On Three Different Types of Subjectlessness and How to Model Them in Lfg

Outside LFG, the term ‘subjectless’ is found referring to a range of phenomena in which the expression of the predicate lacks an overt lexical item (a syntactic constituent) bearing the grammatical function of the subject. In some of these phenomena, for example in prodrop, the architecture of LFG allows us to find the subject at the a-structure and f-structure levels despite there being no categorial element expressing the subject. There are, however, other subjectless constructions for which there are no readily available LFG accounts, and it is not always obvious how they could be analysed adequately. It is constructions of this type – often called ‘impersonal’ in traditional literature – that will be the focus of this paper, exemplified from Polish which is rich in impersonal forms. I will begin with an overview of all Polish constructions which appear to be subjectless. I will identify three types of construction which lack subjects at some level of analysis: pro-drop constructions (including the so-called ‘weather constructions’ and ‘adversity impersonals’), morpholexically derived impersonal constructions, and truly subjectless constructions. I will then demonstrate how they differ by highlighting their morphological and syntactic properties and suggest levels of representation at which the different types of ‘null/missing subjects’ can be captured theoretically.  ‘Subjectless’ constructions in Polish Polish has a large number of different constructions that appear to be subjectless. As will be demonstrated in the further sections of this paper, their morphosyntactic properties allow them to be grouped into the following categories: . pro-drop constructions. These include clauses formed from personal predicates with a dropped personal pronoun (such as ‘[He] saw.3sg.masc that the door was open and [he] went.3sg.masc in’), and from personal predicates with a dropped indefinite pronoun, both the pronoun referring to humans (‘[Someone] was-writing.3sg.masc as if [he] wanted.3sg.masc to warn us’) and the pronoun referring to non-humans, as in ‘weather constructions’ and ‘adversity impersonals’ (‘[Something] was-blowing.3sg.neut as if [it] wanted.3sg.neut to pull out trees with their roots’, ‘[Something] threw.3sg.neut him to the side’). Contrary to the frequently found though unsubstantiated assumption, Polish weather constructions, adversity impersonals and other apparently subjectless clauses involving verbs of physical or psychological states do not lack a syntactic subject, nor do they have a suppressed or other empty category/zero subject. Instead, they result from subject ellipsis, with their omitted subject being the indefinite pronoun referring to non-humans – that is, they are instances of proindef-drop. . Morpholexically derived impersonal constructions. These include clauses formed from personal predicates whose fully operational (binding, controlling, available for raising) and interpretable syntactic subject has been ‘suppressed’ by a morpholexical operation and is not allowed to appear as a constituent in surface syntax. This category includes the socalled ‘-no/-to impersonal’ (Bito Piotra ‘Beat.impers Peter(masc).acc’ meaning ‘[They] beat Peter’) and the reflexive impersonal (Biło sie ø Piotra ‘Beat.3sg.neut refl Peter(masc).acc’ meaning ‘[One] beat Peter’). I gratefully acknowledge the current ESRC grant RES---. This paper draws from my PhD thesis on passive and passive-like constructions in English and Polish (Kibort ). . Subjectless constructions. These are formed from predicates without either an overt or an omitted/covert syntactic subject which could participate in syntactic operations such as control or raising. This category includes inherently impersonal predicates (a small class of defective, non-inflecting verbs as in: Słychać jaø ‘[One] hear.non-personal her.acc’, or ‘Było widać łaøke ø ‘[One] was.3sg.neut see.non-personal meadow(fem).acc’) and predicates which have lost their subjects as a result of derivation. The latter occurs, for instance, when passivisation is applied to an intransitive predicate. . Constructions with non-agreeing subjects. These include predicative adverbial constructions (e.g., Miło z tobaø podróżować ‘Nicely with you travel.inf’ meaning ‘It is nice travelling with you’) or nominativeless clauses with predicates requiring a genitive argument (e.g., Przybywa wody ‘Becomes-more water(fem).gen’). Clauses of this type complete the typology of Polish ‘subjectless’ constructions, but it is important to realise that their subjectlessness is only apparent. They do not lack subjects, but simply have nonagreeing subjects. Thus, they pattern with other clauses whose subjects have some nominal properties but are nevertheless not appropriate agreement controllers. Such subjects are prepositional phrases, quantifier phrases (with quantifiers requiring their complements to appear in genitive case), clausal subjects (including infinitival subjects), and certain indeclinable subjects such as acronyms and foreign place names. This type of construction will not be taken up further in this paper; for some more discussion and analysis see Kibort (:-). The first three types of construction all lack overt subjects but each has different morphosyntactic properties, which will be exemplified in the sections below. I will show that the architecture of LFG makes it possible to identify these different types of subjectlessness at different levels of representation of the predicate, even though the new analyses may require revising some elements of LFG’s theory of argument structure. Constructions of TYPE  (pro-drop) fall under the standard LFG analysis of unexpressed pronouns. Constructions of TYPE  (morpholexical impersonals) need a new analysis: they have an unaltered argument structure, but the categorial expression of their fully operational syntactic subject is suppressed. At functional structure level, the covert subject may be analysed as an obligatory pro analogous to the pro in arbitrary anaphoric control constructions. Finally, constructions of TYPE  (truly subjectless predicates), which have no subject at any level of analysis (a-structure, f-structure, or c-structure), provide a strong argument against LFG’s Subject Condition (‘Every predicator must have a subject’; Bresnan :).  TYPE : pro-drop constructions The following Polish sentences are not usually associated with the pro-drop phenomenon. They exemplify predicates denoting natural or supernatural phenomena including weather phenomena (-), the so-called ‘adversity impersonals’ (), and predicates denoting physical or psychological states (): () Pada/Świta. rains/dawns ‘It is raining/dawning.’ () W in tym this domu house straszy. spooks ‘It haunts in this house.’ (meaning: ‘This house is haunted’) () Wyrzuciło threw-out.3sg.neut łódke ø boat(fem).acc na onto brzeg. shore ‘The boat got thrown onto the shore.’ () a. Mdli/Dusi/Skre øca/Ciaøgnie/Boli/Swe ødzi/Kłuje nauseates/chokes/convulses/pulls/aches/itches/stabs mnie. me.acc ‘[Something] makes me nauseous/choke/convulse/contract my muscles/painful/itch/gives me shooting pains.’ b. Mdli/Dusi/Skre øca nauseates/chokes/convulses mnie me.acc od from tego this zapachu. smell ‘This smell makes me nauseous/choke/convulse.’ c. Mdli/Dusi/Skre øca nauseates/chokes/convulses mnie me.acc z from bólu/zazdrości. pain/envy ‘The pain/envy makes me nauseous/choke/convulse.’ Clauses of this type commonly appear without an overt nominative subject and use verbal forms displaying ‘default’ agreement. They are often treated as impersonal active clauses with covert inanimate subjects – that is, they are taken to contain an empty or ‘zero’ subject. The existence of the ‘zero’ subject is taken to trigger ‘default’ 3sg.neut agreement in the verb and impose on the construction an ‘inherent inanimate force’ interpretation. However, it is straightforward to demonstrate that all predicates used in these constructions can easily appear with an overt nominative subject, whether in the singular or in the plural: () a. Padało. rained.3sg.neut ∼ Deszcz rain(masc).nom padał. rained.3sg.masc ‘It was raining. ∼ The rain was raining.’ b. Świta. dawns ∼ Poranek morning(masc).nom świta. dawns ‘It is dawning. ∼ The morning is dawning.’ c. Cze østo often padajaø rain.3pl tu here ulewne torrential.nonvir.nom deszcze. rains(nonvir).nom ‘Torrential rains often rain here.’ () a. W in tym this domu house straszy. spooks ‘It haunts in this house.’ (meaning: ‘This house is haunted’) b. W in tym this domu house coś something(neut).nom straszy. spooks ‘Something haunts in this house.’ (meaning: ‘This house is haunted by something/some ghost’) c. W in tym this domu house straszy spooks duch ghost(masc).nom pradziadka. great-grandfather(masc).gen ‘This house is haunted by the ghost of the great grandfather.’ () a. Morze sea(neut).nom wyrzuciło threw-out.3sg.neut łódke ø boat(fem).acc na onto brzeg. shore ‘The sea threw the boat onto the shore.’ b. Fale waves(nonvir).nom wyrzuciły threw-out.3pl.nonvir łódke ø boat(fem).acc na onto brzeg. shore ‘The waves threw the boat onto the shore.’ () a. Wszystkie all zapachy smells(nonvir).nom mnie me.acc mdliły. nauseated.3pl.nonvir Nawet even zapach smell(masc).nom kawy coffee(fem).gen mnie me.acc mdlił. nauseated.3sg.masc ‘All smells made me nauseous. Even the smell of coffee made me nauseous.’ b. Ból pain(masc).nom skre øcał convulsed.3sg.masc mnie me.acc niemiłosiernie. mercilessly ‘The pain convulsed me mercilessly.’ c. Bolała/Swe ødziała ached/itched.3sg.fem mnie me.acc głowa. head(fem).nom ‘My head ached/itched.’ d. Coś something(neut).nom mnie me.acc dusi. chokes / Dusiły choked.3pl.nonvir mnie me.acc te these.nonvir.nom zapachy. smells(nonvir).nom ‘Something makes me choke. / Those smells made me choke.’ Furthermore, there are no morphosyntactic restrictions on any of these verbs which would prevent them from agreeing with a subject in a person other than third, e.g.: () Głośno loudly wiejesz, blow.2sg wietrze. wind(masc).voc ‘You are blowing loudly, wind.’ All this suggests that these constructions do not lack a subject at any level of abstract representation of the predicate. They are personal predicates and their superficial subjectlessness results from the familiar pro-drop phenomenon. Wierzbicka () argued against a pro-drop analysis of Polish ‘weather constructions’ assuming that the dropped pronoun would have to be a personal pronoun corresponding in gender to the nominal denoting the particular natural phenomenon, that is: on ‘he[masc]’ for deszcz ‘rain(masc)’ or wiatr ‘wind(masc)’; ono ‘it[neut]’ for niebo ‘sky(neut)’ or powietrze ‘air(neut)’. She assumed that, if the ‘subjectless’ weather sentences were a result of subject ellipsis, the verb would have to display gender agreement with the dropped pronoun corresponding to the nominal denoting the natural phenomenon. Such agreement is indeed not established. However, this hypothesis makes an incorrect assumption about the subject of the weather constructions: the dropped subject is not the personal pronoun, but the indefinite pronoun. All nouns and pronouns in Polish, whether denoting or referring to people, objects, abstract notions or natural phenomena, bear the feature of inherent grammatical gender: masc, fem or neut in the singular, and vir (masculine human) or nonvir (all other, i.e. non-masculine human and all non-human) in the plural. The so-called indefinite pronouns ktoś ‘somebody’, referring to humans (hum), and coś ‘something’, referring to non-humans (non-hum), bear the grammatical features masc and neut, respectively, and these are also the gender agreements that they trigger in the verb. The following is an example of a definite (and referential) use of the indefinite hum pronoun ktoś which is employed here in order to avoid specifying the gender (and number) of the referent of the agent: () Ten this.masc.nom ktoś someone(masc).nom pisał, wrote.3sg.MASC jakby as-if chciał wanted.3sg.MASC nas us ostrzec. warn.inf ‘This person was writing as if he/she wanted to warn us [of something].’ If the pronoun is dropped, as in any other familiar case of ellipsis, the resulting sentence is: () Pisał, wrote.3sg.MASC jakby as-if chciał wanted.3sg.MASC nas us ostrzec. warn.inf ‘He/she was writing as if he/she wanted to warn us [of something].’ Although sentence () taken out of context is ambiguous between a gender non-specific (‘he or she’) and a gender specific (‘he’) interpretation of its agent, both examples () and () show that SG.MASC agreement is used with unspecified singular human subjects, whether overt or dropped. By analogy, the following sentence: () Wieje, blows[3sg].NEUT jakby as-if chciało wanted.3sg.NEUT powyrywać pull-out.inf drzewa trees z with korzeniami. roots ‘[The wind] is blowing as if it wanted to pull out the trees with their roots.’ illustrates the use of SG.NEUT agreement with an unspecified non-human subject. In the sentence above, the subject has remained unexpressed overtly, as in example (). If we choose to specify the number and gender of the agent of the event denoted by the verb, the number and gender agreement corresponding to the unspecified agent is replaced by verbal inflection corresponding to the grammatical number and gender of the subject nominal. Therefore, in case of human agents, we can have, for example: () Piotr Peter.MASC.nom pisał, wrote.3sg.MASC jakby as-if chciał wanted.3sg.MASC nas us ostrzec. warn.inf ‘Peter was writing as if he wanted to warn us [of something].’ () Ta this.FEM.nom kobieta woman.FEM.nom pisała, wrote.3sg.FEM jakby as-if chciała wanted.3sg.FEM nas us ostrzec. warn.inf ‘This woman was writing as if she wanted to warn us [of something].’ This is a simplified view of Polish gender in its interaction with number, but it is sufficient to describe the phenomena discussed in this paper. Moreover, in case of subject ellipsis, the verb retains its agreement with the ‘dropped pro’ denoting a human agent, because personal pronouns are specified for exactly the same features which trigger the agreement as the nominals they correspond to: () (On) (he[MASC].nom) pisał, wrote.3sg.MASC jakby as-if chciał wanted.3sg.MASC nas us ostrzec. warn.inf ‘He was writing as if he wanted to warn us [of something].’ () (Ona) (she[FEM].nom) pisała, wrote.3sg.FEM jakby as-if chciała wanted.3sg.FEM nas us ostrzec. warn.inf ‘She was writing as if she wanted to warn us [of something].’ In case of overtly expressed non-human agents, the gender and number agreement also corresponds to the grammatical gender and number of the subject nominal, as was shown in sentences to the right of the hyphens in example (). However, even though deszcz ‘rain(masc)’ or wiatr ‘wind(masc)’ are grammatically masculine, and niebo ‘sky(neut)’ or powietrze ‘air(neut)’ are grammatically neuter, it is not possible to replace these nominals with the personal pronouns on ‘he[masc]’ or ono ‘it[neut]’ unless we personify the natural phenomena in question. The unacceptability – or, more accurately, the infelicity – of sentences such as: () a. #On he[masc].nom padał. rained.3sg.masc ‘It [he=the rain] was raining.’ b. #Ono it[neut].nom sie ø refl ochłodziło. cooled-down.3sg.neut ‘It [=the air] has become colder.’ follows from the fact that, in addition to being specified for number and gender, personal pronouns in Polish conventionally denote human (hum) agents, while verbs such as ‘rain’, ‘snow’ or ‘cloud over’ imply a non-human (non-hum) ‘agent’ or cause. Since weather verbs in Polish are not normally used with personal pronouns, it is, therefore, not plausible to suggest that weather constructions without an overt subject result from personal pronoun ellipsis. It is, however, reasonable to see them as resulting from the ellipsis of the indefinite pronoun coś ‘something’ which is used to achieve the ‘unspecified agent’ interpretation and which triggers 3sg.neut agreement. In case of subject ellipsis (‘proindef-drop’), the 3sg.neut verbal agreement is retained. One of the conventional uses of the ‘indefinite’ pronouns, both hum and non-hum, is with a definite referent whom/which the speaker chooses not to specify. By omitting the indefinite pronoun coś ‘something’, the identity of the ‘agent’ is not questioned, but left unspecified, since it is in most cases understood from the context. Clauses with proindef-drop do not present problems for LFG. They fall under the standard analysis of unexpressed pronouns, e.g., Bresnan (:-). She analyses prodrop as the functional specification of a pronominal argument by the head to which the pronominal inflection is bound, which entails the absence of the structural expression of For more detailed discussion of the morphosyntax of proindef-drop constructions in Polish, see Kibort (:-). the pronoun as a syntactic NP or DP when the optional semantic and binding features of the pronominal inflection are present.  TYPE : morpholexical impersonals There are two constructions in Polish whose grammaticalised function is to despecify the principal participant of the predicate: the -no/-to impersonal and the reflexive impersonal. The principal participant in these constructions is interpreted as either an unspecified or a generic human agent or experiencer. The constructions have particular morphosyntactic properties and morphological marking. The -no/-to impersonal uses a dedicated, uninflecting verb form ending in -no/-to, and is restricted to past tense, while the reflexive impersonal uses 3sg.neut verb form and the reflexive marker sie ø, and can be used in all tenses. . The -no/-to construction The -no/-to construction is exemplified in () and (): () Budowano built.impers szkołe ø. school(fem).acc ‘A/The school was built. / [They] were building a/the school.’ () Tutaj here tańczono. danced.impers ‘There was dancing here. / [They] danced here.’ One of the key properties of this construction is that it can be used with both intransitive and transitive predicates, and in the case of transitives the accusative object is retained, as in (). Another key property is that it can be formed from both unergative and unaccusative predicates, including the habitual/iterative form of the verb ‘be’. It can be formed from passivised predicates, therefore it has to be treated as independent of passivisation. The following example contains an impersonal form of the auxiliary (bywano) in a periphrastic passive construction with a passive participle (bitymi): () Dostawano received.impers różne various.nonvir.acc kary punishments(nonvir).acc i and bywano wasiterative.impers bitymi. beat.part.pl.instr ‘[They/One] received various punishments and were/was beaten.’ The Polish -no/-to construction does not, under any circumstances, accept the surface expression of a nominative subject (-), nor does it accept the expression of the agent in an oblique phrase as in the passive, (-): () *Władze authorities(nonvir).nom budowano built.impers szkołe ø. school(fem).acc ‘(intended) The authorities were building a/the school.’ () *Uczniowie pupils(vir).nom tutaj here tańczono. danced.impers ‘(intended) The pupils were dancing here.’ () *Budowano built.impers szkołe ø school(fem).acc przez by władze. authorities ‘(intended) A/The school was built by the authorities.’ () *Tutaj here tańczono danced.impers przez by uczniów. pupils ‘(intended) The dancing was done here by pupils.’ However, despite being superficially subjectless, the -no/-to impersonal appears to have a syntactically active ‘covert’ subject which participates in syntactic control and binding. The -no/-to predicate can share its subject with infinitives (), with deverbal adverbials (), and in a subject-raising construction (); the covert subject of -no/-to is also capable of binding reflexive and reflexive possessive pronouns when they need to be bound by the subject (-): () Chciano wanted.impers wyjechać. leave.inf ‘There was eagerness to leave.’ () Wsiadajaøc get-on.partcontemp do into autobusu bus pokazywano showed.impers bilety. tickets(nonvir).acc ‘On getting on the bus [they]/one showed the tickets.’ () Zdawano seemed.impers sie ø refl tego this.masc.gen nie neg dostrzegać. notice.inf ‘[They] seemed not to notice this.’ () Oglaødano looked-at.impers sie ø/siebie refl/self.acc w in lustrze. mirror ‘[They] looked at [them]selves in the mirror. / One looked at oneself in the mirror.’ () Oglaødano looked-at.impers swoje own[refl].nonvir.acc zbiory. collections(nonvir).acc ‘[They] looked at [their] own collections. / One looked at one’s collection.’ The -no/-to impersonal is not agentless, either. Its agent (or experiencer) licenses all sorts of agent-oriented adverbials (e.g., celowo ‘on purpose’) and is invariably interpreted as an unspecified but definite human. The human interpretation of the agent/experiencer, which has been grammaticalised in the usage of this construction, overrides any semantic implications to the contrary that may arise from the meaning of the lexical items used in the clause, or from the context. Therefore, the -no/-to forms of predicates such as ‘bark’ or ‘build nests’ can only be interpreted as involving human activity. The covert subject of the -no/-to impersonal triggers virile (plural) marking in agreeing (adjectival and nominal) predicative complements. Examples () and () show that expressions whose referents are, inflectionally, other than virile (plural) are incompatible with the -no/-to form and produce ill-formed clauses: () (example adapted from Dziwirek :) a. *Pracowano worked.impers jako as nauczyciel. teacher(masc).nom b. *Pracowano worked.impers jako as nauczycielka. teacher(fem).nom c. *Pracowano worked.impers jako as nauczycielki. teachers(nonvir)[fem].nom d. Pracowano worked.impers jako as nauczyciele. teachers(vir).nom ‘[They] worked as teachers. / One worked as a teacher.’ () a. *Wyglaødano looked.impers na to szcze ǿsliwego. happy.masc.acc b. *Wyglaødano looked.impers na to szcze ǿsliwaø. happy.fem.acc c. *Wyglaødano looked.impers na to szcze ǿsliwe. happy.nonvir.acc d. Wyglaødano looked.impers na to szcze ǿsliwych. happy.vir.acc ‘[They/One] looked happy.’ There is no off-the-shelf LFG analysis of impersonals, and therefore none to fit the -no/to construction. The -no/-to impersonal is not a syntactic variant of the passive: it is neither an ill-behaved passive of the transitive, nor equivalent to the passive of the intransitive. It retains the accusative object, can be applied to unaccusatives, and exists alongside the passive – as was shown in (), it can be formed from a passivised transitive predicate if the passive subject can be interpreted as human. It is, therefore, a different morpholexical construction to the passive (for more detailed argumentation against a passive analysis of this construction, see Kibort ). The fact that -no/-to impersonalisation preserves both the grammatical relations and the internal (lexical) semantic structure of the predicate, but only suppresses the surface realisation of the subject, means that, unlike valency-changing operations, it is argumentstructure-neutral – the argument structure of an impersonalised verb is unaltered. Thus, the a-structure representation of the impersonalised transitive verb czytano ‘read.impers’ is the same as that of a personal active verb: () impersonal of the transitive 〈 x y 〉 | | subj obj In the a-structure representations of impersonalised predicates that I had hypothesised prior to this paper, I placed the symbol Ø under the subj to indicate that this grammatical function was prevented from being mapped onto a categorial argument. Cf. the LMT ‘suppression’ rule which says: ‘Do not map an argument to the syntax’ (e.g., Bresnan :-; Falk :) and is notated with Ø. I understand now that this notation was superfluous. If we accept that the subject of the impersonalised predicate is an (obligatory) pro (i.e., the impersonal inflection provides the specification (↑ subj pred) = ‘pro’ in the f-structure), there can be no other NP that could be the subject at the same time. The covert subject is not a phonetically empty pronoun (pro). It is not a null expletive – Polish does not have expletives at all, and the covert subject has a thematic role. It is not a dropped pronominal subject (vir), either – the -no/-to morphology is not equivalent to normal vir morphology, and we would have no way of explaining what prohibits the overt expression of the pronominal subject. However, it is possible to analyse the covert subject of the -no/-to impersonal as a pronominal anaphor analogous to the null, or shared, subject of non-finite clauses in syntactic control contexts (pro). In constructions involving arbitrary anaphoric control the reference of the pronominal element in the clause is not determined syntactically, but the controlled argument finds it referent in a way similar to an ordinary pronoun. Thus, the f-structure of czytano ‘read.impers’ could be represented as in (), with the c-structure appropriately lacking the node for the categorial expression of the subject: ()