Relative Clauses in Native Lower Sorbian and the Relativizer how

Abstract

Native Lower Sorbian, an endangered West Slavic minority language spoken in Germany, possesses a relative clause formation strategy employing the invariant relativizer ak and optional resumption. The focus of this paper lies on the status of ak. In other languages that have them, invariant relativizers are drawn from the set of complementizers, wh-words, or demonstratives. ak seems to differ in that respect because it belongs to neither category. In this paper, I argue that ak is not an outlier. Instead, ak is a variant of the manner wh-word kak ‘how’ in its non-manner use as a complementizer. After I show how the complementizer kak differs from the wh-adverb kak and that relative clauses in Native Lower Sorbian feature empty operator movement, I argue that the empty operator sitting in SpecCP triggers a rule partially deleting the complementizer kak. More specifically, the rule elides the initial [k] of kak, reducing it to ak. This makes Native Lower Sorbian similar to Bern German or West Frisian, both of which also feature the partial deletion of a complementizer in the presence of a moved element in SpecCP. Furthermore, Native Lower Sorbian is yet another language where how has a non-manner use.

1. Introduction

Many languages use invariant relativizers to form relative clauses. In (1), this is illustrated with data from English, Bavarian, Modern Greek, and Early New High German.

(1)	a.	English.
		The gorilla that the lunatic attacked was eating a banana.
	b.	Bavarian.
		I	geb’s	dera	Frau,	wo	d’Muich	bringd
		I	give=it	the	woman	rel	the=milk	brings
		‘I give it to the woman that brings the milk.’
		(Bayer, 1984, p. 216)
	c.	Modern Greek.
		I	jineka,	pu	agapao,	meni	stin	Kriti.
		the	woman	rel	I.love	lives	in	Crete
		‘The woman that I love a lot lives in Crete.’
	d.	Early New High German.
		an	dem	heiligen	evangelio,	so	wir	hiute	lesen
		on	the	holy	gospel	rel	we	today	read
		‘in the holy gospel that we read today.’
		(Behaghel, 1928, p. 731)

There is a generalization covering the shape of invariant relativizers, formulated in (2).

(2)	Invariant relativizers are drawn from the set of function words.

By function words, I refer to all words that have a grammatical function instead of a lexical meaning and that belong to the set of closed word classes. Such word classes include complementizers, wh-words, and simple adverbs. The data in (1) all conform to the generalization in (2). As for English, that is also the standard declarative complementizer. Both Modern Greek and Bavarian use a relativizer that is also used as the locative interrogative adverb ‘where’, namely pu and wo. And the Early New High German invariant relativizer so is also used as a manner adverb (similarly to English so).

Native Lower Sorbian, an endangered West Slavic language spoken in the southeast of the German province Brandenburg, also possesses a system of relative clause formation whereby relative clauses feature an invariant relativizer. This relativizer is ak, which is optionally accompanied by resumption, as shown in (3).

(3)	Ta	žeńska,	ak	ja	som	(ju)	wiźeł,	bydli	Barlinju.
	the	woman	rel	I	am	her	seen	lives	Berlin.loc
	‘The woman that I saw lives in Berlin.’

This relativizer ak poses a problem. According to the generalization in (2), ak should be used as a function word in Native Lower Sorbian. But this is not the case. ak is neither a regular complementizer in Lower Sorbian (this is až ‘that’), nor is ak a wh-word (the one closest to ak is kak ‘how’), nor is ak a simple adverb (the closest to ak would be tak ‘so’). The problem this paper therefore addresses is whether ak is really a problem for the generalization in (2).

In this paper, the claim I want to make is that ak does not violate the generalization in (2). More specifically, I argue that ak is a variant of the wh-word kak ‘how’ in its non-manner use as a complementizer. Since ak is obviously only partially identical to kak (it lacks the initial k), the question arises of how to explain the absence of k. In this paper, I argue that the initial k of the complementizer kak is deleted whenever an empty operator occupies SpecCP, as shown in (4).

(4)

This deletion operation seems ad hoc and raises questions about the interaction between syntax and phonology. I address both concerns and show that they are unfounded.

This paper is structured as follows. In Section 2, I give a brief overview of Lower Sorbian, and in particular about the version reported here, Native Lower Sorbian, and the data sources for this paper. In Section 3, I compare the literature on relative clauses in Lower Sorbian with the situation found in Native Lower Sorbian and briefly review the literature on the status of ak. In Section 4, I discuss the distribution and properties of ak vis-à-vis the properties of the three categories mentioned above and argue that the properties of ak all point to an analysis of ak as a variant of the complementizer kak. In Section 5, I give a detailed account of this analysis and discuss its implications. In Section 6, I discuss a potential problem for the analysis, and in Section 7, I compare Lower Sorbian with Czech. Section 8 concludes this paper.

2. Background on Lower Sorbian

Lower Sorbian is a West Slavic indigenous minority language spoken in a few villages1 of the municipality Peitz, which is located in the Spree-Neiße district of the German province Brandenburg, as shown in Figure 1. Since intergenerational transmission of Lower Sorbian as a native language stopped after World War II, Lower Sorbian counts as severely endangered according to the language endangerment status of UNESCO (it is actively spoken only by grandparents and older generations, with younger generations having only passive knowledge). There is no official number of speakers of Lower Sorbian. Frequently, it is said that there are 7000 speakers of Lower Sorbian. This number eventually goes back to Jodlbauer et al. (2001) and in particular to their chapter 2. This number, however, must be treated with caution. First, the number is based on results obtained from fieldwork in the mid-nineties, that is, it represents, at best, the state of Lower Sorbian from 30 years ago. Second, as the researchers frequently stress themselves, the number is only an estimation. Third, the number is artificially boosted by including Cottbus (Chóśebuz), which they estimate had a population of Lower Sorbian speakers of 1% (approx. 1200 speakers), but this completely ignores the demographic changes in the wake of industrialization that the city underwent after World War II. Based on my personal field trips of the last 10 years and conversations with language activists and other researchers from the community, my estimate is that there are only approx. 200 native speakers left today.

As the title of this paper indicates, I restrict myself to Native Lower Sorbian. By this, I refer to the variety that was natively acquired, either as the first language or as the second language at a very young age. This restriction is important for two reasons. The first reason is that there exists a standardized version of Lower Sorbian, which differs drastically from the native varieties. It is characterized by puristic tendencies to replace loanwords from German with ‘Slavic’ words2, preserve lost morphological categories or formation rules (like the aorist or the formation of the passive with the auxiliary byś ‘to be’ instead of hordowaś ‘to become’), and retain syntactic properties the native varieties have either lost recently (like relative pronouns, more on which is provided in Section 3) or never had to begin with (unrestricted pro-drop, a system without articles). This standard variety is used in basically all Lower Sorbian newspapers, books, and public broadcastings. The second reason for this restriction is that this standard variety is not only used in media, but it is also actively spoken by what one might want to call new speakers (Hornsby, 2015), that is, people who acquire a typically artificial version of an endangered minority language as a second language in order to preserve it. In the case of Lower Sorbian, these are typical Germans but also speakers of other Slavic languages.

Since all types of media use the standard variety, I had to rely on other sources to obtain data from Native Lower Sorbian. The first source is a corpus of conversations with native speakers of Lower Sorbian. This corpus is an outgrowth of the DoBeS project funded by the Volkswagen Stiftung from 20-20, of which the project on Native Lower Sorbian was a subproject3. This corpus is publicly available online4 and comprises approx. 100 h of transcribed and translated conversations of 80 native speakers of Lower Sorbian. The second source of data is judgments from a native speaker that I collected in the summer of 20245.

3. Relative Clauses in Native Lower Sorbian and the Status of ak

3.1. Relative Clauses in Lower Sorbian: The Standard View

According to the grammar of Lower Sorbian in general (Hauptmann, 1761/1984; Mucke, 1891; Janaš, 1984) and articles on relative clauses in Lower Sorbian in particular (Gutschmidt, 1999; Bartels & Spiess, 2012), relative clauses are formed by using one of four relativization strategies.

The first strategy is to use relative pronouns. These are formed based on the adjectival interrogative pronouns kótr-6 or kak-7, to which the suffix ž is added, cf. (5).

(5)	Ta	žeńska,	kótruž/kakuž	ja	som	wiźeł,	bydli	Barlinju.
	the	woman	who.acc.fem.sg	I	am	seen	lives	Berlin.loc
	‘The woman that I saw lives in Berlin.’

This strategy is well-known from other languages: the relative pronoun “stands for” the relativized element inside the relative clause and bears the morphological marking required there, agrees with the head noun for number and gender, and is moved from its TP-internal position to a clause-initial position.

All other three strategies are characterized by the use of an invariant relativizer plus resumption and only differ with respect to the type of invariant relativizer that is used. The second strategy employs the invariant relativizer kenž, the third the invariant relativier což, and the fourth the invariant relativizer ak8. The three strategies are shown in (6).

(6)	a.	Ta	žeńska,	kenž	ja	som	(ju)	wiźeł,	bydli	Barlinju.
		the	woman	rel	I	am	her	seen	lives	Berlin.loc
	b.	Ta	žeńska,	což	ja	som	(ju)	wiźeł,	bydli	Barlinju.
		the	woman	rel	I	am	her	seen	lives	Berlin.loc
	c.	Ta	žeńska,	ak	ja	som	(ju)	wiźeł,	bydli	Barlinju.
		the	woman	rel	I	am	her	seen	lives	Berlin.loc
		‘The woman that I saw lives in Berlin.’

The presence of resumptive pronouns depends on morphological and thematic properties of the head noun vis-à-vis those of the relativized element (cf. Mucke, 1891; Gutschmidt, 1999; Bartels & Spiess, 2012), which are irrelevant for the purpose of this paper.

The strategies are not equally frequently used in the standard language. As shown in

Table 1. Numbers for the relativization strategies in Standard Lower Sorbian (based on the Lower Sorbian text corpus).

kótr-/kak- + ž	kenž	ak	což
9032	5823	≈2000	≈150

9, relative pronouns10 are used most frequently, followed—in descending order—by the use of kenž, ak11, and což12.

3.2. Relative Clauses in Native Lower Sorbian

When turning to Native Lower Sorbian, the situation is very different. Syntactic research carried out in the 1980s in the context of the Sorbischer Sprachatlas (Sorbian Linguistic Atlas, Fasske, 1995), which aimed at a description of the natively spoken varieties, included a section on relative clauses (chapter 35). For Lower Sorbian, the finding was that of the four strategies only one basically survived, namely, the one employing ak (Fasske, 1995, pp. 169–179). The numbers are given in

Table 2. Numbers for the relativization strategies in Native Lower Sorbian (based on Fasske, 1995).

kótr-/kak- + ž	kenž	ak	což
5	0	157	12

.

This finding is corroborated by the DoBeS corpus.

Table 3. Numbers for the relativization strategies in Native Lower Sorbian (based on the DoBeS corpus).

kótr-/kak- + ž	kenž	ak	což
5	13	≈850	≈30

shows the numbers for the relative pronouns, kenž, což, and ak.

As the table shows, the use of ak as a relativizer with approx. 850 hits (cf. note 11) greatly exceeds the numbers of all other strategies. Of the five hits for relative pronouns (all using the strategy kót(a)ry + ž), four are from speakers that are heavily influenced from the standard language; the one informant is a linguist, and the other worked for radio broadcasting. kenž is only used when the speakers quote from archaic fixed expressions, like old folk songs or the Lord’s Prayer. což also occurs; but with approx. only 30 hits (cf. note 12), it clearly represents a minor strategy. It is therefore fair to conclude that relativization with the relativizer ak plus resumption is the default and preferred strategy to form relative clauses in Native Lower Sorbian13.

3.3. The Status of ak

The use of ak as a relativizer is not a recent development of Lower Sorbian. A search in the historical text corpus of Lower Sorbian (comprising nearly all written texts up until 1945) gives the results shown in

Table 4. Numbers for the relativization strategies in Standard Lower Sorbian before 1945 (based on Lower Sorbian text corpus).

kótr-/kak- + ž	kenž	ak	což
47,440	32,285	≈9700	≈750

.

As the numbers show, ak is used less frequently than relative pronouns and kenž. Still, the use of ak as a relativizer is attested already in the oldest printed document of Lower Sorbian, namely, in Albin Moller’s hymnal and catechism from 1574. In it, one finds the following example (Schuster-Šewc, 1959, p. 172)14:

(7)	O	Tij	neymotzneyssy	nymerny	Knijss	Bog	/	kenss	ssij
	O	ty	nejmocnjejšy	njeměrny	kněz	bog		kenž	sy
	oh	you	mightiest	immortal	Lord	God		rel	are
	Pomotznyk		schyknym	ak   se	natebe	spuszayu.
	pomocnik		wšyknym	ak   se	na tebje	spušćaju
	helper		for all	rel refl	on you	trust
	‘Oh you almighty immortal Lord God, who you are the helper for all who trust you.’

Given that the use of ak as a relativizer is so ubiquitous in Lower Sorbian, it comes as a bit of a surprise that the issue of how to analyze ak has hardly been addressed so far. In total, there are two views regarding the status of ak.

The first is given by Schuster-Šewc (2024) in his etymological dictionary of Sorbian. According to him, ak is a reduced variant of ako, which, in turn, presumably derives from the “originally emphatic particle *a, extended via an old interrogative element *kъ or *ko”15. The loss of the final o is similar to the loss of the final o of the inflectional suffixes -ogo and -ego16, which get reduced to -og and -eg, respectively. There are various problems with this view. On the one hand, Albin Moller’s hymnal and catechism contains numerous instances of ak, but none of ako; if ako represented the original stage, it is surprising that this stage is not represented in the oldest available document. On the other hand, this document nevertheless contains variants of -ogo and -ego both with the final o and without it; so, whatever is responsible for the dropping of o in -ego and -ogo is different from what connects ak and ako17. Furthermore, neither the emphatic particle a nor the suffix *kъ or *ko are otherwise attested in Lower Sorbian18.

The second idea is only sketched in Mucke’s comparative grammar of Lower Sorbian and its dialects (Mucke, 1891). He too assumes that ako is the older variant and connects it to jako, which he assumes is an older version of the wh-adverb kak ‘how’ (Mucke, 1891, pp. 396, 456). But he also connects it functionally to the German demonstrative pronoun so ‘so’, which corresponds to tak in Lower Sorbian (Mucke, 1891, p. 431). The analysis faces two problems. First, it leaves the missing initial consonant (either j or t) of ako unaccounted for. Secondly, in addition to the fact that Albin Moller’s hymnal and catechism contains no instance for ako, it neither contains any instance of jako, the alleged source for ak. Instead, the wh-word for ‘how’ used there is kak only.

In sum, the only two existing analyses for the status of ak neither capture the shape of this relativizer nor are their accounts for the diachronic development compatible with the earliest historical records from Lower Sorbian.

4. ak as a Variant of kak

Having shown so far that ak is the standard relativizer in Native Lower Sorbian and that the status of ak is unresolved, I now turn to the analysis of ak. I argue that ak is a variant of the wh-word kak ‘how’ in its non-manner use of a complementizer. I first show that kak ‘how’ can be used as a complementizer in Native Lower Sorbian. I then argue that since the complementizer kak behaves differently from the manner wh-adverb kak, the complementizer kak sits in C°. Finally, I argue that based on this difference, ak cannot be a variant of the demonstratives kak- ‘which’ or tak- ‘such’ either.

Similar to other Slavic (Šimík & Sláma, 2023) and non-Slavic languages, like English (Legate, 2010) and German (Hinterhölzl, 2023), Native Lower Sorbian allows the usage of the manner wh-adverb kak as a complementizer. Three examples from the DoBeS corpus are given in (8).

(8)

a.

Pón

sy

mógał

wiźeś,

kak

su

byli

how

karony.

then

are

could

see

how

are

been

here

crows

‘Then you could see that there are crows here.’

b.

Gaž

wón

to

wiźi,

kak

te

faraje

su

musali

to serbske

wuknuś.

as

he

it

sees

how

the

pastors

are

must

Sorbian

learn

‘As he saw it that the pastors had to learn Sorbian.’

c.

To

jo

naša

mama

wulicowała,

kak

wóni

su šli

pó jsy.

it

is

our

mum

told

how

they

are gone

in village

‘Our mother used to tell us that they were walking in the village.’

In all three examples, a manner reading is unlikely due to the semantics of the verb. In the first example (8a), the speaker only reports that there were crows but not in what way they existed. Similarly, in the second example (8b), the speaker only reports that the priests arrived but not in what way they arrived. Particularly telling is the third example (8c). The speaker starts reporting that his mother and her friends used to walk along the road in order to find out who had the most impressive Christmas pyramid. The manner of walking is clearly irrelevant to achieving this aim.

Apart from this argument from interpretation, there is a further and stronger syntactic argument that the kak’s in (8) are complementizers. As observed by Pankau (2024), Native Lower Sorbian belongs to the few languages where the manner adverb how can agree with the subject with respect to number and gender (the other languages reported being Övdalian (Vangsnes, 2015) and Luyia (Carstens & Diercks, 2013)). Two examples of this phenomenon are given in (9). For ease of understanding,

Table 5. Inflectional paradigm for kak in Native Lower Sorbian19.

	sg			du	pl
	m	neut	fem
nom	kaki	kake	kaka	kakej	kake
acc	kakeg(o)/kaki		kaku	kakeju/kakej	kakich/kake
dat	kakem(u)		kakej	kakima	kakim
gen	kakeg(o)		kakej(e)	kakeju	kakich
ins	kakim		kakeju	kakima	kakimi
loc			kakej		kakich

gives the inflectional paradigm for kak.

(9)	a.	Kake	su	wuglědali?
		how.pl	are	looked
		‘How did they look?’
	b.	Kaka	wóna	wuglěda?
		how.fem.sg	she	looks
		‘How does she look?’

Importantly, when changing the examples in (8) from kak to agreeing versions of kak, according to my informant, the meaning of the sentences changes, as shown in (10).

(10)

a.

Gaž

wón

to

wiźi,

kake

te

faraje

su

musali

to serbske

wuknuś.

as

he

it

sees

how.pl

the

pastors

are

must

Sorbian

learn

‘As he saw it in what way the pastors had to learn Sorbian.’

b.

To

jo

naša

mama

wulicowała,

kake

wóni

su

šli pó

jsy.

it

is

our

mum

told

how.pl

they

are

gone in

village

‘Our mother used to tell us in what way they were walking in the village.’

As indicated, agreeing kak’s only have a manner reading. In (10a), kake only refers to the manner in which the priests arrived. Equally, in (10b), where kake only refers to the way in which the girls were walking along the road. The contrast between (8) and (10) is summarized in (11).

(11)	The manner wh-adverb kak can agree, the complementizer kak must not agree.

In order to capture this difference, I assume that the manner wh-adverb kak sits in SpecCP, whereas the complementizer kak sits in C°, as shown in (12).

(12)	a.	manner wh-adverb kak	b.	complementizer kak

Given this diagnostic, one arrives at different predictions concerning the status of ak, as formulated in (13).

(13)	a.	If ak is the manner wh-adverb kak, ak can agree.
	b.	If ak is the complementizer kak, ak must not agree.

As the examples in (14) show, the second prediction is correct: relative clauses with agreeing ak are highly ungrammatical.

(14)	a. *	Ta	žeńska,	ake	wóne	su	(ju)	wiźeli,	bydli	Barlinju.
		the	woman	rel.pl	they	are	her	seen	lives	Berlin.loc
		‘The woman that they saw lives in Berlin.’
	b. *	Ten	muski,	aka	wóna	jo	(go)	wiźeła,	bydli	Barlinju.
		the	man	rel.fem.sg	she	is	him	seen	lives	Berlin.loc
		‘The man that she saw lives in Berlin.’

Having shown that ak is a variant of the complementizer kak and not of the manner wh-adverb kak, the relevant diagnostic (agreement vs. non-agreement) can also be used to show that ak is not a variant of the demonstrative kak or tak either. There are two options to consider. ak could either be analyzed as a variant of the demonstrative tak or kak that is fronted to SpecCP, that is, ak would be analyzed as a relative pronoun, as shown in (15) with tak (the variant with kak would be identical, except that kak were fronted).

(15)

Alternatively, if one assumes that relative clauses to be derived through the Matching Analysis or the Promotion Analysis (Pankau, 2018), whereby the relativized element contains a silent representation of the head noun, ak could be taken as a DP-internal demonstrative, as shown in (16) again with tak (also, here, the variant with kak would be identical, except that kak were fronted).

(16)

The reason that excludes both options is identical to the one that excludes the option that ak is the manner wh-adverb kak, namely, agreement20. Demonstratives always agree in Native Lower Sorbian, as shown in (17)–(18) based on tak. When used pronominally, they agree with their antecedent (cf. 17), when used as determiners, they agree with the noun (cf. 18) (the inflectional paradigm for tak is identical to the one for kak).

(17)	a.	Ja	znam	jadnog’	muskeg’,	a	wóna	teke	takog’/*  tak	znajo.
		I	know	a	man	and	she	also	such+agr/ such-agr	knows
		‘I know a man, and she also knows such a man.’
	b.	Ja	mam	jadnu	žeńsku,	a	wón	teke	taku /*  tak	ma.
		I	have	a	woman	and	he	also	such+agr/ such-agr	has
		‘I have a wife, and he also has such a wife.’

(18)	a.	taki /* tak	muski.
		such+agr/such-agr	man
		‘such a man’
	b.	taka /* tak	žeńska.
		such+agr/such-agr	woman
		‘such a woman’

Consequently, if ak had a demonstrative source, it should be able to agree. But as the examples in (19) show, ak’s that agree with the head noun are as ungrammatical as the examples in (14), where ak agrees with the subject of the relative clause.

(19)	a. *	Ta	žeńska,	aka	wóne	su	(ju)	wiźeli,	bydli	Barlinju.
		the	woman	rel.fem.sg	they	are	her	seen	lives	Berlin.loc
		‘The woman that they saw lives in Berlin.’
	b. *	Ten	muski,	aki	wóna	jo	(go)	wiźeła,	bydli	Barlinju.
		the	man	rel.m.sg	she	is	him	seen	lives	Berlin.loc
		‘The man that she saw lives in Berlin.’

In sum, since an analysis of ak as a variant of the complementizer kak equates the invariant shape of ak with that of the complementizer kak, it is superior to an analysis of ak as having a demonstrative source where this invariance must be stipulated.

5. A Partial COMP Deletion Analysis

Having shown that ak behaves like the complementizer kak, I will turn to the obvious shortcoming equating these two has, namely, that ak is phonologically different from kak. The solution to this problem comes from a closer inspection of the structure of relative clauses in Native Lower Sorbian. I, therefore, first investigate their structure and then show how this paves the way for understanding the phonological difference.

5.1. The Structure of Relative Clauses Based on ak in Native Lower Sorbian

As already mentioned, the formation of relative clauses in Native Lower Sorbian involves ak and optional resumption. Cross-linguistically, two structures are attested for resumption in relative clauses (cf. Korsah & Murphy, 2024 for a recent overview). The resumptive pronoun ju in (20a) could either enter into some kind of binding relation with the head noun, as shown in (20b), or the relation could be mediated by some moved phrase, with the trace being realized as a pronoun, as shown in (20c).

(20)	a.	Ta	žeńska,	ak	ja	som	ju	wiźeł,	bydli	Barlinju.
		the	woman	rel	I	am	her	seen	lives	Berlin.loc
		‘The woman that I saw lives in Berlin.’
	b.	… žeńskai … [CP … jui …].
	c.	… žeńskai … [CP … XPi,1 … __i,1 → jui,1 …].

For Native Lower Sorbian, the available evidence points to the movement structure. First, if binding were involved, relativization should not be island-sensitive in Native Lower Sorbian. But relativization in Native Lower Sorbian is island-sensitive, whether it involves a gap or a resumptive, as (21) shows for a coordinate structure21.

(21) *

Ta

žeńska,

ak

ja

som

[(ju)

a

móju

mamu]

wiźeł,

bydli

Barlinju.

the

woman

rel

I

am

her

and

my

mother

seen

lives

Berlin.loc

‘The woman that I saw her and my mother lives in Berlin.’

Second, movement in relative clauses in Native Lower Sorbian is more restricted than other movements. As shown in (22), Native Lower Sorbian allows both extraction out of DPs, namely, left-branch extraction (LBE, cf. Pankau, 2021), and topicalization from embedded clauses (the example showing this is from the DoBeS corpus).

(22)	a.	Wšykne	su	chwalili,	kaku	my	mamy	[t dobru	wódu ].
		all	are	praised	what	we	have	good	water
		‘All praised what good water we have.’
	b.	Běle	štrumpy	wěm,	až	njejsom	t měła	.
		white	socks	I.know	that	not.am	had
		‘White socks I know that I didn’t have.’

However, relativization of a possessor out of a DP, which would be an instance of LBE, and of a phrase out of embedded clauses is ungrammatical in Native Lower Sorbian, as shown in (23).

(23)

a. *

Ta

žeńska,

ak

znaju

[ jej

gólca],

bydli

Barlinju.

the

woman

rel

I.know

her

boy

lives

Berlin.loc

‘The woman whose son I know lives in Berlin.’

b. *

Ta

žeńska,

ak

se

myslim,

až

som

(ju)

wiźeł,

bydli

Barlinju.

the

woman

rel

refl

I.think

that

am

her

seen

lives

Berlin.loc

‘The woman that I think I have seen lives in Berlin.’

So, with respect to relativization, Native Lower Sorbian shows what Ross (1984) called inner island effects: the domain for the application of a syntactic operation is smaller than the usual domain of islands.

At the same time, there is no overtly moved phrase detectable in relative clauses in Native Lower Sorbian. In order to resolve this paradox, I assume that relative clauses in Native Lower Sorbian feature movement of an empty operator, as shown in (24).

(24)

This captures the island-sensitivity, but not the inner island effects. Although it is beyond the scope of this paper to capture these, let me briefly mention two options. The first is that empty operator movement is subject to different locality restrictions than the movement of non-empty phrases. The second—and, in my view, more promising route to take—is that the empty operator undergoes subextraction (cf. Corver, 2017 for an excellent overview). According to this idea, the empty operator occupies a high specifier or adjoined position inside the phrase corresponding to the relativized element. The operator marks the XP as [+rel] and is then moved to SpecCP (the remnant XP is elided and optionally turned into a pronoun), as shown in (25).

(25)

This makes the structure of the relativized element in Native Lower Sorbian structurally similar to another structure involving subextraction, namely, wat-voor splits in Dutch, where the minimal wh-word wat ‘what’ marks an XP as being [+wh]. Movement can either target the whole phrase or only subextract the wat, as shown in (26).

(26)	a.	Wat	voor	boeken	heeft	Jan t	verkocht?
		what	for	books	has	Jan	sold
	b.	Wat	heeft	Jan	[t voor	boeken]	verkocht?
		what	has	Jan	for	books	sold
		‘What books did Jan sell?’

The reason why this is more promising, in my view, is that the subextraction of wat shows inner island effects, whereas movement of the whole wh-phrase does not (Corver, 1990, 2017). Two such differences are illustrated in (27)–(28). Example (27) involves extraction out of a [+wh]-complement clause, and example (28) involves extraction across negation (Corver, 2017, p. 44).

(27)

a. ?

Wat

voor

boeken

vraag

jij

je

af,

wanneer

Jan

t

gekocht

heeft?

what

for

books

ask

you

refl

prt

when

Jan

bought

has

b. *

Wat

vraag

jij je

af,

wanneer

Jan

[t voor

boeken ]

gekocht

heeft?

what

ask

you refl

prt

when

Jan

for

books

bought

has

‘What books do you wonder when Jan has bought?’

(28)	a.	Wat	voor	boeken	heeft	Jan	niet	t	gelezen?
		what	for	books	has	Jan	not		read
	b. *	Wat	heeft	Jan	niet	[t voor boeken]			gelezen?
		what	has	Jan	not	for books			read
		‘What books didn’t Jan read?’

Whichever of the two options eventually turns out to be on the right track, what moves in relative clauses in Native Lower Sorbian is an empty operator.

5.2. The Analysis

Given the structure for relative clauses in Native Lower Sorbian with a moved empty operator, I want to suggest that it is this moved empty operator that is responsible for the change from kak to ak. More specifically, I suggest that the empty operator moved to SpecCP triggers a phonological operation targeting the complementizer kak occupying C°, namely, the deletion of the initial k of kak, as formulated in (29a) and illustrated in (29b).22

(29)	a.	k → Ø/[CP Op [C° [_ak]] ] k is deleted when it is the first consonant of the complementizer kak
	b.

This analysis captures that ak behaves like kak because ak is kak in disguise. So whatever properties kak has, ak will also have, like the inability to agree.

Since the empty operator deletes only part of the complementizer, I follow Brand and Penner (1995) in referring to this as Partial COMP deletion. Brand and Penner (1995) discuss Bern German in their paper (the variety of Swiss German spoken in Bern), which has an operation similar to k-deletion, namely, d-deletion. d-deletion optionally deletes the initial d of the complementizer dass ‘that’ under the condition (30b), illustrated in (30a-a’).

(30)

a.

De

Hans

hed

nid

gwüsst,

wem

(d)ass

är

t

söll

aalüte.

the

Hans

had

not

known

who

that

he

should

call

‘Hans didn’t know who he should call?’

a’.

Wem

hed

är

gseid,

t’  (d)ass

är

das

t

wiitersägi?

who

had

he

said

that

he

that

tells

‘Who did he say that he should tell that?’

b.

d-deletion may apply if SpecCP is filled by (i) an A’-moved XP or (ii) its trace.

Bern German is not the only language with partial COMP deletion. Another one is West Frisian (Reuland, 1979, pp. 167–171; Tiersma, 1985, pp. 97–101), where the complementizer dat ‘that’ is obligatorily reduced to ‘t under the condition in (31b), as illustrated in (31a-a’).

(31)	a.	Ik	wit  net,	hwa	*dat/*Ø/√‘t	hjirre	so’n	rommel	make hat.
		I	know not	who	that	here	such=a	mess	made has
		‘I don’t know who made such a mess here.’
	a’.	de	minister,	dy	*dat/*Ø/√‘t	in protte	tasein	hie.
		the	minister	rel	that	much	promised	had
		‘the minister that promised a lot.’
	b.	dat is reduced to ‘t if SpecCP is occupied by an overt element.

The deletion of part of a complementizer in C° when SpecCP is occupied is therefore a phenomenon that is independently attested and not a quirk of the analysis in this paper.

Partial COMP deletion is most likely an instance of a much broader set of phenomena, namely, so-called external sandhi (cf. Kaisse, 1985; Andersen, 1986; Kaisse & Zwicky, 1987). By external sandhi, one refers to phonological operations that exclusively occur at word or phrase boundaries. One example is found in Kimatuumbi, a Bantu language spoken in Tanzania. In Kimatuumbi, a long vowel of the head of a phrase is shortened when overt material follows the head inside the phrase. The data in (32) illustrate this.

(32)	a.	[NP kikóloombé]
		‘cleaning shell’
	b.	[NP kikólombe chaángu]
		‘my cleaning shell’
	c.	[NP kikólombé [AP kikeéle/*kikelé] chaángu]
		‘my red cleaning shell’
		(Kaisse, 1985, p. 178)

In (32a), the NP contains the noun kikóloombé and it appears in its unshortened form because nothing follows it. When the NP also contains the possessive pronoun chaángu, the long vowel shortens (oo → o), turning kikóloombé into kikólombé, as in (32b). (32c) shows that the rule is sensitive to syntactic information and not mere surface adjacency. The vowel of the noun kikóloombé shortens because the NP has overt material following the head noun. The long vowel of the adjective kikeéle, however, does not shorten because it is not the head of the NP, and the AP it heads does not contain any further overt material. Odden (1996, pp. 218–233) discusses this rule at greater length and also shows that adjectives that follow a noun but are not part of the same NP do not trigger shortening, strengthening the relevance of syntactic information for the application of this rule.

This aspect—the sensitivity to syntactic information—is similar to k-deletion in Native Lower Sorbian. It is not the case that any kak is reduced to ak when preceded by silence; kak is only reduced to ak when preceded by silence that corresponds to an empty operator.

Furthermore, vowel shortening and k-deletion are similar in that something has to be adjacent phrase-internally to the head that is affected by a phonological operation: in Kimatuumbi, something has to follow, in Native Lower Sorbian, something has to precede. Kaisse (1985) developed a model where this is not accidental. More specifically, the trigger and target must be in a c-command relation such that one c-commands the other. In Kimatuumbi, the target (the head of a phrase containing a long vowel) must c-command the trigger (the head’s complement); in Native Lower Sorbian, the trigger (the empty operator) must c-command the target (kak in C°).

Finally, the interaction required for this analysis to work is compatible with recent approaches to the interaction between syntax and phonology, namely, cyclic spell-out. This refers to the idea that just as syntax builds up structure cyclically via phases, transfer to PF also applies cyclically. Since the moved empty operator and the complementizer it affects are part of the same CP-phase, the analysis developed does not depend on novel ideas regarding the syntax–phonology connection.

6. Fronted Resumptive Pronouns

There are data both reported in Fasske (1995, pp. 170–171), cf. (33), and found in the DoBeS corpus, cf. (34), that seem to pose a problem for the analysis in (29), namely, relative clauses where the resumptive pronoun is fronted.

(33)

a.

Źo

jo

ten

luź,

tomu

ak

ja

som

cora

te pjenjeze

pžyzycyła.

where

is

the

man

him

rel

I

am

yesterday

the money

given

‘Where is the man who I yesterday gave the money to?’

b.

To

jo

ta

droga,

na

tej

ak

smy

pjerjei

šli.

that

is

the

road

on

it

rel

we.are

earlier

walked

‘This is the road on which we used to walk.’

(34)

a.

Mój

stary

nan,

wot

togo

ak

něnt

som

pówědał.

my

old

father

of

him

rel

now

am

told

‘my grandfather who I told you about.’

b.

Smy

měli

jadnu

flašku,

z teju

ak

smy

se

pón

chwalili

pó jsy.

we.are

had

a

bottle

with it

rel

we.are

refl

then

walked

in village

‘We had a bottle with which we were walking through the village.’

These data obviously clash with the analysis in (29) because the fronted resumptive pronoun occupying SpecCP is not an empty operator and, therefore, should not trigger k-deletion, contrary to fact.

This problem vanishes once the context for relative clauses with fronted resumptive pronouns is taken into consideration. According to my informant, such relative clauses are very natural but always involve some sort of emphasis on the fronted resumptive pronoun. In (35), three such contexts are given that, according to my informant, favor the use of a fronted resumptive pronoun.

(35)	a.	Ja	znam	jadnu	žeńsku,	wót teje	ak	cu	śi	něco	wulicyś.
		I	know	a	woman	of her	rel	I.want	you	something	tell
		‘I know a woman that I want to tell you something about.’
	b.	A:	Madlena		jo žeńska,	ak njocoš		z neju		rejowaś.
			Madlena		is woman	rel not.you.want		with her		dance
			‘Madlena is a woman that you don’t want to dance with.’
		B:	Ně, ně,		Madlena	jo žeńska		z teju		ak cu	rejowaś.
			no no		Madlena	is woman		with her		rel I.want	dance
			‘No no, Madlena is a woman that I do want to dance with.’
	c.	A:	Z kakeju		žeńsku	mógu hyś		na reju?
			with which		woman	I.can go		on dance
			‘With which woman can I go dancing?’
		B:	Madlena		jo žeńska,	z teju  ak		móžoš   hyś		na reju!
			Madlena		is woman	with her rel		you.can go		on dance
			‘Madlena is a woman with whom you can go dancing!’

What these data show is that there is always some additional pragmatic effect that goes along with fronting the resumptive, either frame-setting (35a), contrast (35b), or new information (35c). In order to express the difference between relative clauses with fronting and relative clauses without fronting, I assume that the fronted resumptive pronoun is moved separately from the empty operator to the specifier of a functional projection in the extended left periphery that encodes these pragmatic functions (cf. Rizzi, 1997 for an influential approach). This is shown in (36) for the relative clause in (35c).

(36)

Here, I remain agnostic about whether there is a single FP comprising all these pragmatic functions or a unique FP for each function since this is ultimately of no concern. What is relevant is that the structure in (36) does not clash with the analysis in (29): the fronted resumptive pronoun occupies SpecFP and hence cannot possibly block k-deletion.

Evidence that relative clauses with fronted resumptive pronouns also feature movement of an empty operator comes from the observation that this movement must not cross the clause boundary, cf. (37).

(37)

*

Ta

žeńska,

tu

ak

se

myslim,

až

som

t

wiźeł,

bydli

Barlinju.

the

woman

her

rel

refl

I.think

that

am

seen

lives

Berlin.loc

‘The woman that I think I have seen lives in Berlin.’

If only movement of the contrastive resumptive pronoun took place, it should behave like (22b), which shows that a contrastive DP can be moved out of an embedded clause. Instead, (37) is similar to relative clauses without the additional fronting of a resumptive, indicating that also relative clauses with fronting feature movement of an empty operator.

7. Native Lower Sorbian vs. Czech and German

Before concluding, I want to discuss the differences between the sort of relative clause described in this paper and two types of relative clauses found in Czech and German that superficially appear similar. More specifically, they both involve NP-internal clauses introduced by the wh-word how. Since their respective analyses differ from the one for ak-relatives in Native Lower Sorbian suggested here, the question arises whether a separate treatment of ak-relatives in Native Lower Sorbian is justified. I argue that this separate treatment is justified because, upon closer inspection, both types of NP-internal clauses in Czech and German turn out to be very different from ak-relatives. I start with Czech and then turn to German.

As reported in Šimík and Sláma (2023), Czech too has relative clauses introduced by how, namely, jak. This is shown in (38).

(38)	A	vochutnalas	tu	buchtu,	jak	(ji)	dělali	ty	maldý?
	and	you.tasted	the	cake	how	it	made	the	young ones
	‘And did you taste the cake that the young ones made?’

jak-relatives in Czech are, in many respects, similar to ak-relatives in Native Lower Sorbian. For example, jak-relatives are island-sensitive and feature optional resumptive pronouns (Šimík & Sláma, 2023, pp. 243, 239).

However, there is an important difference between jak-relatives in Czech and ak-relatives in Native Lower Sorbian. Šimík and Sláma (2023, p. 240) observe that jak-relatives have an evidential implication, that is, the use of a jak-relative implies that the addressee has evidence for the truth of the relative clause and the clause containing it. For this reason, they dub them evidential relatives. One consequence of this implicature is that only non-quantificational head nouns are compatible with jak-relatives (Šimík & Sláma 2023, pp. 264–266), as shown in (39).

(39)	a. *	Každej	teenager,	jak	dneska	ráno	vstane	si,	pustí	Evropu 2.
		every	teenager	how	today	morning	gets.up	refl	turns.on	Europe 2
		‘Every teenager that gets up today in the morning turns on Europe 2.’
	b. *	Nejdelší	šichtu,	jak	můžeš	mít tak,	to je	do	vod 9	do 11.
		longest	shift	how	you.can	have so	that is	of	from 9	to 11
		‘The longest shift you can have is from 9 to 11.’

Crucially, ak-relatives in Native Lower Sorbian do not show this restriction at all. They are fine with all sorts of quantificational head nouns, including negatives, superlatives, all sorts of quantifiers, and even non-restrictive relative clauses, as shown in (40).

(40)	a.	Wacen	njej	how	nicht,   ak	serbske	pówědał.
		else	not.is	here	no one rel	Sorbian	spoken
		‘Here is no one else that would speak Sorbian.’
	b.	Tam	su	te	rědnjejše	štuki,	ak  ja	som	namakała.
		there	are	the	most beautiful	pieces	rel I	am	found
		‘There are the most beautiful pieces that I found.’
	c.	Kuždy	gólc,	ak	ma  sam	doma	konja.
		every	boy	rel	has  self	at home	horse
		‘every boy that himself owns a horse.’
	d.	Ta	slědna	šotša	mojog nana,	ak jo	doma	krydnuła.
		the	last	sister	of my father	rel is	house	got
		‘the last sister of my father, who inherited the house.’

So, despite surface similarities, the two types of relative clauses differ too much from each other, precluding a uniform analysis for them on empirical grounds.

The same conclusion must be drawn with respect to a type of relative clause found in German as described by Pankau (2023) and illustrated in (41).

(41)

Maria

hat

einen

Pullover

gekauft,

wie

sie

ihn

bei

Karstadt

gesehen

hat.

Maria

has

a

sweater

bought

how

she

it

at

Karstadt

seen

has

‘Maria bought a sweater as she saw at Karstadt.’

Such clauses are introduced by wie ‘how’ and, given their modification function, could be taken as relative clauses. However, as described in detail by Pankau (2023), such clauses are better analyzed as DP-internal equative subcomparatives. For this reason, Pankau (2023) calls them comparative relatives. In a nutshell, comparative relatives do not restrict the nominal itself but only specify the kind of nominal. Consequently, the sentence in (36) does not entail that Maria bought a sweater from Karstadt (a German department store chain) but only that she bought one similar to one she saw there, as already indicated by the translation. It is in this respect that comparative relatives differ from ak-relatives in Native Lower Sorbian: the relative clause in (42) does entail that Maria bought a sweater from Karstadt.

(42)	Marija	jo	jaden	pulower	kupiła,	ak	wóna	jo	jen	wiziła	Karstaśe.
	Marija	is	a	sweater	bought	how	she	is	it	seen	Karstadt.loc
	‘Marija bought a sweater that she saw at Karstadt.’

Also, here, surface similarities are again deceptive.

8. Summary and Conclusions

In this paper, I argued that a seemingly strange property of Native Lower Sorbian, namely, having a particle that is only used as a relativizer, is only apparent. The offending relativizer ak is a variant of the complementizer kak, whose initial k is deleted due to the movement of an empty operator, as argued in Section 4 and Section 5. There, I also rejected the alternative that ak is a variant of the wh-adverb kak based on agreement patterns that differentiate the wh-adverb kak from the complementizer kak. In Section 6, I demonstrated that the existence of fronted resumptive pronouns, which are expected to bleed the deletion of k, is compatible with the analysis. Furthermore, I showed that ak-relatives differ from the superficially similar constructions in Czech and German in Section 7.

There are two conclusions to be drawn from this paper. First, on the empirical side, Native Lower Sorbian is compatible with what is already known about the syntax of relative clauses but also adds a novel aspect. Although the particle ak has other uses and is, therefore, not designated to function as a relativizer, the use of the wh-word how as a relativizer is so far unattested. This broadens the set of elements used in this function and opens up the question of what unites how and where that makes them so prone to be used as relativizers (note also the more frequent homophony of how and where, cf. Vangsnes, 2008). Second, on the more theoretical side, the analysis argued for here suggests both that the interaction between syntax and phonology is more interconnected than usually assumed (cf. Sande, 2024) and that relative pronouns can have additional pragmatic functions, similar to other A’-moved elements, suggesting an extended left periphery also in relative clauses.