1.1 Deficient pronouns

Before we introduce Polish data, let us specify the notion of deficient pronouns. Since the publication of Cardinaletti & Starke (1994), the peeling hypothesis has become popular, on which pronouns can be arranged in three classes showing the following levels of structural deficiency:

Structural impoverishment of deficient pronouns (weak pronouns and clitics) determines their morpho-syntax:

Commonly, the classification in (1) constitutes only a set of guidelines and a point of reference, as concrete examples of pro_DFs in particular languages fail to show all the properties expected of a given class, show mixed properties or properties opposite to what is expected.¹ We also treat it as reference and we are not committed to it, and specifically, to the view that structural peeling is a process through which pro_DFs have come into existence. In particular, the class of weak pronouns is not as homogeneous as Cardinaletti and Starke would have liked.²

Other pronominal classifications are also available. Dechaine & Wiltschko (2002) group pronouns into three categories but along different lines:

Their criteria include internal syntax (whether they are complex, with elements of D, or simple), distribution (whether they occur in argument or predicate positions), semantic properties (i.e. whether they can function as bound variables) and binding-theoretic status (whether they are subject to Conditions C or B of the Binding Theory of Chomsky 1981). Upon closer scrutiny, Dechaine & Wiltschko classify the English indefinite pronoun one as a pro-NP, English 1^st and 2^nd person pronouns as pro-DPs and English 3^rd person pronouns as pro-ΦPs.³

A four-way division is proposed for Slavic, and Polish in particular, in Cetnarowska (2003; 2004) and Jung & Migdalski (2019), where a certain class of pronouns straddles the borderline between strong and weak pronouns:

Jung & Migdalski (2019) conduct a comprehensive study of pro_DFs in Slavic from a diachronic perspective and point to a correlation between the loss of the morphological past tense (aorist) and the loss of pronominal clitics. They submit that certain pro_DFs have been subjected to ‘degrammaticalization’, a shift from X^o, like adverbal clitics in Bulgarian and Macedonian (or Romance languages), to a maximal projection (XP). These gravitate towards the 2^nd position (South and West Slavic) or turn to hybrid weak/strong pronouns. Following Jung (2017), they propose the following scale of pronominal deficiency:

This division corresponds to Cetnarowska’s in (4) and distinguishes between stressed strong pronouns, unstressed strong pronouns, weak pronouns and clitics, arranged along a declining scale of prominence. We endorse the proposal in (5) and in the ensuing analysis groups (5b–5d) are subsumed under the term deficient pronoun (pro_DF). In the cases we consider, the deficiency of the pro_DF concerns its feature make-up: lack of value for its [π] feature. This deficiency requires the pro_DF to undergo movement for feature valuation, while a strong pronoun does not move to value its [iπ] feature.

We adopt the following definition of pro_DFs:

In general, pro_DFs tend to move out of VP and to the left periphery, while strong pronouns do so to a much lesser degree. The syntactic deficiency driving movement of pro_DFs to the left periphery, may be construed in other ways as well. For instance Pescarini (2021: 157–158), adopting the cartographic approach (Rizzi 1997; 2013), argues that the original driving force behind displacement of clitic/weak pronouns in Old Romance was their feature triggering g-topic interpretation. The clitic pronoun used to move as XP to the Wackernagel position (initially, a criterial position for g-topics) through an intermediate landing site in the Infl-area. Subsequently, it underwent reanalysis into an adverbal clitic (X^o) in the Infl-area.

In-depth studies of pronominal clitics in modern Slavic show that, unlike their equivalents in present day Romance, pronominal cliticization in these languages follows two general paths: the adverbal one, where the clitics are heads adjacent to the verb (Bulgarian and Macedonian) and the second position one (exploring the Wackernagel position), where the so-called pronominal clitics (in fact showing properties of maximal projections) occupy the Wackernagel position (Czech, Slovak, Slovenian, Bosnian, Serbian, Croatian and Montenegro). Both types of pro_DFs are subject to ordering constraints (Progovac 1993; Tomić 1996; Stjepanović 1998; Franks & King 2000; Bošković 2002; 2004; Migdalski 2006; 2016; 2021; Zubizarreta & Pancheva 2013; Jung & Migdalski 2019; Nevins 2007; Franks 2017; 2020; Pancheva & Zubizarreta 2018; Stegovec 2020, etc.).⁵ As we show below, Polish pro_DFs behave more like maximal projections than heads.

1.2 Pronominal clusters and the Person Case Constraint

Deficient pronouns tend to cluster in particular positions and show a particular cluster-internal order. The limitation concerns expression of local person on the direct object pronoun; it is determined by presence of local person on the indirect object (the Person Case Constraint):

This constraint has been defined in different ways, with a few prominent formulations provided below:

In minimalist analyses (Bejar & Rezac 2003; 2009; Anagnostopoulou 2005; Nevins 2007; Franks 2017; 2018; Stegovec 2020; Pancheva & Zubizaretta 2018; Foley & Toosavarandani 2022, etc.), the PCC applies to ditransitive predicates with clitic/weak pronouns.⁸ They are defective (in lacking independent 1^st/2^nd person marking) but their deficiency can be repaired by moving them, or linking them via Agree, to a dedicated functional head (Agr/v/Asp), where the 1^st /2^nd person marking is obtained. The diagram below illustrates an application of PCC, with β as the head where the 1^st/2^nd person are licensed:

Pronouns P₁ and P₂ require person marking; they can obtain it via movement to [spec, βP], a position from which they c-command β. Yet, P₁ and P₂ do not stand equal chances, as P₁ c-commands P₂ and is closer to β. Syntactic operations such as Move or Agree are subject to locality constraints, favoring minimal operations (Relativized Minimality of Rizzi 1990, Minimal Link Condition of Chomsky 1995, Agree of Chomsky 2001). P₁ wins the competition against P₂ and obtains the 1^st /2^nd person marking by moving to [spec, βP]. When P₂ relates to β, β’s person features are already used up by P₁, so P₂ cannot obtain 1^st/2^nd person marking and its person marking is reduced to the 3^rd person, a default value. P₂ is clearly underprivileged in this context. If P₁ is dative and P₂ accusative, this derivation leads to effects visible in (9–10).

Although not all languages that have pronominal clitics force clitic clustering, specifically in constructions involving climbing, the formation of the cluster is a strong tendency.⁹ Particular languages respect PCC slightly differently; below two most frequent varieties are illustrated, the strong PCC and the weak PCC (Pancheva & Zubizarreta 2018: 1306–1309):

In languages/varieties with strong PCC the direct object must be 3^rd person, while in the languages/varieties allowing for weak PCC, the direct object can be 1^st or 2^nd person if the indirect object is 1^st or 2^nd person.¹⁰

If a given language has pro_DFs but its pronominal clusters do not obey any variant of the PCC, such pronouns tend to be described as deficient phonologically and morphologically but not syntactically. Polish was said to provide an example of such a circumstance.

2.1 Pronominal clusters

Franks & King (2000) assume that Polish pro_DFs are phonological enclitics that show syntactic distribution common to other DPs, although they also optionally appear in post- and preverbal clusters similar to South and West Slavic (Franks & King 2000: 156):¹²

This remark reflects the gist of Person Ordering Constraint (POC).

The fact, that both pro_DF orders are acceptable in (19), was taken by Franks & King to indicate that Polish does not observe PCC:

In (19a) the acc pronoun is 2^nd person and it follows a 3^rd person dat, violating both PCC/POC in (10–13). The order in (19b) observes POC, as the local 2^nd person precedes the non-local 3^rd person but it violates PCC. If PCC/POC is regarded as condition on the representation at spell-out, then its inapplicability to Polish is a logical conclusion.

Cetnarowska (2003; 2004) credits the composition of clusters to various factors and explicitly claims that Polish disregards PCC:

For her, (20) is grammatical because the cluster is not placed in the pre-pausal position and it is right-adjacent to the inflected verb. The order within the cluster violates PCC, as in (19a). Cetnarowska’s conclusion, also adopted and highlighted in Haspelmath (2004), is that PCC does not hold of Polish. Her argument is valid with respect to the canonical PCC, seen as a condition on the representation at spell-out but it becomes weaker if PCC is a condition applying intra-derivationally and subsequently obscured by reordering.

Following Aissen (1999) and Haspelmath (2001), Cetnarowska captures data concerning orders of pro_DFs through harmonic alignment of different scales:

Harmonic alignment of deficient pronouns strongly prefers the order where local dative pronouns (more prominent) precede non-local accusative pronouns.

Cetnarowska observes that auxiliary and weak pronominals fall into two sub-groups with respect to the amount of phrase stress they can bear.¹³ The weakest elements (reflexive się and auxiliary clitics -ście, -śmy) do not carry stress at all, while the stronger ones (pronominal mu, jej, go and conditional by) can carry secondary or tertiary stress. Cetnarowska proposes that cię be added to the set of weaker elements. She claims that a diagnostic for the strength of the clitic comes from its ability to support AUX clitics. Cetnarowska shows that while cię cannot support AUX clitic -śmy in (13), the stronger pro_DF mu supports -ś in (12):

On the basis of an informal poll, we add the 1^st person dat mi and the 2^nd person dat ci to the same group as cię:

In sum, four categories of pronominal elements emerge in the prosody sensitive classification: the weakest ones cannot accept any degree of prosodic highlighting (mi ‘1p.dat’, ci ‘2p.dat’, cię ‘2.P.acc’ plus person/number agreement AUX -ś ‘2p.sg, -śmy ‘1p.pl’, -ście ‘2p.pl’) and się REFL, the ‘stronger’ pro_DFs can accept secondary stress (mu ‘3.sg.m.dat’, go ‘3.sg.m.acc’ plus the conditional auxiliary -by ‘would’) and the strong pronouns, which can either bear full word and phrase stress or secondary stress (was ‘you.2^nd pl.acc’, jego ‘him.acc’, ją ‘her.acc, etc.).

This four-way distinction is also confirmed in Jung & Migdalski (2019). On the assumptions above, the prosodic pattern of Polish impacts the order in the cluster:

Example (27a) sounds better with jej ‘3.f.dat’ bearing secondary stress in the final position than (27b) due to Selkirk’s (1995) observation that unstressed function words at the end of the prosodic domain should be penalised. At the same time the universal thematic (case) hierarchy in (21b) favours the dat > acc order in (27b). It apparently loses out to the prosodic condition. Yet, it seems, not every pronominal cluster is relevant for PCC, (Cetnarowska 2003: 12):

The strong pronoun (jej ‘her.dat’) is placed in highlighted and destressed positions, used in isolation and individually modified. It also topicalizes/scrambles across the boundary of the subjunctive clause. It is thus different from pro_DFs and it is the ordering of pro_DFs only, that contributes to the discussion of PCC, destressing is not enough. We focus on pro_DFs which have morphologically-marked strong counterparts, see (17): 1dat.sg (mi), 2dat.sg (ci), 2acc.sg (cię), 3dat.sg.ms (mu) and 3acc.sg.ms (go).

This section outlines idiosyncrasies of Polish pro_DFs. Four conclusions come to the fore. First, Polish pro_DFs show ill-disciplined distribution, distinct from the clitic-second position and the adverbal position.¹⁴ Second, pro_DFs still show some residue of special syntax, i.e. they tend to be right-adjacent to the verb. Third, there is a partition within the class of deficient pronouns between 1^st and 2^nd pronouns and 3^rd person masculine pronouns, with the latter able to accommodate secondary stress and the former unable to show any individual highlighting. This distinction is relevant not only for the phonological and morphological properties but also for syntax. Fourth, an arrangement of deficient pronouns in which 3^rd ms. dat precedes 2^nd acc in (19–20) flouts PCC on the assumption that this constraint is representational and applies at the surface.

2.2 Research questions

In view of the discussion above, we formulated research questions helping determine the status of pro_DF clusters in Polish. First, we check if the cluster is sensitive to the local person – 3^rd person distinction. If it is, then the cluster-internal order is driven by the PCC to some extent. Second, we check if the cluster allows for both members to be local persons and if so, in what order they preferably appear (1^st – 2^nd and 2^nd – 1^st). This is a diagnostics for weak PCC. Third, we check for the case order preference; to what extent the dat- acc order is fixed and whether it is alternated with local and non-local persons to the same extent. More specific questions pertaining to the experiments are detailed below.

3.1 The logic of the design

Our experiment was designed to test acceptability of pro_DF clusters in Polish to determine if they show acceptability patterns predicted by the PCC/POC. As discussed in previous section, we selected five deficient pronouns to be tested in our experiments which make up 5 clusters presented in Figure 2. Since we tested all five clusters in two orders, dat-acc and acc-dat, the figure presents compatibility of all cluster options with the PCC/POC. The predictions of the PCC presented in the left column are the same for both pronoun orders. As for the POC, for dat-acc order (left column), the predictions of this constraint align with those of the PCC. For acc-dat order (right column), POC provides a different set of predictions than PCC, as shown. Note that here, we take PCC and POC as they are defined in (11) and (13) respectively, so as purely observational constraints, not as syntactic rules, which would make us commit to specific theory-internal assumptions, depending on the syntactic incarnation of these rules.

Figure 2

The list of pro_DF clusters in Polish and their compatibility with weak/strong PCC and POC.

Importantly, depending on the nature of PCC and POC and the approach one takes on grammaticality, we would expect different levels of unacceptability caused by violation of these rules. In a traditional sense, violation of core syntactic rules should result in categorical unacceptability (ungrammaticality, e.g., Adger 2003).¹⁵ However, syntactic constraints such as PCC/POC could possibly have a gradable effect, whereby incompatibility should result in milder unacceptability.¹⁶ Such gradient acceptability can be potentially modeled via gradient grammar (e.g., Keller 2000; Sorace & Keller 2005; Featherston 2007). Yet, as one of the reviewers points out, gradience in acceptability of our clusters could also be attributed to discourse/processing effects. We test this hypothesis in Exp3 and, as we explain in the discussion section, discourse effects partially account for lower acceptability of only some of our clusters, while the rest of the effects we obtained is due to person constraints.

Since a comparison of all 5 clusters within one paradigm could compromise the naturalness and plausibility of the sentence materials, we divided the clusters into 3 sub-experiments, called Exp1, Exp2, and Exp3. Figure 3 presents the summary of variables used in all experiments. Exp1 and Exp2 were based on two binary variables (2 × 2, pronoun cluster and pronoun order) while Exp3 was built on one variable (pronoun order).

Figure 3

A summary of variables used in Exp1–3.

In the next section, we present the materials used in all three sub-experiments.

3.2 Materials and survey construction

Each participant completed a survey of 84 sentences: 30 experimental items and 54 fillers which were pseudorandomized to avoid the same condition appear in succession. Each participant rated 3 tokens of each condition to avoid the risk of noise and increase the statistical power of the analysis. The survey included experimental items of all 3 experiments described above. For Exp1 and Exp2, we created 12 lexically matched sets (of 4 conditions) each, and for Exp3 we created 6 such sets (of 2 conditions). The items were distributed across 4 experimental lists using a Latin square procedure so that participants saw a unique lexical item in each condition. Each list included 12 items of Exp1, 12 items of Exp2 and 6 items of Exp3 (3 items per condition in each case), which sums up to 30 experimental items in total. Since Exp3 included only 2 conditions, its items were distributed into 2 experimental lists and were copied to the other 2 lists, which means that there were two times more observations per condition for Exp3 than Exp1–2.

Exp1 included experimental sentences with clusters 1p.dat – 2p.acc mi cię and 1p.dat – 3p.acc mi go in two orders dat-acc and acc-dat. To facilitate the processing and understanding of the target sentences, a context sentence (typically a question) with referents for the pronouns was provided for each item in the survey. Importantly, the context questions were neutral in terms of information structure, and they did not impose any specific pronoun order. Experimental items in Exp1 included verbs which take dat and acc object, namely polecić (recommend), opisać (describe), pokazać (show), zarekomendować (recommend), przypomnieć (remind), obrzydzić (make disgusting), zasłonić (conceal), przynieść (bring), sprowadzić (bring), przywiać (bring over), przedstawić (introduce), powierzyć (entrust). Since we had to keep the animacy of the pronoun referents the same (dat-animate, acc-animate), the selected verbs had to be compatible with two animate objects, which limited our choices, see (29)–(30).

Exp2 included sentences with clusters 2p.dat – 3p.acc ci go and 3p.dat – 2p.acc mu cię in two orders dat-acc and acc-dat with a context sentence. Similarly to Exp1, materials in Exp2 included double object verbs, namely przedstawić (introduce), przyprowadzić (bring over), pokazać (show), zachwalać (praise), narysować (draw), zaprezentować (present), wytatuować (tattoo), dodać (add), wykraść (steal), przeznaczyć (predestine), przytrzymać (hold), przydzielić (allocate). Also in this experiment, verbs had to be compatible with two animate objects, see (31)–(32).

Exp3 included sentences with one cluster 3p.dat – 2p.acc mu go in two orders dat-acc and acc-dat, with a context sentence. Since in this experiment, there was no animacy restriction on pronoun referent (no 1/2p pronouns), we used double object verbs which are compatible with animate dat object and inanimate acc object, to introduce more variation to our experimental items. Importantly, we did not include animacy as a variable in any of the experiments and in Exp1–2, we used only same animacy referents for pronouns (animate) to avoid any potential effect it might have on the order of pronouns. However, since animacy effects are typically found for nouns that are inherently marked as animate/inanimate, we did not predict pronouns (that only refer to animate/inanimate nouns) to show the same effect. Thus, although in Exp3 different animacy referents were used, we believe it did not affect the ratings (see discussion in 3.6.2). We used the following verbs: pożyczyć (lend), sprzedać (sell), dostarczyć (deliver), ukraść (steal), przynieść (bring), kupić (buy), see (33).

Filler sentences consisted of 15 acceptable, 15 unacceptable, and 24 unrelated sentences. The acceptable/unacceptable fillers included various pronouns (strong pronouns, plural pronouns etc.), and in some cases, also pronoun clusters, but unacceptability was not related to the use of pronouns/clusters, as we aimed to mask the purpose of the experiment. Also, acceptable/unacceptable fillers served as accuracy control, i.e., it was used to potentially exclude participants whose accuracy was too low. The unrelated fillers consisted of simple sentences without pronouns. All filler sentences were provided with a context sentence, see (34)–(36).

3.3 Predictions

3.4 Participants and procedure

Our survey was completed by 59 participants, all of whom are students at the Faculty of English, Adam Mickiewicz University in Poznań. In exchange for completing the survey, participants received credit points required for the ‘WA experiments’ course. For all participants mean accuracy was calculated based on their ratings for acceptable/unacceptable fillers. Namely, incorrect responses for ungrammatical fillers included ratings between 4–7, and for grammatical fillers, ratings between 1–4. Out of 59 participants, 3 were excluded in order to balance the number of participants per experimental list. The choice of participants to be excluded was based on their mean accuracy, i.e. one participant with 87% accuracy and 2 participants with 90% accuracy were excluded. This resulted in a total of 56 participants, whose mean accuracy was 97%, mean age 21.5, SD=1.55 (42 women, 12 men, 1 non-binary, 1 undisclosed).

The experiment was administered online using Google Forms. Each sentence and its context were presented on their own presentation screen together with a scale from 1 to 7 (1 – całkowicie nieakceptowalne ‘completely unacceptable’, 7 – całkowicie akceptowalne ‘completely acceptable’). The context sentence and the experimental item were presented in a form of a dialog between speakers A and B. The participants were instructed to rate acceptability of B’s response alone, without making any judgements as to whether the response is a plausible answer to the question/context. The instruction made it clear that the A’s utterance is only to facilitate the understanding of the B’s response and in principle, the two should be plausibly connected. Since we wanted to control for the effect of context on pronoun order, we decided to provide a fixed discourse-neutral context for each sentence. The participants were to rate the sentence by clicking on a bullet with the corresponding number (1–7). An example trial is presented in Figure 4 together with a short instruction that accompanied each trial. The full instruction given to participants at the beginning of the survey is provided in the Appendix.

Figure 4

An example experimental trial with a short instruction which translates to “Rate how acceptable the Polish sentence in B is. Follow your intuition”.

3.5 Results

3.5.2 Results: experiment 1

The results obtained in Exp1 yielded a significant main effect for pronoun cluster and pronoun order. Overall, cluster 1p.dat – 3p.acc mi go was rated higher than the cluster 1p.dat – 2p.acc mi cię (z = –4.318, p < 0.001) and the pronoun order dat-acc was rated higher than acc-dat (z = –7.088, p < 0.001). Also, a significant interaction was found for pronoun cluster and pronoun order (z = –2.077, p = 0.038). Pairwise comparisons showed that acc-dat order was rated significantly lower than dat-acc in both clusters, 1p.dat – 3p.acc mi go (z = –5.250, p < 0.001) and 1p.dat – 2p.acc mi cię (z = –7.089, p < 0.001). Also, cluster 1p.dat – 3p.acc mi go was rated significantly higher than 1p.dat – 2p.acc mi cię in acc-dat order (z = –6.164, p < 0.001), but not in dat-acc order (z = –1.215, p = 0.617). Figure 5 presents the proportion of ratings per condition. We indicate significant effects with (*) and no effect with ( ).

Figure 5

Proportion of ratings per condition in Exp1.

3.5.3 Results: experiment 2

In Exp2, a significant effect was found for both pronoun cluster and pronoun order. Generally, cluster 2p.dat – 3p.acc ci go was rated higher than 3p.dat – 2p.acc mu cię (z = 2.688, p = 0.007) and dat-acc order was rated higher than acc-dat (z = –5.462, p < 0.001). Also, a significant interaction was found for pronoun cluster and pronoun order (z = –2.045, p = 0.041). Pairwise comparisons showed that acc-dat order was rated significantly lower than dat-acc in cluster 2p.dat – 3p.acc ci go (z = –5.496, p < 0.001) but not in cluster 3p.dat – 2p.acc mu cię (z = –2.279, p = 0.103). Also, cluster 2p.dat – 3p.acc ci go was rated significantly higher than 3p.dat – 2p.acc mu cię in dat-acc order (z = 2.843, p = 0.023), but not in acc-dat order (z = 1.548, p = 0.409). Additionally, the results showed no significant difference between 2p.dat – 3p.acc ci go in acc-dat order and 3p.dat – 2p.acc mu cię in dat-acc order (z = –0.478, p = 0.964). We present the proportion of ratings in Figure 6.

Figure 6

Proportion of ratings per condition in Exp2.

3.5.4 Results: experiment 3

A significant effect of pronoun order was obtained in Exp3, namely dat-acc order was rated higher than acc-dat (z = –6.768, p < 0.001). We present the proportion of ratings in Figure 7.

Figure 7

Proportion of ratings per condition in Exp3.

3.6 Discussion of the results

3.6.1 Descriptive data

Based on descriptive results of EXP1–3, we identify tree degrees of pronoun cluster acceptability, represented in Figure 8, where mean scores and medians are provided for each cluster in both orders. The green highlight marks fully acceptable clusters (with mean rating between 6–7 and median 7), the light green highlight marks still acceptable but degraded clusters (mean between 5–6, median 5–6) and the orange highlight marks marginally acceptable clusters (mean 4–5, median 5). Generally, clusters in dat-acc order belong to the first group, with the exception of cluster 3dat-2acc mu cię. The second group comprises clusters in acc-dat order except clusters with 2acc pronoun cię, and plus a 3p.dat –2p.acc mu cię cluster. The lowest rated group consists of clusters with 2acc pronoun cię in acc-dat order. These results show that clusters following basic dat-acc order and abiding by weak PCC are fully acceptable, while those PCC-incompatible (3p.dat – 2p.acc mu cię), are only slightly degraded. Also, their scrambled acc-dat counterparts are rated as slightly degraded, but when the acc pronoun involves 2acc cię, a more deficient weak pronoun, those ratings are even lower and make the clusters only marginally acceptable.

Figure 8

Summary of means and medians for EXP1–3.