What the giant tells us about agreeing post-verbal subjects in Xhosa

Constructions with the subject following the verb are a widely studied topic in Bantu linguistics. One such construction, in which the subject is dislocated, is considered not as core subject inversion, but generally as an afterthought construction. This study takes a spoken text, in this case a narrative, as its point of departure to examine this kind of construction more carefully in terms of its function and morphosyntactic structure in Xhosa, a Bantu language of South Africa. The paper shows that such agreeing post-verbal subject constructions are used to re-activate semi-active concepts that have been mentioned previously in the narrative. They reintroduce a concept which then becomes the topic of the current sentence and of subsequent phrases, in which the subject is often pronominalized. It is also shown that the expected penultimate lengthening, one of the diagnostics used to differentiate core inversion from constructions with a dislocated subject, is often not present. The function of the construction and other morphosyntactic diagnostics point to the subject being dislocated, however. Furthermore, it is argued, based on a few examples from the narrative and follow-up grammaticality judgements, that there is agreeing inversion in Xhosa, with the subject in the immediately-after-verb position. This inversion construction has not previously been attested in Xhosa and further research is needed in order to corroborate the results. The appendix presents the recorded, transcribed, glossed and translated narrative on which the analysis is based.


Introduction
This paper analyses constructions in Xhosa where the subject follows the verb and there is agreement on the verb with this post-verbal subject1 .Constructions with a post-verbal agreeing subject are not uncommon in Xhosa and other Bantu languages; for an overview, see Marten and van der Wal (2014).There are in fact two kinds of such constructions, superficially similar but different in their syntax and information structure.The first has been analysed as having a subject that is right-dislocated out of the verb phrase, as in an afterthought construction, comparable to English "In the morning they crowed, the roosters".They are traditionally described as being used for predicate focus, with the post-verbal subject as the dislocated topic or anti-topic (Chafe 1976;Lambrecht 1994).For ease of reference, I will call these structures "agreeing post-verbal subject constructions" in this paper.An example is the following, from Otjiherero (Marten 2011:801): (1) V-è-yá, òvà-éndà.SM2-PAST-come 2-visitors 'They came, the visitors.' [Otjiherero] Secondly, many Bantu languages have inversion constructions with the (logical) subject inside the verb phrase in a close bond with the verb.These are "core inversion constructions".When the verb agrees with the post-verbal subject, they are referred to as "agreeing inversion" and are used for subjects that lack topicality; i.e. subject focus and thetic 2 constructions (Marten and van der Wal 2014).For Xhosa, data are lacking with regard to the agreeing inversion construction, and it is reported not to be allowed in Zulu (Marten and van der Wal 2014).The following phrase is an example from the Matengo language.It can be the answer to "What has fallen down?", i.e. subject focus, or it can be used as a thetic sentence whereby all parts of the sentence are new information (Nobuko 2011:756): (2) Gu-hábwiki ńko:ngo.3SM-fall.PERF 3.tree 'A tree has fallen down.' [Matengo] The aim of this paper is to analyse inversion constructions in more natural spoken language in order to critically examine the diagnostics for agreeing inversion vs. agreeing post-verbal subject.The analysis takes as its point of departure a transcribed, glossed and translated narration of a story 3 .Firstly, the data presented in this paper show that there is agreeing inversion in Xhosa.Consider the following example 4 : (3) kwâ:sa za-khál(a) ínkukhu in.the.morning10SM.PST-cry 10.rooster 'In the morning the roosters crowed.' [PSJ150517O] 5 2 Thetic sentences lack a topic-comment structure and are used for the presentation of an event as a whole, also referred to as "all-new" or "sentence focus" (Sasse 1996;Lambrecht 2000;van der Wal 2008).
3 I.e.more natural than elicited examples.A folktale story is of course a specific kind of genre not much used in everyday speech.Nevertheless, it is not a written text, but is freely told from the mind of the narrator. 4Parentheses indicate that the vowel is not heard, often written as zakhal' inkhuku in Xhosa texts.For glossing reasons, it is important to know what the missing vowel is.A colon indicates lengthening of the preceding vowel.
A complete list of glossing abbreviations is included in the appendix.The final vowel (FV) is not always glossed separately in the examples. 5All Xhosa examples are followed by a code consisting of an abbreviation of the place name, plus the date of the recording and a letter indicating how the data have been obtained.In the case of our story, this letter is O=oral tradition, story, legend.The subject follows the verb, and agreement on the verb is with the post-verbal subject íinkúkhu 'rooster' (in noun class 10, but here with a short initial vowel).As further explained in section 3.3, phonological phrasing is evidence of a close link between the verb and the post-verbal noun phrase.As the penultimate vowel of the verb -khâla 'crow' is short, this is an indication that such a close link exists and that this phrase is an example of agreeing inversion 6 .Moreover, the phrase in (3) is used in a thetic construction.
Secondly, the paper aims to determine in more precise terms the function and discourse properties of the agreeing post-verbal subject construction.In terms of information structure, it is shownas expectedthat the agreeing post-verbal subject is often used to re-activate concepts that have recently been mentioned and to introduce them as the topic of the current sentence.In terms of the morphosyntactic structure of the agreeing post-verbal subject, however, we will see that the diagnostics generally used in order to distinguish agreeing postverbal subject from agreeing inversion in Bantu languages are not always available.One of the most important clues as to whether a subject is dislocated is phonological phrasing.A subject that is inside the verb phrase is phrased phonologically with the verb.In Otjiherero, the evidence for such phrasing is tonal, and in Matengo there are several clues to the phrasing, including tone and penultimate lengthening.In Xhosa, phonological phrasing is evidenced by penultimate vowel lengthening.
This paper will show that penultimate lengthening is not always a clear indication of dislocation in agreeing post-verbal subject constructions in Xhosa.This makes it very difficult to tell agreeing post-verbal subject and agreeing inversion apart in tenses where there is no distinction between conjoint and disjoint verb formsanother clue to the distinction between agreeing inversion and agreeing post-verbal subject.Most examples in the narrative are in the remote past, as in the following: (4) wá-bás(a) uNomaha:mle 1SM.PST-kindle uNomahamle 'Nomahamle kindled the fire.' [PSJ150517O] The subject follows the verb, and agreement on the verb is with the post-verbal subject uNomahamle (a name, triggering agreement of noun class 1).There appears to be a close link between the verb and the post-verbal noun phrase, as evidenced by the phonological phrasing: the penultimate vowel of the verb basa 'kindle' is not lengthened.The subject is semi-active in terms of its information structure, as is typical for agreeing post-verbal subject (see section 4.1).
This paper will address the following questions: What are the information-structural contexts in which agreeing post-verbal subject is used?What is the phonological phrasing of the construction?How do we differentiate this construction from agreeing inversion when the phonological phrasing is often not the expected one?What about other diagnostics used to determine whether the subject is dislocated out of the verb phrase?The paper also contributes a transcribed, glossed and translated text in which the agreeing post-verbal subject construction is amply used: "The two girls who escaped from a giant".
the initials of the speaker rather than the place name.Freely-given phrases based on stimuli (e.g.cartoons) have the letter S. I am deeply grateful to all speakers who have contributed to the study. 6The penultimate vowel in the last word is also not long.This is due to the speaker continuing with the sound of the roosters (see the appendix).
The paper is structured as follows: firstly, in section 2, the micro-variation project within which this research was conducted is briefly introduced, together with a discussion of the methodology.Section 3 argues that there is agreeing inversion in Xhosa and gives a background to subject inversion, information structure, and the diagnostics needed to distinguish core inversion constructions from agreeing post-verbal subjects.Section 4 analyses the agreeing post-verbal subject construction, firstly in terms of information structure in 4.1, then examining the diagnostics for subject inversion vs. dislocation in 4.2 (phonological phrasing) and 4.3 (other diagnostics).A summary and conclusions are offered in section 5, and the story is presented in the appendix.

Subject inversion in a micro-variation project
This paper reports results from the project Morphosyntactic variation in the dialects of Xhosa, based on data collected through research in the Eastern Cape in South Africa.Xhosa is a Bantu language of the Nguni group, closely related to other Nguni varieties and especially to Zulu.As with most languages, Xhosa is a cluster of dialects, bordering on the Zulu cluster of dialects and without a clear-cut boundary between the two.The Xhosa standard is based on a few of the dialects (Doke 1954), and most publications on the Xhosa language take this standard as the object of study.Variation in other parts of the Xhosa-speaking area remains relatively understudied (Nyamende 1994).This project does not focus on determining what the Xhosa dialects are, their interrelatedness, and the boundaries between them; rather, it is concerned with any variation encountered, be it geographical or based on e.g.socio-economic factors, gender, age or register.Geographical bordersthose of the Eastern Capehave however been used in order to narrow down the scope of the project.The Eastern Cape is the nucleus of the Xhosaspeaking area.
Subject inversion is one of the morphosyntactic categories studied for this project.It is a complex phenomenon involving phonology (phrasing), morphology (conjoint/disjoint form of the verb), syntax (order of elements), and information structure.Studies carried out on subject inversion in the Bantu language family have shown how the different constructions appear across the family, but also how closely related languages can differ in this domain, making it an interesting candidate for a micro-variation project (Marten and van der Wal 2014).
The micro-variation project is based entirely on fieldwork carried out across the Eastern Cape.The analysis of different morphosyntactic constructions is based on transcribed recordings of spoken Xhosa, which represent how the language functions in actual daily use.Additional elicitation is based on examples from texts, i.e. follow-up questions have been asked in order to test the borders of grammaticality.Traditional elicitation of translated phrases is avoided in this way.For the specific kind of inversion construction presented in this article ̶ agreeing post-verbal subject ̶ a narration of a story to children (see appendix) has been used as the point of departure.Transcription and analysis of further texts is in progress, and the current study therefore focuses on one text with additional elicitation for clarification.The findings presented here will be checked against more data in future analyses.

Agreeing inversion in Xhosa
This section will look into examples from the text that can be labelled "agreeing inversion"; a category of inversion constructions that have so far not been attested in Xhosa.In order to analyse these constructions, the diagnostics used to establish core inversion constructions are considered in section 3.3.Firstly, however, section 3.1 introduces subject inversion more generally in the language, and section 3.2 discusses the effects of subject inversion with respect to the information structure of a sentence.

Inverted subjects
The default word order of Xhosa and other Bantu languages is SV(O).There is an obligatory subject agreement marker (SM) on the verb, agreeing with the (optional) preverbal subject in person and noun class: (5) uSíph(o) u-rhálel-(a) úmphoko:qo Sipho 1SM-like-FV 3.umphokoqo 'Sipho likes umphokoqo' (a dish with maize and milk). [SN150925E] Bantu languages are relatively free with the manipulation of word order for informationstructural reasons, and the subject frequently occurs in the post-verbal position.This has been extensively studied in Bantu linguistics, both for individual languages and comparatively (e.g.Bresnan and Kanerva 1989;Demuth 1990;Demuth and Mmusi 1997;Marten 2006;Zerbian 2006;Buell 2007;Marten 2011;Creissels 2011;Zeller 2013;Marten and van der Wal 2014).
The core inversion constructions are recognised through certain diagnostics, to be presented in section 3.3, and have in common that the post-verbal subject is non-topical.There are different kinds of subject inversion constructions.In the first category, a pre-verbal noun referring to semantic roles such as location, instrument, and patient ̶ for a full overview see Marten and van der Wal (2014) ̶ determines the agreement on the verb, with the logical subject following the verb.The noun referring to location can be marked by locative morphology (noun classes 16-18), referred to as "formal locative inversion", or the noun can lack such locative morphology, in which case it is referred to as "semantic locative inversion", as in the following example from Zulu (Zeller 2013): (6) i-n-gadi i-mil-a u-tshani AUG-9-garden 9.SM-grow-FV AUG-14.grass'In the garden grows grass.' [Zulu] The semantic locative inversion construction, as exemplified in (6), is also accepted in Xhosa, according to my consultants.
In the second category of inversion constructions, there is no pre-verbal argument.Either the agreement on the verb is with the post-verbal subject, referred to as "agreeing inversion" and discussed below, or the agreement is with the default subject marker ku-, followed by the conjoint form of the verb (du Plessis and Visser 1992:130;Carstens and Mletshe 2015).This is a commonly used construction in Xhosa and referred to as "default agreement inversion" (Marten and van der Wal 2014)7 : (7) kú-khála nê:-ntá:ka 17SM-sing and-10.birds(na+iintaka) 'and the birds were singing..

.' [PSJ150517O]
The construction with the non-agreeing subject is used in thetic (all-new or presentational focus) sentences, as in (7), as well as in subject focus (Carstens and Mletshe 2015).The thetic construction can be elicited by asking Kutheni?'What happened?',and is used in out-of-theblue contexts.The subject focus construction is elicited by asking, for example, Kulime bani?'Who cultivated?'.What these two have in common ̶ as with other core inversion constructions ̶ is the non-topicality of the subject.Default agreement inversion can be used with both intransitive and transitive verbs in Xhosa (du Plessis and Visser 1992:130-133).
We now turn to agreeing inversion, in which the verb agrees with the post-verbal subject.This was exemplified in (2) for Matengo.As the remainder of section 3 will show, such constructions also exist in Xhosa.In the text, a new situation is introduced a few times in the following way: (8) lâ:-tshona íla:nga 5SM.PST-set 5.sun 'The sun set.' [PSJ150517O] This is a thetic expression where the whole sentence is in focus (no topic-comment structure).
As agreeing inversion constructions have not previously been established for Xhosa, I expected that a default agreement inversion construction, as in (7), would be used.However, the construction in (8) has a formal resemblance with the agreeing post-verbal subject construction, where the subject noun phrase follows the verb in a dislocated position and the subject agreement on the verb is with this post-verbal subject, as shown for Otjiherero in (1) and for Xhosa in (9).The difference between the two can be determined by a set of diagnostics discussed in section 3.3.Firstly, as these constructions are said to differ in information structure, the following section will examine the way in which this difference manifests itself.

Information structure of VS constructions
Information structure is concerned with how a language packages an idea in different ways, depending on (i) what the speaker assumes the hearer already has in her mind and can relate to, and (ii) the intentions of the speaker with regard to what is more important in the message and what is backgrounded.The idea or message in itself has the same truth value regardless of the different ways that it can be packaged (see for example Lambrecht (1994) and Krifka (2006)).
We have seen that the canonical word order in Xhosa is SV(O), and that the word order can be changed without changing the truth value of the message.The speaker's choice to use a VS construction is a matter of packaging.
Such packaging can represent the speaker's assumption about the mental representation of the referent, i.e. whether the referent is identifiable by the hearer and how recently a representation of this identifiable referent has been activated (Lambrecht 1994), or, to put it another way, what the cognitive status of the referent is in the hearer's mind (Gundel, Hedberg and Zacharski 1993).In Chafe's (1987) terms, a concept can be active at the moment of speaking.Such concepts are often pronominalised or not verbalised at all.An example is when a speaker says "..this was uh... a biology course", where "this" refers to an undergraduate course already activated previously in the discourse (Chafe 1987:27).A concept is semi-active or accessible if it is in background awareness because it has been mentioned before or because it is expected based on its relatedness to other things that have been mentioned; i.e. it is part of a schema of concepts related to a certain mentioned concept (e.g. the concept 'students' is expected in relation to the undergraduate course mentioned above).Semi-active concepts can also be present in the text-external world (Lambrecht 1994).Inactive concepts are newly brought into the conversation, as with "biology course" in the example above from Chafe (1987).
These ideas of identifiability, activation, and cognitive status are called "pragmatic states" (Lambrecht 1994) and are interrelated with the ideas of topic and focus, which concern the pragmatic roles that a referentwhich has certain identifiability and activation propertiescan play in a proposition (Lambrecht 1994).Topic and focus are relational categories, where the topic of a sentence is the thing that the proposition is about, while focus indicates that the referent is unpredictable in the situation, and draws specific attention to it.
For Bantu languages, as summarised in Marten and van der Wal (2014), the information structure of the sentence determines the use of the different structures available with a postverbal subject, as post-verbal subjects are used to express (non-)topicalisation of subjects.In core inversion constructions, the subject is not the topic in terms of 'what the sentence is about'.Rather, the subject is either in focus, or it is part of a thetic construction that introduces new information.A pre-verbal expression typically functions as the topic.Agreeing inversion is one of these core inversion constructions.The following example is from Makhuwa (Marten and van der Wal 2014): (10) ni-hoó-wá n-láikha SM5-PRF-DJ-come 5-angel 'There came an angel.' [Makhuwa] Agreeing inversion thus differs from agreeing post-verbal subject, where the subject comes after the verb but is dislocated, as we shall see in the following section.In such constructions, the subject is discourse-old (or active), and the verbal agreement marker "can be analysed as agreeing with a contextually given discourse topic rather than with the right-dislocated DP directly" (Marten and van der Wal 2014:36).It is this function of the agreeing post-verbal subject that we will examine further in the text.

Diagnostics for VS constructions
Our understanding, then, of the function of a certain VS construction depends on the information structure of the sentence and the discourse.Moreover, there is morphosyntactic and phonological evidence for the existence of the specific inversion construction concerned.
Marten and van der Wal (2014) summarise the diagnostics that have been set out in previous studies.Such diagnostics help us to determine whether the construction in question concerns core inversion, where the post-verbal subject is phrased with the verb, or whether the post-verbal subject is right-dislocated out of the verb phrase.These diagnostics will be presented here.
Firstly, in a core inversion construction, the inverted subject cannot be left out (Marten and van der Wal 2014).Secondly, object marking is not possible (van der Spuy 1993;Buell 2008;Zeller 2015).Thirdly, there is a close bond between the verb and the post-verbal subject.Evidence for this bond comes from phonological phrasing.In Xhosa, penultimate lengthening indicates the end of a phonological phrase (Jokweni 1995;Downing 2003;Zerbian 2004).An utterance such as in the following example makes up a single phonological phrase, and can be interpreted as in broad focus, meaning that the whole utterance is focused.It is the answer to the question "What are they doing?"Example from Jokweni (1995:31): The phrase in ( 11) can also imply a narrow focus on the object with no change in phonological phrasing.In the case of predicate focus, phrasing is split up.In the following example, the verb is in focus and the object noun is right-dislocated (Jokweni 1995): book 'They open it, as for the book.' [Xhosa] We see this close phonological bond in default agreement inversion, which is a core inversion construction.For example, in (7), the verb and the subject are in one phonological phrase and the penultimate vowel of the verb is not lengthened8 .This was checked with another speaker, who also pronounces the default agreement inversion construction without lengthening on the verb, as expected: (13) ku-khála né-ntá:ka 17SM-sing and-10.birds'and the birds were singing..

.' [NF160404E]
Further evidence of the close bond between the verb and its inverted subject is provided by the conjoint/disjoint distinction.The Xhosa language exhibits a distinction between the conjoint (short) and disjoint (long) forms of the present and perfect verb forms (van der Spuy 1993; van der Wal and Hyman 2017).The perfect disjoint form with -ile is illustrated in ( 14) and the conjoint perfect with -e is illustrated in ( 15): (14) í-bá-tháth-iˑle ke ya-ba-léqísa e-síbhedlé:(le) 9SM-2OM-take-PRF.DJT then 9SM.PST-2OM-hurry LOC-hospital 'it (the ambulance) took them then and hurried them to the hospital.' [BU151210S_b] (15) kw-íːntsásá yá-namhláːnje sí-bon-é inqwéːlo… LOC-morning 9.AS-today 1pSM-see-PRF.CJ 9.vehicle 'In the morning of today we saw a vehicle…' [BU151210S_d] The same distinction also exists in the present; see (11) and ( 12).That the distinction between the conjoint and disjoint form relates to information structure and focus has repeatedly been shown for the Bantu language family; see Downing and Hyman (2015) and van der Wal and Hyman (2017) for an overview.This includes other languages of the Nguni group (van der Spuy 1993; Buell 2006) and also Xhosa (Jokweni 1995;Downing 2003;Zerbian 2004).Specifically, the position directly following the conjoint form is important for expressing narrow focus.This position is referred to as the "immediately after verb" (IAV) position, following Watters (1979), and all elements occurring there are verb-phrase internal, as has been shown to be the case in several Bantu languages (e.g.van  Therefore, IAV can be said to be the position for non-topical elements, as has been shown for Zulu (Buell 2009).We will come back to these diagnostics in sections 4.2 and 4.3, after analysing the agreeing post-verbal subject in terms of information structure.
The conjoint/disjoint distinction, however, is not relevant in all tenses, and specifically not in the remote past tense, the tense in which most sentences in the narrative are set.We can however try to change the tense of a sentence in elicitation in order to test the distinction between agreeing inversion and agreeing post-verbal subject, as we shall see in section 4.3.

Agreeing inversion in Xhosa
Coming back to agreeing inversion in Xhosa, we see that for (8), repeated here as ( 18), the postverbal subject ilanga 'sun' is not active or semi-active.The situation is new; after walking for a while, the event of the sun having set is introduced in the story.
(18) lâ:-tshona íla:nga 5SM.PST-set 5.sun 'The sun set.' [PSJ150517O] The same applies to (3), repeated as ( 19).After being caught by the giant and put into bags for the whole night, the next day is introduced: (19) kwâ:sa za-khál(a) ínkukhu in.the.morning10SM.PST-cry 10.rooster 'In the morning the roosters crowed.' [PSJ150517O] Regarding the diagnostics for core inversion constructions, we find that there is no penultimate lengthening of the verb, indicating a close bond between verb and subject.In order to check the agreeing inversion further, a speaker was asked in elicitation whether the phrase in ( 13) could be changed to replace the default agreement inversion construction.This is possible, and there is no phrase break after the verb: (20) za-khála né-nta:ka 10SM.PST-sing and-10.birds'and the birds sang.' [NF160404E] These first results point to the existence of agreeing post-verbal subject as well as agreeing inversion in Xhosa.If this is correct, examples like ( 18) and ( 20) represent agreeing inversion.
These have been checked with other speakers and new phrases have been formed through transformational elicitation, with other verbs and subjects.However, agreeing inversion has not shown up spontaneously in elicited examples as the answer to the question "What happened?".This could be due to the question asked: Kwenzeke ntoni?'What happened?' has noun class 17 agreement, and this could trigger the noun class 17 agreement in default agreement inversion9 .Agreeing inversion also does not occur in my data as subject focus.When a subject is questioned, as in Ngubani ogijimayo?'Who is running?', the answer can be given with a default agreement inversion construction, Kugijima uNontle 'Nontle is running ' [BN150927E].It is possible that agreeing inversion only functions as a thetic expression in Xhosa, and maybe also only as a certain type of thetic expression.More investigation is needed in order to determine the kinds of contexts in which agreeing inversion occurs.
As discussed in section 3.3, an indicator of the close bond between verb and subject in core inversion constructions is the use of a conjoint form.The phrases in the story are mainly expressed in the remote past, where there is no distinction between the conjoint and disjoint form.The perfect form is also used in the story, and in such cases it is the conjoint form that is used: The construction occurs several times, but only with the verb ukuthi 'to say'.This is an irregular verb in Xhosa, which makes it difficult to draw any conclusions regarding the kind of inversion construction this represents.Other verbs were tested with consultants.A conjoint present form is accepted in ( 22), with the object following the subject.It is assumed that the subject is in focus in this example and that this is agreeing inversion; however, this should be confirmed by looking for similar examples in natural speech.That is also definitely required to establish the VOS order in ( 23).This phrase is highly unexpected with a conjoint form: The example in (20) can also be changed to the present conjoint form.According to the consultant, the last vowel of the verb is obligatorily dropped in such a construction: (24) zí-khál' ínta:ka 10SM-sing 10.birds 'The birds are singing.' [SQ160422E] *zikhala iintaka

The function and form of the agreeing post-verbal subject
We now turn to the constructions in which the subject is dislocated out of the verb phrase.Firstly, information structure will be considered: how the speaker uses this construction in discourse as a referring expression.Secondly, the diagnostics, as introduced in section 3.3, are applied to determine the structure of the phrase.The construction occurs frequently in the narrative in question, a folktale in the oral tradition.It should be noted that most of the examples are in the past, as expected for this genre of speech.Further analysis of other types of spoken language is needed in order to test the ideas presented here.In this section, all examples are from the story [PSJ150517O], unless otherwise indicated.

4.1
Information structure in "The two girls who escaped from a giant" The story starts with a default agreement inversion construction.As already mentioned in section 3.1, this is a strategy frequently used to introduce new information, where both the subject and the predicate are in focus, referred to as a "thetic sentence" or "sentence focus" (Lambrecht 2000): (25) kwá-kú-ngu-mâ:ma ó-wá-yé-ná-mántombazana áma-bi:ní 17SM.PST-17SM.CT-COP-1a.mother1RC-1SM.PST-1SM.CT-with-6.girls6AC-two 'There was a mother who had two daughters.' The daughters introduced in ( 25) are now the active concept about which the following statement is made.After this, the concept 'mother' is picked up again.In order to do this, the subject umama 'mother' follows the verb: (26) wá:-wá-yálela: ké uma:ma índlela é:-ya kwa-malû:me 1SM.PST-6OM-instruct then 1a.mother 9.road 9.RC-go LOC -1a.uncle 'The mother then explained to them the way to get to uncle.'After she has explained how to get there, an agreeing post-verbal subject construction is used again to refer to the girls.This pattern repeats itself throughout the story, in a manner that I propose is related to the pragmatic states of the referents, as introduced in section 3.2.The concepts have been previously mentioned in discourse and can be said to be semi-active or accessible.To bring them into an active state, either the agreeing post-verbal subject construction is used or some other non-canonical construction (for example 'they went while their mother said goodbye', with a switch to mother as active concept).When a concept has been brought into the active state, a few sentences (normally one to four) refer to the active concept, until it changes to another.
The agreeing post-verbal subject construction is also common when introducing and changing the utterer of direct speech.This is called "quotative inversion": ( 27) wáth(i) úNóntsókólo: 1SM.PST-say Nontsokolo 'Nontsokolo said:' Because of its function of changing a semi-active concept to the active concept in the story, transitions from one event to another are also often marked by means of the agreeing postverbal subject construction.For example, after discussing for a long time what road to take, the construction is used when someone finally makes a move: (28) wâ:-thatha é-ya-sé-kunene uNontsokolo… 1SM.PST-take 9.RC-9AS-LOC-right Nontsokolo 'Nontsokolo took the right one (road).'Inactive concepts are not introduced by this construction.When the giant comes into the picture, it is as the complement of a verb: 'they heard the growling of a giant'.After this introduction, the actions of the giant are often introduced with the agreeing post-verbal subject construction, which brings the concept 'giant' into the picture and turns it into an active concept again.Two examples are the following: (29) l-othúk (a ) ízi:m, lá:-vúka lá-hlála ng-éempu:ndu 5SM.PST-be.startled5.ogre 5SM.PST-wake.up5SM.PST-sit with-10.buttocks'the ogre got startled, woke up and sat on his buttocks.'This construction, therefore, is used to re-activate an accessible concept, which becomes the topic of the sentence.It is interesting to note that due to variation in phonological phrasing, the agreeing post-verbal subject construction appears morphosyntactically similar to agreeing inversion.Recall from section 3.2 that agreeing inversion is a core inversion construction used for thetic sentences or for subject focus.Apart from this function, it is characterised by diagnostics relating to the close bond between the verb and the subject.In order to establish the properties of agreeing post-verbal subject and the differences between this and agreeing inversion, it is necessary to go into more detail regarding the diagnostics that are used to establish if a subject is dislocated or if there is a close bond between the verb and the subject.

Agreeing post-verbal subject and phonological phrasing
As shown in section 3.3, a strong indication of a phonological phrase in Xhosa is penultimate lengthening (Jokweni 1995;Downing 2003;Zerbian 2004).The end of a phonological phrase following the predicate indicates that what appears after that phonological phrase is dislocated.
In the examples in the story, we find utterances where the subject is dislocated, in the sense that it is preceded by a verb with penultimate lengthening.
(31) la-go:dû:k(a) izim 5SM.PST-go.home5.giant 'The giant went home.' (32) lá:-ba:lé:k(a) izim 5SM.PST-run 5.giant 'The giant ran.' In many other cases, however, there is no penultimate lengthening.We see this in (29), and also in the following: (33) bâ:-zo-phúm(a) ábantwana kw-ézi ngxo:wa 2SM-FUT-come.out2.children LOC-10.DEM.PRX 10.bag 'The children will come out through these holes.'This is unexpected.It should be noted that this narrative is told by an amateur storyteller, but a very talented and charismatic one.Therefore, the intonation is sometimes marked with high pitches and other ways of making the story sound even more interesting to the children.Lengthening is sometimes also exaggerated for such purposes, as is the case in (32).In order to test the lengthening of agreeing post-verbal subject phrases further, the same sentences were pronounced by another speaker, as mentioned in section 2. For many of the phrases taken from the text, the speaker accepts a version with vowel liason, as well as a more carefully pronounced version where the vowel remains.The penultimate vowel, however, remains short: (34) lâ-súkum' í:zim lâ-súkuma í:zim 5SM.PST-stand.up5.giant 'the giant stood up.' [NF160404E] As there is no penultimate lengthening on the verb, the verb and the subject can be considered to be in the same phonological phrase.When changed to a present disjoint form, there is a clear phrase break after the verb.The present conjoint form is disallowed in this specific case: (35) lí-ya-súkû:ma í:zim 5SM-DJT-stand.up5.giant 'the giant stands up.' [NF160404E]

* lisukum(a) izim
It could be that this kind of phonological phrasing is specific to the past.In summary, the penultimate vowel of the verb followed by the subject is sometimes lengthened, and sometimes is not.It appears that the relationship between phonological phrasing and syntactic phrasing is an imperfect one.This is also shown in a recent paper on phrasing in Zulu, in which penultimate lengthening was systematically measured in verbs followed by an in situ direct object or a dislocated object in tense forms without the conjoint/disjoint alternation (remote past and future) (Zeller, Zerbian and Cook 2017).When comparing the length of penultimate vowels across the whole sample of tokens per speaker, penultimate lengthening is a relevant and significant indicator of phonological phrasing.However, not all penultimate vowels are lengthened in the relevant contexts.Medial and final vowels differ more clearly in their length in the morphologically marked present forms, but show some overlap in the remote past.This means that there is some variation in the length of penultimate lengthened vowels, and this variation is greater for some speakers than for others.
Although these findings concern Zulu and are not related to subject inversion constructions, they give an indication of why penultimate lengthening is not always crystal clear.In measuring vowels in an utterance, it is fairly easy to distinguish between a short and a lengthened vowel.But in many instances, a vowel may be slightly longer than a short one would be, but is not a clearly lengthened vowel.As vowel lengthening is important for our understanding of phonological phrasing and morphosyntactic structure, a detailed analysis of the phonetic properties of vowel lengthening in Xhosa would be much needed.We now turn to other diagnostics of dislocation, as introduced in section 3.3.

Other diagnostics of dislocation
The first diagnostic concerns the omission of the subject.This is allowed in terms of the morphosyntactic structure of the agreeing post-verbal subject construction, according to my consultant (although it might make the story less clear).As the omission of the subject in the following example is based on (34), we can conclude that it is the post-verbal rather than a preverbal subject that is left out: (36) lá-súkú:ma 5SM.PST-stand.up'(the giant) stood up.' Secondly, the occurrence of an object marker is evidence of dislocation.There is no example of an agreeing post-verbal subject with an object marker in the text.However, the language consultant accepts adding an overt post-verbal subject to a construction with an object marker from the text.In the original text, izim 'giant' does not follow the verb, but it has been added in the example: (37) lá-m-fák(a) ízim 5SM.PST-1OM-put 5.giant 'The giant put her (inside).' [NF160404E] The consultant also accepts this construction without the deletion of the last vowel of the verb.
In her pronunciation, the penultimate vowel is now lengthened: (38) lá-m-fá:ka í:zim 5SM.PST-1OM-put 5.giant 'The giant put her (inside).' [NF160404E] Also, an adverb can intervene between the verb and the subject.In the case of core subject inversion, this is not possible, as the subject has to appear in IAV: (39) lá-m-fák(a) é-ngxow-é:ni om-nye i:zim 5SM.PST-1OM-put LOC-9.bag-LOC1AC-one 5.giant 'The giant put one of them in the bag.' [NF160404E] All this clearly points to a dislocated subject, and, in such cases at least, the VS constructions in the story are not agreeing inversion but involve a dislocated subject, however confusing the phonological phrasing seems.Another diagnostic related to the close bond between verb and subject is the conjoint/disjoint distinction.In elicitation, agreeing post-verbal subject constructions with a right-dislocated subject are easily elicited by asking a question about the action (see also (35)): (40) Q: benza ntoni?'What are they doing?' (the children) A: bá-yá-dla:l-a abántwa:na 2SM-PRS.DJT-play-FV 2.children 'they are playing, the children.' [NF151210E] As we can see from the use of the disjoint form and the phonological phrasing, the subject abantwana 'children' is outside of the verb phrase and is right-dislocated.The subject is therefore in a non-focused position, as has been pointed out for Zulu (Buell 2008).The subject is the topic and there is predicate focus.

Concluding remarks
This paper has examined constructions in which the subject follows the verb in a narrative in Xhosa.These inverted constructions are of three kinds in the narrative.One of these is default agreeing inversion, whereby a default agreement marker ku-is used with the conjoint form of the verb and is followed by the subject immediately after the verb in the same verb phrase.Such constructions are commonly used to introduce a new situation and can also be found in the narrative.Surprisingly, constructions in which the verb agrees with the post-verbal subject are also found in this narrative with the same kind of function, i.e. as thetic sentences.These constructions were tested in elicitation and this paper concludes that Xhosa makes use of agreeing inversion, although this has not previously been attested for this language, nor for other Nguni languages.In agreeing inversion, the post-verbal subject is phonologically phrased with the verb and is not dislocated.However, these conclusions are drawn based on limited data and need to be tested further in future research.
In the most commonly used inversion construction in the narrative, the subject is ̶ in contrast to the agreeing inversion construction ̶ dislocated out of the verb phrase, as evidenced by penultimate lengthening on the verb.It is concluded that this construction is used when the concept denoted by the subject is semi-active; i.e. it has been mentioned previously in the text or it can be inferred from the context.The agreeing post-verbal subject construction is therefore commonly used when something new happens in the story and it needs to be made clear who the agent of the action is.It appears frequently in turn-taking; i.e. to change the person who is saying something in the narrative.However, the diagnostics used in Bantu language studies to determine the dislocation of the subject are not always in place, specifically with regard to penultimate lengthening.In several of the sentences in the narrative, there is no lengthening of the penultimate vowel of the verb preceding the subject.These sentences make use of the remote past form of the verb.This result fits in nicely with recent findings with regard to penultimate lengthening in closely-related Zulu, for which it has been concluded that such lengthening is not a clear sign of phonological phrasing in the tenses that do not have a morphological distinction between the conjoint and the disjoint forms, such as the remote past.

U-th-e
uNontsokolo "umama e-beth-e si-hamb-e nge-ya-se-kunene".1SM-say-PRF Nontsokolo 1a.mother 1PRT-instruct-PRF 1pSM-walk-SBJV INSTR-9AS-LOC-right Yho, uVathiswa a-ka-zange a-yi-vume no-ku-yi-vuma yho Vathiswa NEG-1NEG-never NEG-9NEG-agree CONN-INF-9OM-agree      Narrator: …there was a woman who had two daughters.One day, these daughters asked to visit their uncle.Their mother instructed them about how to get to their uncle: "When you reach the crossroad, you will see two roads diverging; one goes to the right and the other one goes to the left."The girls prepared themselves.Their mother cooked snacks for them and slaughtered a chicken so that, when they got hungry on their way, they could eat.They prepared themselves with joy because this would be their first time to visit their uncle.They woke up early in the morning; they prepared, washed themselves, took their small bags and put their clothes inside.One of them was older and the other one was younger.They left while their mother was saying goodbye to them: "Goodbye my children, travel safely.""Remember, especially you, Nontsokolo, take the left road.Don't take the right road, you will get lost."They both agreed: "Yes mama!"They travelled in happiness.They were jumping up and down and tickling each other: "Yho!We are going to uncle!" "Wow, we are going to see uncle!"The sun set.When they arrived at those two diverging roadsone disappearing to the left and one disappearing to the rightthey stopped.The young one said, "Nontsokolo, what way did mother instruct us to take, on which road should we walk?" Nontsokolo said: "Mother instructed us to take the one to the right."Yho, Vathiswa totally disagreed with that.She said, "No Nontsokolo, mother instructed us to take the road to the left, we should not take the one to the right."They argued until it became dusk.
Nontsokolo took the one to the right because she was the oldest.Vathiswa, the small one, had no alternative but to follow Nontsokolo.They walked the road to the right, the one they were told not to take because they would get lost.They walked, walked and walked until they got into the forest.
(Narrator is talking to the audience) "Tell me my children, when kids enter into the forest after sunset without knowing (the way), are they going to get to their uncle?"Yho, the trees were shaking because the animals of the forest were trying to sleep at that time.Birds were singing: tyululukuthu,tyululukuthu. Yho! Monkeys: tyo! tyo!, tyo! tyo! Yho!!! Vathiswa was afraid as she was the youngest one.Nontsokolo said, "Come Vathiswa and let me hold you by your hand, because I can see that you are afraid."They walked, walked, walked and walked.Vathiswa said, "No, Nontsokolo, why is it like we don't get out of this forest, we have just been walking around, yho!" Nontsokolo said, "No, be patient daughter of my mother, Vathiswa, we are close to an exit."It happened that at night as they were walking, they saw lights glimmering in a distance.Vathiswa, the youngest, said, "It seems like there's a light over there, is it where uncle stays?"Nontsokolo said "Yes, we are almost there."They walked, walked and walked until they came out of the forest.
It happened that when they came close to the house with the light, they heard the growling of a giant.The giant was exhausted and sleeping close to the door and its bag was lying next to it because it was coming from hunting.Yhiyho-o-o!!! What is the giant saying?"Hee yhee phephephe!!!" Vathiswa was very scared: "Yhooo!!! Nontsokolo!What are we going to do, we will be eaten by the giant!"It happened that when she said that, the giant was startled, woke up and sat down, saying "Fakatha, fakatha, why is there a smell of a human here?Fakatha, fakatha, why is there a smell of a human here?"Yho-o!Nontsokolo, the older one, was also scared now, she said: "Yho!Mother said we should not take this route."Yho!The giant stood up, jumped and caught both of them.The giant put them into the two bags, Vathiswa in one bag and Nontsokolo in another.The giant opened holes in the bags so that they could breathe.It tied the bags and guarded them.The giant said, "I will eat you on a rainy day."Yho, they were scared inside the bags, they were really scared inside the bags.
The giant slept at night holding the bags; it did not want them to escape.Yho, in the morning the roosters crowed Kurukugu!!! Yho!In the morning, the giant woke up and went to fetch water where it was available.This was when the giant passed a house where a cow was being slaughtered.It was called, "Come this side man27 , we will give you pieces of meat." The giant passed the house carrying these bags, it walked, gqu, gqu, gqu, gqu.When it got there, it put the bags down.It was told "Take this here so that we can pour you something to drink."The giant drunk hi-i hi-i hi-i, and that is when they got suspicious that there might be people in those bags.They sent the giant to get them water.One of the men said: "I will watch your bags and no one will touch them.Go and fetch us water."But they gave him a leaking bucket.
The giant went to fetch water, zigqu, zigqu, and during this time they opened the bags, took out the children and locked them in another house.They put reptiles, snakes, and frogs in the bags.The giant was confused.Fetching water with a bucket that is leaking he-e-e-e!!!It tried to fetch water again and again but the water was leaking out.
The giant gave up and went back making loud noises on the way: "They gave me a bucket that leaks, they gave me a bucket that leaks."It arrived at the homestead and threw down the bucket because it realised that it was just leaking.The giant said "Give me my bags so that I can walk home!There's nothing you have sent me to fetch water for!"They gave the bags to the giant, it went home, the giant was happy, knowing that it was going to slaughter.
When the giant got home, it shouted: "Nomahamle-e-e-e-e!!!" It had a child who was a girl."Ma-a-ma!""Make a very big fire!!" Nomahamle kindled a fire, Nomahamle kindled a fire, the giant arrived and closed the door.Look, the fire was burning.The giant was going to cook the children.It said, "I don't want any hole that these children can use to get out of these bags and run."The giant locked the door and opened both bags.A-a-yhu!! Nasty!The reptiles, bees, and snakes were jumping up and down, and the giant was crying: "Yho-o-o!!! Yho-o-o!!!"There was a big container inside this house.This big container was filled with water.The giant rushed to the big container to put its head inside, "Aaaah!"Zum! Yho my children, the giant was really crying, but it drowned inside the container and died.There in the backyard, the children were taken home.
Here is our story.The end.