3.1 Possessive structures

The two possessive strategies mentioned above for ESG are found across Gaelic varieties. Adger (Reference Adger2013) suggests a way of structurally unifying the two possessive strategies in Gaelic. The structure proposed there for mo bhràthair, ‘my brother’ looks roughly as in (13), abstracting away from details of the theoretical implementation not relevant here.

1. (13)

   

The first aspect of the proposal to note is that the possessive proclitic mo is treated as a determiner. Evidence for this is that it is in complementary distribution with definite determiners, and also functions like a determiner in other ways; see Adger (Reference Adger2013) for details.

1. (14)

   1. a. a  mhac

      his son

      ‘his son’

   2. b. *an a  mhac

      the his son

      ‘his son’

There is good evidence that the proclitic structure involves a null pronominal after the noun, as shown in (13). Hale and McCloskey (Reference Hale and McCloskey1984) provide a battery of arguments for a similar structure in Modern Irish; I will replicate just one here for Scottish Gaelic. Gaelic has a series of emphatic particles that attach to pronouns, as in (15b).

1. (15)

   1. a. Tha mi toilichte

      be. pres I happy

      ‘I am happy’

   2. b. Tha  mise  toilichte

      be.pres I-emph happy

      ‘I am happy’

The generalization about these elements is that they attach to pronouns. Whenever we see a pronoun, we can attach an emphatic particle to emphasize the pronoun. This can be seen when pronouns are in subject position, as in (15), as well as in object position or in a clefted position, as in (16).

1. (16)

   1. a. Bhuail  Iain thusa

      hit.past Iain you-emph

      ‘Ian hit YOU.’

   2. b. Is  e esan  a  bhuail mi

      cop it he-emph that hit I

      ‘It's HIM that I hit.’

Recall that prepositions inflect for agreement. If the pronoun is either a null pronoun, or somehow morphologically melded with the preposition, then we would expect that the emphatic elements should attach to the preposition. This is correct, as (17) shows.

1. (17) Tha  cat agadsa

   be.pres cat at.2sg-emph

   ‘YOU have a cat.’

Turning to the proclitic possessives, if these were pronouns, we would expect to find the emphatic particle following them, but (18) is resolutely ungrammatical.

1. (18) *Tha mo-se bhràthair tinn

   be.pres my-emph brother  sick

   ‘MY brother is happy.’

Instead we find the emphatic element attached to the noun, as in (19).

1. (19) Tha  mo bhràthair-se  tinn

   be.pres my brother-emph sick

   ‘MY brother is happy.’

Assuming then, that emphatic particles always attach to pronouns, the grammaticality of (19) suggests that there is a null pronoun after the noun, and the structure is as sketched above in (13).

In (13), I have specified the pro with a set of features marking person, number and gender. pro is the element that directly specifies a semantic referent, and so it bears a number of semantically relevant features. However, being null, it obviously cannot express those features morphologically. That job is taken over by D, which bears a set of features that agree with those on pro, just as in familiar pro-drop languages like Italian (Rizzi Reference Rizzi1982).

I assume that person, number and gender are encoded by binary features following Noyer's (Reference Noyer1992) typological investigation of agreement systems: [part:±] (distinguishing speech act participants (first and second person) from non-participants (third person), [auth:±] (author—distinguishing first from second), and [fem:±] (distinguishing feminine and masculine). In Gaelic, [auth] is only relevant when [part] has a + value, and [fem] is only relevant when [part] has a − value. This motivates two versions of D, shown in (20).

1. (20)

   1. a. D[part:+, sing:±, auth:±]

   2. b. D[part:−, sing:±, fem:±]

These Ds are entered into the syntactic construction of the noun phrase, and agree with the semantically active features on pro. The agreement operation values the features, so they can be realized in the appropriate morphological form.

The fundamental syntactic condition on pro is that it enters into an agreement relation with some element that can identify it. Under the analysis proposed here, the D head bearing unvalued agreement features plays that role, so Gaelic is analysed as pro-drop in these possessive constructions.

The final aspect of the structure in (13) is the element poss. I will follow Adger's (Reference Adger2013) analysis for Gaelic possessives, simplifying the account for my purposes here. In that proposal, possessive semantics is introduced by a relational functional category, which I notate here as poss and which takes the possessor as one argument and the possessee as another. Where the possessive relationship is one of kinship, as in mo bhràthair, ‘my brother’, the relevant lexical item is as shown in (21).

1. (21) ∅::poss::λxλy.kin(x, y)

Here the phonological realization is zero (∅ before ::), its syntactic category is poss, and its semantics is the lambda-term to the right (following ::).

Syntactically, and semantically, poss combines with the possessee and returns a semantic function that combines with the possessor to give a meaning where the possessee is kin to the possessor. The function of the D (mo in mo bhràthair) is syntactically to license the presence of a null pronominal, and semantically to act as a definite determiner.

Speakers across dialects in the Western Isles have a preference for the genitive/proclitic structure for uniquely and inalienably possessed elements, such as bodyparts and kinship terms.

1. (22)

   1. a. cridhe Sheumais

      heart  Seumas.gen

      ‘Seumas’ heart’

   2. b. *an cridhe aig Seumas

      the heart  at  Seumas

      ‘Seumas’ heart

   3. c. corrag Sheumais

      finger Seumas.gen

      ‘Seumas’ finger’

   4. d. a’  chorrag aig Seumas

      the finger  at  Seumas

      ‘Seumas’ finger’

The difference here between ‘finger’ and ‘heart’ appears to be uniqueness, similar to the difference in English between ‘the finger I hurt’ and ‘the heart I hurt’, where the latter loses its possessive bodypart reading. We see a similar pattern in (23) for non-unique kinship terms.

1. (23)

   1. a. bràthair Sheumais

      brother Seumas.gen

      ‘Seumas’ brother.’

   2. b. ??am bràthair aig Seumas

      the brother at Seumas

      for ‘Seumas’ brother’

   3. c. bràthairean Sheumais

      brother.pl Seumas.gen

      ‘Seumas’ brothers.’

   4. d. na bràthairean aig Seumas

      the brother.pl at Seumas

      for ‘Seumas’ brothers.’

Here again, the difference between ‘the brother’ and ‘the brothers’ appears to be related to uniqueness. For example, (24) is generally considered unacceptable.

1. (24) *a’ mhàthair aig Mòrag

   the mother  at  Morag

   ‘Morag's mother’

This means that we effectively have some constrained variability in the system. Inalienably possessed but non-unique elements allow both structures with, according to my consultants, no detectable difference in meaning, as illustrated in (22c)–(22d) and (23c)–(23d). We capture this distribution in the analysis partly via the semantics of poss. The various details about definiteness are not crucial to the rest of the analysis here (see Adger Reference Adger2013 for discussion).

For alienables, the genitive structure is extremely high register. (25) is strikingly odd in spoken Gaelic, and even in most written registers.

1. (25)

   1. a. peann Sheumais

      pen  Seumas.gen

      ‘Seumas’ pen’

   2. b. a  pheann

      his pen

      ‘His pen’

In vernacular speech, then, variability is restricted to inalienably possessed but non-unique elements.

Given the various components of the structure just motivated, Adger (Reference Adger2013) suggests a way of structurally unifying the two possessive strategies, as follows.

Barker (Reference Barker1995) and many others argue for the existence of a semantic function that has a generalized relational semantics, pragmatically determined, which is used to capture the wide range of meaning relations in the English genitive construction.

1. (26) Anson's ball

The relation between Anson and ball can be any pragmatically reasonable one: Anson could own the ball, have just kicked it, be temporarily holding it, etc. This same range of relations is available for the prepositional possessor in Gaelic, motivating this general relational semantics for poss in the prepositional possessive construction.

This gives us two varieties of the category poss: one is phonologically zero, lacks unvalued agreement features, and has a restricted set of semantic relations associated with it (the lexical item introducing the kin relation we saw above). The other, in the unified analysis, bears agreement features, has a general relational semantics, and is realized via the usual morphology of the preposition aig. For concreteness, (27b) shows the version for first person singular generalized possessor.

1. (27)

   1. a. ∅::poss::λxλy.kin(x, y)

   2. b. agam::poss[sg:+, part:+, auth:+]::λxλy.R(x, y).

The same basic syntax that we used for mo bhràthair is sufficient to capture am bràthair agam, which is literally ‘the brother at-me’, using the generalised possessor instead of the specific kinship possessor.Footnote ⁴

1. (28)

   

Here it is poss that will agree with pro, licensing it, rather than D. It follows from the fact that pro is then licensed that D can have no agreement features (or else we would have double agreement, something which is not used for the licensing of pro) and a simple definite article is the only available item that will be syntactically well formed.Footnote ⁵

Using the generalized possessor poss rather than the kinship poss, am bràthair agam is derived as in (29).

1. (29)

   

The definite article introduces an iota operator in the semantics, and the possP combines with the head noun by predicate modification, deriving the representation in (30).

1. (30) ιx. brother(x) ∧ R(x, speaker)

The pragmatics of the situation resolve the R relation to one of kinship in ordinary contexts (or to another pragmatically relevant relation).

This system derives the range of form-meaning relations in the vernacular Gaelic possessive straightforwardly (the high register written varieties simply lack the prepositional version). I give examples here for versions of ‘her brother/pen’.

1. (31)

   1. a. D[sg:+,part:−, fem:+] brother poss pro

      a  bràthair ∅  ∅

      ιx. brother(x) ∧ kin(x, y), where y is an identifiable female.

   2. b. D[def] brother poss[sg:+, part:−, fem:+] pro

      am  bràthair aice ∅

      ιx. brother(x) ∧ R(x, y), where y is an identifiable female. R is resolved pragmatically

   3. c. D[sg:+,part:−, fem:+] pen  poss pro

      a peann ∅ ∅

      ιx. pen(x) ∧ kin(x, y), where y is an identifiable female. Pragmatic clash.

   4. d. D[def] pen poss[sg:+,part:−, fem:+] pro

      am  peann aice ∅

      ιx. pen(x) ∧ R(x, y), where y is an identifiable female. R is resolved pragmatically.

3.2 Change in ESG possessives

Let us now turn to the changes Dorian reports for ESG. As I mentioned in the first section, Dorian reports that her fluent speakers appear to know the relevant possessive forms, though, except for what Dorian takes to be fixed expressions, they are not easily accessible. The younger speakers generalize the prepositional possessive strategy throughout, and Dorian reports that these speakers even seem unaware of the relevant proclitic forms. It's worth quoting the relevant passage from Dorian (Reference Dorian1973: 419) in full.Footnote ⁶

For Dorian's younger speakers, even the first and second singular give way to the prepositional strategy, so that the alienable/inalienable distinction is no longer grammatically marked. Why should this be the change that happens here? Dorian later suggests that it may be because the alienable/inalienable distinction is not strong elsewhere in the grammar of the language. The use of possessive clitics is

This proposal indeed leads to the expectation that one of the two structures would be lost, but does not explain which one should be lost. It requires us to assume that there is something marked about ‘inalienability’ which is not marked for ‘alienability’, such that the former is lost and the latter maintained. Without some further stipulation, this proposal doesn't lead to an explanation of why the maintained structure is not analogous to English, the speaker's other language. Why do we find an leabhar agam, literally, ‘the book at me’ rather than *mo leabhar, analogous to ‘my book’? More generally, why do the structures not converge to the structure of English, which is, for the younger speakers, the dominant language (Winford Reference Winford1993)?

The answer is twofold: first, the richness of phonological forms associated with different values of agreement features is reduced, leading to a reduction in the features themselves; second, reduced agreement features pose a problem for licensing pro, and the solution is given by what other resources are available in the language to express the relevant meanings. In the case at hand, the prepositional possessor can be used to express anything that the inalienable possessor can express, so there is a route to solving the problem raised by licensing pro in inalienables: to use a structure that does the job, even though that structure is quite unlike possessives in English, the speakers’ dominant language.

The lexical forms found for the ESG proclitics, which Dorian reports are available in fixed expressions, are given in (32).

1. (32) ESG Possessive Proclitic Paradigm

   

We can see that there are syncretisms already evident in the plural. ESG is distinct from Western varieties of Gaelic in not distinguishing 1st plural from 2nd plural (in Western Gaelic this would be ar, vs. ur, while these have been neutralised to na in ESG). The morphological distinctions motivate a less rich set of D's than is available in Western varieties.

1. (33) Morphosyntactic features for possessive proclitics in fixed expressions

   

Given Dorian's outline of the changes, for all speakers only the first and second persons have an accessible phonology outside these fixed expressions. For ESG, then, D ends up having the relations between the features and the phonology listed in (34).

1. (34) Morphosyntactic features for possessive proclitics outside fixed expressions

   

The absence of the ↔ symbol here expresses the lack of a phonological form for this feature specification.

There are, then, two aspects to feature loss here. One is the existence of a reduced morphosyntactic feature set compared to other varieties of Gaelic (including earlier varieties of the language). The other is related to an inability to access phonological forms for the feature bundles, a result of reduced exposure to the full range of morphological forms during acquisition. For these speakers, although pro can be syntactically licensed, there is simply no phonological way of expressing its licenser. However, D is obligatory in Gaelic noun phrases, leading to a problem of how to express the structure.

1. (35)

   

This weakening of feature specification can be seen as part of the same change that has already gone to completion with the objects of non-finite verbs. Recall that in other varieties of Gaelic, aspectual elements inflect for agreement features, but ESG instead uses an overt pronoun, as shown in (36) and (37).

1. (36) An robh  a’  ghrian gan  dalladh?

   Q be.past  the  sun simp.3pl  blind.Vn

   ‘Was the sun blinding them?’ Western Gaelic

2. (37) An do  robh a’ ghrian  dalladh iad

   Q be.past  the sun blind.vn  them

   ‘Was the sun blinding them?’   ESG

We can understand this as a variation in the range of categories that can be specified with licencing agreement features. Western varieties of Gaelic have the relevant features on the element that signifies simple aspect. ESG has lost that specification entirely, so that pro cannot be licensed in object position of non-finite verbs, leading to a grammatical difference between the varieties. The solution in the aspectual constructions is to use an overt pronoun. The same loss appears to be extending to D in possessives. At the time of Dorian's fieldwork, aside from fixed expressions, the only two versions of D which have a phonological form are those that express the difference between first and second person.

This proposal gives us a model for Dorian's observation that the older fluent speakers had difficulty accessing the relevant possessive D, even though they could, with effort, access the right form. That would suggest a cognitive organization of the feature matrix above, where although the items are grammatically available, they are close to impossible to access (one could model this via a probabilistic version of lexical access). However, as the next generation acquires the language, if there is not sufficient data to give them access to the forms, the relevant feature specifications are simply never acquired and language change takes place: the feature matrices are simply not associated with a phonological form. As the change progresses, the version of D bearing the relevant features is totally lost, leading to an inability to license pro.

If the language loses agreement feature specification over time, then, given that pro needs agreement features to be present in the structure in order to be licensed, what are the implications? One might guess that the grammar would just break down here, and the speakers would have only a fragmentary system, but that is not what Dorian reports for even the young speakers. Another possibility would be for speakers to reanalyse the proclitics, not as agreeing Ds, but rather as full pronouns, following the model of English, and this appears to be what has happened in the aspectual construction.

One might imagine, analogizing with the extant aspectual construction, that the language would resort to an overt pronoun, perhaps moving this to the prenominal position under pressure to converge with English. Such a reanalysis would result in structures such as (38).

1. (38) *iad bràthair

   them brother

   ‘their brother’

However, this did not happen, and the question is why.

The explanation I propose here appeals to the internal workings of the language as opposed to external pressure from English. Recall that the prepositional poss item, which competes for the same structural position as the inalienable poss item, semantically subsumes it: inalienable poss is a hyponym of prepositional poss, which can be used for inalienables as well. Since the range of possible uses of the prepositional poss category semantically contains the range of use of the inalienable poss category, speakers acquiring it maintain full expressive capacity: they can say all the things they need to say.

Consider a situation of normal acquisition of the Western Isles structure. Here the child has to figure out that there are two lexical items that express possessive relationships in the syntax: one is usable for almost all possessive relations, while the other is usable for a narrower range of kinship and bodypart relations. Following the idea that the learning mechanism posits as few featural distinctions as it can, a child will conservatively posit just one of the two possible poss heads. If she posits the narrower one, then general facts about what the child wants to express will immediately require her to also adopt the broader one, simply to deal with inalienables. If, on the other hand, she posits the more general one, the data will, at some point, suggest that there is a bifurcation in the feature space. The child will notice that a cognitively important category, plausibly available as part of UG, will consistently correlate with one structure, while the other structure is available for all kinds of possession. Since there will be sufficient data to suggest a one-to-one match between proclitic possessive syntax and unique kin or bodyparts, the child will posit two lexical items for possession in the language: the two we have seen above.

However, in a situation of reduced input, such as when a language is obsolescing, if the child posits the more generally applicable lexical item, there will simply not be enough data to also posit the more specific lexical item. Further, the general lexical item will perform all of the expressive linguistic tasks that the speaker needs, so the speaker is not left with any gaps in their expressive capacity. Of course the child will encounter a small number of cases where the possessive proclitics are found, but these can be learned as exceptions or idiomatic special cases, as Dorian suggested. Dorian (Reference Dorian1978) provides some evidence that even in fixed expressions, certain forms (the special form of the feminine singular preceding a vowel, for example) are lost, leaving lexical gaps. This proposal follows Yang's suggestion that there is a quantity threshold for when a child posits a rule (Yang Reference Yang2005). In the case of ESG kinship possessives, my proposal is that there is simply not enough data to permit the child to acquire the kinship poss lexical item.

There are then four aspects of this situation that are relevant, listed in (39).

1. (39)

   1. a. Reduction in the range of specification of agreement features, following a reduction in the range of phonological forms distinguishing agreement feature values, and hence a syntactic ‘problem’ of what to do with pro.

   2. b. The generalized possessor item is morphosyntactically realized as a preposition, and prepositional agreement features are robustly present, so can step in.

   3. c. The generalized possessor item is sufficient to accomplish all of the tasks for expressing the meanings the speaker requires.

   4. d. In an obsolescing situation, there is not enough data to reach the threshold where a specialised kinship poss is posited.

This gives rise to a problem raised by structures like (40), where the licenser of pro has no possible expression:

1. (40)

   

This problem is solved by using a structure such as (41), where the licenser has a wider range of meanings and can be expressed.

1. (41)

   

The reason that, in the possessive, speakers have not analogized to English and reanalysed the proclitic as a full pronoun is that an expressively equivalent alternative is already available. We saw that one change, a loss of agreeing aspectual particles, had already run to completion, and in that case the solution to pro not being licensed was to use an overt pronoun. In the possessive construction, a more general structure has been substituted, so that there is still a pro, but it is licensed by prepositional inflection, a type of inflection that Dorian reports is very robust, even for her semi-speakers.

The idea that speakers opt for a more general form in situations of language change, or contact, has been proposed for other situations. Silva-Corvalán (Reference Silva-Corvalán1994), for example, reports on a situation in Los Angeles Spanish where the syntax of clitics changes for internal reasons, even though it is in close contact with English.