_GEN-poss.

Semantic properties of N

5. If N denotes the quantity of N_GEN or a set of N_GENs,

then this N_GEN precedes all other cosubordinated N_GENs.

3) Linear order of the same-type cosubordinated N_GENs

Several same-type cosubordinated N_GENs are possible only for two repeatable SSyntRels: qual-adnom and attr-adnom. The mutual order of same-type N_GENs is determined by the following semantic hierarchy:

Figure 2. Semantic hierarchy of same-type NGENs

This hierarchy, based on [Vendler 1968: 128],⁴ is underlain by the Inherence Principle:

The modifiers of an N cosubordinated to N by the same SSyntRel are linearly arranged according to the degree of their semantic “inherence” with respect to N: a more inherent characterization stands closer to N.

We cannot formally define “degree of semantic inherence” of modifiers, but we think that the proposed hierarchy reflects this property well enough. Thus, the “objective” characteristics are more inherent than the “subjective” ones, the internal properties are more inherent than the external ones, and «KIND» is the most inherent characteristic.

Let it be emphasized that, although this hierarchy is introduced for the same-type N_GENs, it is also partially valid for the different-type N_GENs. More precisely, Rules 2 are based on the same Inherence Principle: thus, the N_GEN expressing «KIND» precedes all other N_GENs, etc.

3. Illustrations of N_GEN ordering rules

We will illustrate the above rules, proceeding as follows.

• The SSyntRels are considered pairwise, one after another, from left to right (in conformity with the template in Figure 1).
• Each pair of SSyntRels is illustrated by phrases featuring the standard order of the two N_GEN nouns.
• Each deviation from this standard order is explicitly indicated.
• Each of the two repeatable SSyntRels — that is, qual-adnom and attr-adnom — is also considered in combination with itself.
• More than three cosubordinated N_GENs are practically unacceptable.

The rules in question specify the best ordering possible. Deviations from it can be characterized by different degrees of ill-formedness, of which we distinguish three: ungrammatical (*), hardly acceptable (^??), and jarring (^?). We are aware that our judgments of grammaticality can be challenged; however, for our purposes here it is sufficient if a difference in the degree of correctness is perceived — as we have indicated at the beginning of Section 1.

This SSyntRel is repeatable.

With –qual-adnom→N:

The order of N_GEN-quals in (7) corresponds to the semantic hierarchy in Figure 2.

With –metaph→N; N_GEN-qual precedes N_GEN-metaph:

With –obj-adnom→N; N_GEN-qual precedes N_GEN-obj:

The N_GEN raspredelenija [toka] is an N_GEN-obj (being DSynt-actant II of the noun SISTEMA); according to the standard template (Figure 1), it should follow an N_GEN-qual — as in (9a). However, a semantic factor perturbs the standard order: this N_GEN-obj identifies a kind of system (≈ a particular device), not an individual system, so that in conformity with Rule 2.1 it must precede the N_GEN-qual.

With –attr-adnom→N; N_GEN-qual precedes N_GEN-attr:

In (10b), we see again the impact of a semantic factor: according to Rule 2.2, the N_GEN denoting material precedes all other N_GENs (except the one denoting kind).

With –subj-adnom→N; N_GEN-qual precedes N_GEN-subj:

*rjumka pričudlivoj formy krasnogo vina⁵.

The deviation from the standard order in (11b–с) is imposed by Rule 2.5.

With –gen-poss→N; N_GEN-qual precedes N_GEN-poss:

This SSyntRel is non-repeatable and combines only with an NGEN-qual (see above) and with an N_GEN-attr.

This SSyntRel is non-repeatable.

With –attr-adnom→N; N_GEN-obj precedes N_GEN-attr:

The variation of the placement of the N_GEN-attr denoting time is allowed by Rule 2.3.

With –subj-adnom→N; N_GEN-obj precedes N_GEN-subj:

NB: The violation of the standard order in (15b) is worse than that in (15a) because of Rule 2.1: N_GEN-obj in (15b) denotes a kind of talent. In other words, if (15a) violates only a syntactic rule, (15b) violates both a syntactic rule and a semantic constraint.

With –gen-poss→N; N_GEN-obj precedes N_GEN-poss:

This SSyntRel is repeatable.

With –attr-adnom→N:

The freedom of the placement of the N_GEN-attr denoting time is ensured by Rule 2.3.

The impossibility of the last phrase is also determined by semantic hierarchy: the N_GEN-attr denoting «MATERIAL» should precede other N_GENs (except «KIND»). With –subj-adnom→N; N_GEN-attr precedes N_GEN-subj:

The freedom of placement of the N_GEN-attr denoting time corresponds to Rule 2.3.

With –gen-poss→N; N_GEN-attr precedes N_GEN-poss:

The freedom of placement of the N_GEN-attr denoting time (20b) or the causer (20c) corresponds, respectively, to Rules 2.3 and 2.4.

This SSyntRel is non-repeatable.

With –gen-poss→N; N_GEN-subj precedes N_GEN-poss:

To demonstrate how the rules proposed can be applied, let us return to example (1), repeated here as (22):

• First, the mutual arrangement of cosubordinated N_GENs is specified by the standard template (Figure 1) for different-type N_GENs: N_GEN-qual precedes N_GEN-subj. The phrase russkogo jazyka ‘of. Russian language’ is an N_GEN-subj that expresses DSyntA I of glagoly ‘verbs’, which are elements of the set ‘Russian language’; according to the N_GEN order template, it must follow the phrase soveršennogo vida ‘of.perfective aspect’ (an N_GEN-qual).

• Second, the mutual arrangement of N_GEN-attr and N_GEN-qual is specified by Rule 2.1: in the standard case (= according to the template), N_GEN-qual precedes; but if N_GEN-attr denotes the kind of N, then N_GEN-qual follows. And in (22), the phrase napravlennogo dviženija denotes a particular kind of verbs.

4. Ordering of cosubordinated N_GENs vs. ordering of cosubordinated ADJs

It is interesting to compare the ordering of Russian postposed cosubordinated N_GENs with the ordering of Russian anteposed cosubordinated adjectives. As is to be expected, N_GENs and adjectives, both being noun modifiers and on multiple occasions synonymous, show significant parallelism in their ordering. We will first present the rules for the ordering of cosubordinated adjectives (see Section 4.1) and then compare them with the corresponding rules for N_GENs (see Section 4.2).

4.1. Ordering of cosubordinated ADJs

The papers [Iordanskaja 2000; 2003] propose a hierarchical semantic classification of Russian adjectives that determines their mutual linear ordering — more precisely, their relative closeness to the modified noun. Figure 3 below presents this classification. The higher in the table an adjective semantic class is (i.e., the higher its rank), the closer its instance must be to the modified noun. This is so since an adjective’s rank corresponds to the degree of inherence of the characteristic the adjective expresses: the more inherent the characteristic, the closer to the noun is the adjective.

Figure 3. Hierarchical semantic classification of adjectives [Iordanskaja 2003: 161–162]

Now let us give some examples.

▪ Adjectives that express an objective characteristic are closer to the modified noun than adjectives expressing a subjective characteristic:

▪ Adjectives that express a qualitative characteristic are closer to the modified noun than adjectives expressing a quantitative characteristic:

▪ Adjectives that express a permanent characteristic are closer to the modified noun than adjectives expressing a temporary characteristic:

▪ Adjectives that express an internal property are closer to the modified noun than adjectives expressing an external property:

▪ Hierarchy of internal property adjectives: for instance, «material» adjectives are closer to the modified noun than «color» adjectives; «kind» adjectives are closer to the modified noun than any other adjectives; etc.:

To sum up: The linear ordering of cosubordinated adjectives is determined semantically — by their meaning — of course, everything else being equal, the same as with N_GENs (see the end of Section 1): that is, the weight of genitive phrases being compared and discourse factors.

However, this is not true for Russian possessive adjectives, such as MAMIN ‘Mom’s’ or PETIN ‘Pete’s’: their mutual linear arrangement is determined by their syntactic role, cf.:

To account for this fact, in addition to the modificative SSyntRel, three more SSyntRels for possessive adjectives in Russian are needed: possessive-modificative, subjectival-modificative, and objectival-modificative. As can be seen from (28)–(29), the order of possessive adjectives with respect to the modified N is as follows:

ADJ←poss-modif– + ADJ←subj-modif– + ADJ←obj-modif– + N.

For instance, Petin repinskij mamin portret lit. ‘Pete’s Repin’s Mom’s portrait’ = ‘Mom’s portrait by Repin owned by Pete’.

NB: The cooccurrence of two or more possessive adjectives is rare, so that, generally speaking, it could be ignored. However, this case is interesting from a theoretical viewpoint.

The cooccurrence of possessive adjectives with “normal” ones is determined by two general rules:

1) The possessive-modificative ADJ precedes all “normal” ADJs

2) The subjectival-modificative and objectival-modificative ADJs follow all “normal” ADJs:

4.2. Comparison of the two orderings: Similarities and differences

Recall that cosubordinated N_GENs follow the governing N, while cosubordinated ADJs precede it. Therefore, the order of N_GENs is a mirror image of that of ADJs. This means that comparing these two orderings we actually speak of the degree of closeness of an N_GEN or an ADJ to its governor N.

The ordering of cosubordinated N_GENs and that of cosubordinated ADJs in Russian are similar in the following two respects:

• The mutual ordering of Russian possessive ADJs (ADJ←poss-modif– + ADJ← subj-modif– + ADJ←obj-modif– + N) is the same (of course, mirror-wise) as the mutual ordering of the corresponding N_GENs (that is, N + –obj-adnom→N_GEN + –subj-adnom→N_GEN + –poss-adnom→N_GEN; Figure 1). The possessive adjectival modifier is the outermost, and the objectival adjectival modifier is closer to the modified noun than the subjectival one. This is natural, since possessive ADJs are simply adjectivalizations of N_GENs.
• The mutual ordering of repeatable N_GENs (that is, qual-adnom and attr-adnom N_GENs) is the same as the mutual ordering of non-possessive ADJs, since it is determined by the same hierarchical semantic classification of the corresponding lexical units. This is also natural, since the closeness of a modifier to its governor N is determined by the degree of semantic inherence of the characteristic expressed: a more inherent characterization stands closer to N.

The difference between the ordering of cosubordinated N_GENs and that of cosubordinated ADJs in Russian is as follows. The cosubordinated N_GENs are ordered syntactically — according to different SSyntRels that link them to the governor, with several semantic “corrections” imposed by their meaning. To put it differently, the linear arrangement of Russian N_GENs is based on the relations between the governing N and the depending N_GENs being ordered. The cosubordinated ADJs, however, are ordered semantically — according to their meaning, with several syntactic “corrections” concerning possessive ADJs, which are positioned in conformity with the governing SSyntRels.

The Appendix

Here are the examples illustrating the three cases of impossibility of N_GEN-qual’santeposition:

1. *moej mečty devuška ‘of.my dream girl’, where ‘girl’is Sem-actant 2 of ‘dream’;
2. *cveta morskoj volny pojas ‘of.color of.sea wave belt’;
3. *prošedšego vremeni glagol ‘of.past tense verb’.

Incases 1) and 2) the anteposition of an N_GEN-qual  can be possible under additional communicative and/or syntactic conditions.