Aspects of the Arabic middle – an exploration of voice, aspect and valency

Running head: ASPECTS OF THE ARABIC MIDDLE

Aspects of the Arabic middle

- an exploration of voice, aspect and valency

Hanna Danbolt Ajer

Department of Theoretical and Applied Linguistics

Trinity Hall, University of Cambridge

June 2015

This dissertation is submitted for the degree of Master of Philosophy

ASPECTS OF THE ARABIC MIDDLE

of 95

This dissertation is the result of my own work and includes nothing which is the outcome of

work done in collaboration except where explicitly indicated in the text.

This dissertation contains 19,946 words.

With thanks to my supervisor, Professor Ian Roberts, for his support and advice.


of 95

Table of contents

1 Introduction 5

2 Theoretical background 7

2.1 The Arabic verb 7

2.2 The middle voice 11

2.2.1 Clarifications of terminology 11

2.2.2 The approach of Kemmer (1993) 12

2.2.3 Affectedness 14

2.2.4 The analysis of the Arabic middle in Ajer (2014) 15

2.2.5 An extended notion of subject-affectedness in Arabic 20

2.3 Lexical aspect 28

2.3.1 Lexical aspect in Arabic 34

2.4 Valency 35

2.4.1 Valency in Arabic 40

2.5 A hierarchy of affectedness 42

3 Methodology 47

4 Analysis 52

4.1 Subject-affectedness in the middle Forms 52

4.2 A feature approach to the Arabic middle 53

4.3 The U-pattern 56


of 95

4.4 Form VII 60

4.5 The I-pattern 63

4.6 Form VIII 67

4.7 Form V 72

5 Conclusion 76

6 Bibliography 80

7 Appendix A – Transliteration chart 82

8 Appendix B – Abbreviations used in example glosses 83

9 Appendix C – Key to data sample 84

10 Appendix D – Data sample 88

11 Appendix E – Questionnaire 89

11.1 Procedure 89

11.2 Consent form 90

11.3 Background information form 92

11.4 Questionnaire 93


of 95

1 Introduction

This paper will provide an analysis of several derivational verb Forms in Modern Standard

Arabic, hereafter ‘Arabic’. The Forms in question are the ones I will term the U- and I-

patterns, as well as Forms V, VII and VIII.

All these Forms have previously been identified as encoding the middle voice. I will argue

that the middle Forms in Arabic are unified by always allowing affected subjects, and that the

distinctions between them mainly relate to differences in type of affectedness. In order to

account for the fact that these different types of arguments pattern as affected, the notion of

affectedness will need to be extended. It is hoped that the more fine-grained distinctions of

affectedness proposed here might inform our view of the middle voice.

As we will see, these affectedness distinctions are used to capture the different middle Forms

in Arabic. They might therefore also bring us closer to understanding the semantic and

syntactic contributions of Arabic verbal morphology, a matter which has long puzzled

Arabists.

Furthermore, I will seek to show how each Form’s valency and aspectual properties largely

follow from the type of affectedness it requires. It seems as though the middle in Arabic for

the most part does not impose restrictions on valency and aspect, but that these properties

rather arise as by-products of the middle readings favoured. Thus, the study also aims to

contribute to the discussion of the interaction between the three phenomena of aspect, voice

and valency.

The fact that there seems to be no requirement on valency is particularly notable, as middle


of 95

formation is often analysed as a valency-reducing process. My findings show that this is not

always the case, and that this property therefore cannot be crucial to the middle.

Furthermore, my analysis of the middle Forms supports the idea that there is a distinction

between affected and non-affected objects, as the former seem to be more strongly transitive

than the latter. This indicates a need for developing a more complex notion of valency, which

takes into account both the number and type of arguments for which the verb s-selects.

In order to follow my analysis of the middle Forms, some theoretical background is needed.

This will be introduced in chapter 2. The chapter will open with an overview of Arabic verbal

morphology in 2.1, before turning to a discussion of the middle voice in 2.2. Thereafter, we

will look at the concepts of aspect and valency in sections 2.3 and 2.4 respectively. The idea

of affectedness and the need for extending it will appear throughout the chapter, before being

discussed more thoroughly on its own in 2.5.

Chapter 3 outlines the methodology used in the study at hand, before chapter 4 moves into

the analysis itself. Section 4.1 looks at how the property of subject-affectedness relates to the

Arabic middle. I then explore how a feature approach might be useful to the analysis of the

verb Forms in 4.2. After that, I go through each of the verb Forms in sections 4.3 to 4.7,

seeking to identify their key properties and showing which effects these properties have on

their syntactic and semantic behaviour.

Finally, chapter 5 concludes the paper.


of 95

2 Theoretical background

2.1 The Arabic verb

Arabic has a root-based morphology, where verb stems are formed by combining the root

consonants with set derivational patterns of vowels and consonants. Most roots contain three

consonants, which in reference forms are represented by f, ʕ and l respectively (Wright

1962[1859-1862], vol. 1:pt. 2,§35). Thus, a root such as h r s can be combined with the

pattern faʕala, creating the verb harasa (“guard s.th.”).

It is important to note that most verbs have many different meanings, of which I can normally

only quote a few when giving examples. Unless otherwise specified, all example verbs are

from my sample, which can be consulted in Appendix D. The key to the sample is provided

in Appendix C, whereas a description of how it was constructed can be found in chapter 3.

The transliteration system adopted is laid out in Appendix A. It is based on the one used in

Hans Wehr’s Dictionary of Modern Written Arabic (1979[1961]), henceforth ‘Wehr’.

Verbs created from the same pattern are said to belong to the same ‘Form’. The intuition is

that in addition to sharing the same derivational morphology, verbs of the same Form share

some component of meaning (Badawi, Carter, and Gully 2004:60). However, it is unclear

exactly what each Form contributes, and most grammars simply list some of the meanings

commonly associated with them. Furthermore, linguistic studies of the Forms are rare, though

the recent studies of Ajer (2014), Danks (2011) and Fehri (2012:chap. 2) all investigate some

of them. Their analyses of the Forms in question relate to the phenomena of the middle voice,

lexical aspect and verbal plurality respectively. Verbal plurality will not be discussed in the

present study, but the relevant literature on middle voice and lexical aspect will be reviewed

in 2.2 and 2.3. First, however, an overview of the most common verb Forms might be helpful.


of 95

For three-consonant roots, there are considered to be fifteen different Forms, which Arabists

refer to using Roman numerals (Danks 2011:18). Our overview will include Forms I to X,

which are the ones that are still in common usage (Badawi et al. 2004:62; Danks 2011:19). A

majority of roots combine with several different Forms to create verb stems, though no root

gives rise to verbs in all Forms (Danks 2011:23,29).

Whereas the other Forms only have one pattern each, Form I gives rise to the three patterns

faʕala, faʕila and faʕula. Henceforth, I will term these the ‘A-pattern’, ‘I-pattern’ and ‘U-

pattern’ respectively. There is disagreement about whether there are any semantic and

syntactic differences between these three patterns (Badawi et al. 2004:60; Holes 2004:101;

Wright 1962, vol. 1:pt. 2,§37-38). Some of the supposed differences can be observed in Table

1, which presents some of the meanings most commonly attributed to Forms I-X. It is

important to note that the table gives a highly simplified picture of the meanings each Form

expresses.


of 95

Table 1. Overview of the most important Arabic verb Forms and their meanings (based on

Badawi, Carter, and Gully 2004:59–62; Fehri 2012:32; Holes 2004:100–105; Ryding

2005:455–595; Wright 1962, vol. 1:pt. 2,§36–65).

Form Pattern Meaning Example Source of

example

I: A-pattern faʕala dynamic

reflects root meaning

ʾarasa (“till the land”) my sample

I: I-pattern faʕila temporary (change

of) state

jadila (“be happy”) my sample

I: U-pattern faʕula permanent (change

of) state

raquʕa (“be stupid”) my sample

II faʕʕala causative

intensive

ballaʕa (“make s.o. swallow s.th.”)

mazzaqa (“tear s.th. to pieces”)

my sample

III faʕala conative (attempting

to do s.th.)

qatala (“fight s.o.”, seen as

conative of qatala (“kill s.o.”))

Badawi et

al.

(2004:60)

IV ʾafʕala causative ʾajra (“cause s.th. to flow”) my sample

V tafaʕʕala reflexive tahaffaza (“prepare o.s.”) my sample

VI tafaʕala reciprocal tawafaqa (“come to an agreement”) Holes

(2004:103)

VII infaʕala passive inbahara (“be dazzled, blinded”) my sample

VIII iftaʕala benefactive ibtaʕa (“buy s.th.”) my sample

IX ifʕalla colours and physical

defects

ihmarra (“be or go red”) Badawi et

al.

(2004:61)

X istafʕala seeking s.th. istagfara (“ask pardon”) Wright

(1962, vol.

1:pt.

2,§63)


of 95

To demonstrate how different Forms might modify the meaning of a root, we might take a

look at the verbs listed for the root h k m in Wehr. The root conveys a sense of judgement,

and the Form I verb hakama simply means to pass a judgement. The Form II verb hakkama

can be interpreted as a causative of Form I, as it means to make someone a judge. The Form

III stem hakama has the conative sense of prosecuting someone, whereas ʾahkama from Form

IV expresses the event of causing a state of firmness. The Form V verb tahakkama displays

the reflexive meaning of making oneself the judge over something, and the Form VI verb

tahakama has the reciprocal sense of bringing each other before the judge. The Form VIII

verb ihtakama can express a self-benefactive notion, as it can denote appealing for a legal

decision. Finally, the Form X verb istahkama means to be strengthened, and thus seems to

denote the result of the Form IV verb.

Even though the verbs derived from the root h k m exemplify many of the meanings

attributed to the Forms in Table 1, it is already clear that the generalisations about the Forms

are too simplified. For instance, the resultative meaning of the Form X verb does not fit the

sense of seeking something which was highlighted in the overview. The reality is that most

Forms can express a variety of meanings, which might be hard to unify.

In addition, the distinctions are obscured by the fact that all the above stems have various

readings, some of which are the same for different Forms. For instance, the verbs of both

Forms V and VIII include the meaning of having one’s way. My sample shows that a fair

amount of synonymy is not uncommon.

Many roots also give rise to meanings which seem unrelated. This is the case for the root q m

r, which is used to form the Form I verb qamara (“gamble”), as well as ʾaqmara (“be


of 95

moonlit”). It can even be hard to see the link between different readings of the same stem, as

is evidenced by the verb qabbala (“kiss s.o., go south”). When analysing the verb Forms, we

must thus take care to make sure any characterisation of them is flexible enough to allow for

the above facts.

2.2 The middle voice

The term ‘middle voice’ has been used in many different ways across languages, and many

different approaches to it have been developed. We will start by looking at the approach of

Kemmer (1993) in 2.2.2, before moving on to the affectedness approach in 2.2.3. These are

the approaches most relevant to the study at hand, and we will look at how they might relate

to the middle Forms in Arabic. The ‘middle Forms’ investigated here will be Forms V, VII

and VIII, as well as the I- and U-patterns, which will also be referred to as ‘Forms’ for

convenience. I identified these Forms as belonging to the middle in Ajer (2014), a study

which will be discussed in its own right in 2.2.4. The chapter is concluded by section 2.2.5,

where I argue that the notion of affectedness we encountered in 2.2.3 should be expanded to

account for the different classes of middle subjects in Arabic.

Before embarking on a discussion of these approaches, however, some clarifications of

terminology are in order.

2.2.1 Clarifications of terminology

All predicates will be said to describe ‘eventualities’. This term subsumes both ‘events’,

which involve change, and ‘states’, which do not (Smith 1997:xiii–xiv).


of 95

When referring to the thematic roles of the participants of an event, I will rely on the term

‘initiator’ to mean the entity effecting change, and ‘undergoer’ to mean an entity which

undergoes change as a result of the event (Ramchand 2008:24,64). My data does not seem to

justify more fine-grained distinctions between ‘agent’ and ‘causer’, and ‘patient’ and ‘theme’.

Thus, in (1) below, John is the initiator, and the cart the undergoer. The example is taken

from Ramchand (2008:64).

(1) John pushed the cart.

A participant benefiting from an event is a ‘beneficiary’, and one of which a state is

predicated to hold is a ‘stateholder’, (Carnie 2013:231; Kratzer 1996:123).

For reasons that will be explained in 2.4, I will not use the terms ‘external’ and ‘internal’

argument when referring to the specific roles of the arguments. I will rather rely on the terms

defined above. I will use the terms ‘subject’ and ‘object’ not only to mean the syntactic

subject and object of the verb in a given construction, but also to refer to those of the verb’s

arguments which are allowed to fill these syntactic positions. Unless otherwise specified, I

am referring to surface subjects and objects. In the discussion of the argument roles in

Ramchand (1997) in 2.4, reference will be made to deep subjects and objects. This will be

made clear by describing them as ‘canonical’.

2.2.2 The approach of Kemmer (1993)

Kemmer (1993:210) seeks to unify all the ‘middle voice’ phenomena across languages under

the notion of ‘low elaboration of events’. Verbs with this property are characterised by the


of 95

fact that the different components of the events they describe are not properly distinguished

from each other. In Arabic, low elaboration of events seems to arise due to what she terms

‘low distinguishability of participants’ (Kemmer 1993:73).

According to Kemmer (1993:243), the core area of the middle is what she terms ‘inherently

reflexive verbs’, for which the roles of initiator and undergoer are expected to devolve upon

the same entity. This makes the two roles less conceptually distinct from each other than

usual, and thus gives rise to low participant distinguishability. As we saw in Table 1, Form V

is often associated with reflexive meanings. Many verbs of this Form are inherently reflexive,

as is the case for tatahhara (“clean o.s.”) (Kemmer 1993:66,72–73).

Kemmer (1993:142–145) identifies several other classes of verbs displaying low participant

distinguishability. One of these classes is ‘spontaneous verbs’, which assign no initiator role.

I will follow her in using that term rather than the term ‘anticausative’, as such verbs do not

necessarily have causative counterparts (Alexiadou and Doron 2012:24–25). Their low

participant distinguishability stems from the fact they construe the events as occurring

spontaneously, whereas there is always something bringing about events in the real world

(Kemmer 1993:144–145). The Form VII verb insaha (“pour forth”) is an example of a

spontaneous verb.

Kemmer’s semantic property of low elaboration of events is meant to encompass all

categories of verbs subsumed under the middle voice crosslinguistically. However, she

highlights the fact that languages vary as to which of these categories are marked

morphosyntactically as belonging to the middle voice (Kemmer 1993:21,238–239). In my

investigation of the Arabic middle Forms in Ajer (2014), which will be discussed in 2.2.4, I


of 95

established that the Arabic middle seems to mark some of the categories Kemmer identifies

as having low participant distinguishability. The question is therefore which exact subset of

the semantic middle it is the Arabic middle Forms encode.

In this paper, I argue that the relevant subset can be picked out by the notion of ‘subject-

affectedness’, which applies to some categories within the semantic area of the middle

(Kemmer1993:73). We will therefore turn to a discussion of affectedness next.

2.2.3 Affectedness

As defined earlier, an undergoer argument is one the event causes a change in. Such

arguments have often been said to be ‘affected’ (Anderson 1979; Jaeggli 1986). For instance,

the Form II verb jallaka (“sharpen s.th.”) has an affected object, as the event leads to a

change in its property of sharpness. Likewise, the movement predicated of the object of the

Form I verb nabada (“remove s.o./s.th.”) is also seen as a type of change, and thus marks it

out as an affected object.

The notion of affectedness has often been related to the middle voice, for instance by Roberts

(1987). According to his analysis, the subject of the English middle must be an affected

argument (1987:187–192,210). This can be demonstrated by the middle construction in (2),

which is taken from Roberts (1987:206). The baseball is the subject of the verb, and is an

affected argument as it undergoes movement in the event.

(2) This baseball throws well.


of 95

The subjects of some middle verbs in Arabic seem to fit this notion of affectedness. For

instance, we see that the subject of the Form V verb taraddada (“be thrown back”) would be

affected. As we will see in 2.2.4, however, the middle Forms in Arabic contain many verbs

where the subject does not undergo change as a direct result of the event. Some middle

subjects do not undergo change at all, which is the case for subjects of stative verbs, of which

the Form VIII verb ibtaraza (“be conspicuous”) is an example. Such stateholder subjects

would not fit Anderson’s (1979) idea of an affected argument, yet we will see that stative

verbs are common across the middle Forms.

It is thus clear that middle subjects in Arabic cannot be unified by this traditional notion of

affectedness. Rather than abandoning it, however, I argue in 2.2.5 that the notion of

affectedness must be extended to account for the different types of middle subjects found in

Arabic.

The approaches we have looked at so far seem either too broad or too narrow to capture the

Arabic middle. In order to see what characterises the middle voice in Arabic, we will now

take a look at the analysis of Ajer (2014).

2.2.4 The analysis of the Arabic middle in Ajer (2014)

When investigating the Arabic middle in Ajer (2014), I focused on Forms V, VII and VIII. It

seemed likely that these Forms belong to the middle, as many of the meanings associated

with them in the Arabic grammars seem to fit Kemmer’s notion of low event elaboration.

For instance, they are all mentioned to have reflexive meanings (Ryding 2005:530,555,565;

Wright 1962, vol. 1:pt. 2,§47,52,57). As we have seen, reflexive verbs are characterised by


of 95

low event elaboration, as the roles of initiator and undergoer overlap. All these Forms are

also claimed to give rise to resultative verbs (Ryding 2005:555,565; Wright 1962, vol. 1:pt.

2,§47,57). Resultatives encode states which have been brought about by an initiator, though

the focus is on the states themselves rather than on what caused them (Kaufmann

2007:1683,1700). These verbs might be exemplified by the Form V verb tahaddada (“be

determined”). The fact that the initiator is ignored gives rise to low event elaboration.

The descriptions of several other verb Forms make it seem as though they might also express

middle meanings. For instance, we have seen that Form VI is said to encode reciprocals, a

category Kemmer (1993:112) relates to the middle. However, I chose to look at Forms V, VII

and VIII, as these seemed to have the strongest connection to the middle (Ajer 2014:12-13).

For comparison, I also included Forms I, II and IV in my study.

I found that Forms V, VII and VIII, as well as the I- and U-patterns of Form I, seem to

encode the middle. An overwhelming majority of the verbs of these Forms in my sample

conform to Kemmer’s criterion of low event elaboration. It seems that low event elaboration

in Arabic middle verbs arises either through the lack of an initiator, or through the fact that

the initiator also plays another role in the event (Ajer 2014:14,40-41). In other words, the

subjects of these verbs are not prototypical initiators.

By contrast, Forms II and IV have a strong tendency to express active meanings. The A-

pattern, however, commonly encodes both active and middle meanings (Ajer 2014:34,41). As

it is widely assumed that the A-pattern gives rise to the readings which are the closest to the

root readings, as we saw in section 2.1, this indicates that the middle voice is as basic as the

active, rather than being derived from it. A similar viewpoint has been argued by Kaufmann


of 95

(2007:1687) and Klaiman (1991:24) among others.

Even though Forms V, VII and VIII and the I- and U-patterns all seem to encode the middle

voice, my findings indicate that there are some differences between them.

Form V is the only of the middle Forms to include many verbs where the initiator and

undergoer overlap, which are the core readings of the middle according to Kemmer. A

reflexive verb such as tarawwaha (“fan o.s.”) can exemplify this class of verbs. However,

Form V can also convey a variety of other middle meanings. It commonly expresses events

which involve no initiator, which is the case for stative, resultative and spontaneous verbs. In

addition, many Form V verbs denote events which benefit the subject, or in which the subject

is emotionally involved (Ajer 2014:24–26). These types of verbs will be discussed below.

Form VII verbs mainly encode events that are construed as having no initiator, with

resultative, stative and spontaneous readings being the most common. The intransitive verb

insalaʕa (“split, break open”) has a spontaneous meaning typical of this Form (Ajer 2014:22).

Stative and spontaneous verbs are also common in Form VIII. However, most Form VIII

verbs are what I will term ‘involved’ verbs (Ajer 2014:19). One type of involved verbs

denote events that take place in the mind of the subject. Examples are cognition verbs such as

ihtasaba (“take into account”) and emotion verbs such as ihtadda (“be furious, agitated”). As

it is often not specified what brings about these eventualities, they are characterised by low

event elaboration (Ajer 2014:5,19–20; Kemmer 1993:128–130).

Another type of involved verbs is those I term ‘emotive’ verbs. Even though the events


of 95

described by these verbs take place outside the subject’s mind, they are still associated with

the emotions or cognition of the subject. For instance, a verb such as ihtala (“strive”) entails

that the subject is working hard, and thus encodes a more concrete, physical event than an

emotion or cognition verb. However, part of the verb’s meaning is that the subject is

particularly committed to the action, or is working towards a specific goal. Thus, the event of

striving differs from the event of working in that only the former implies a mental

involvement on the part of the subject (Ajer 2014:19–20).

Kemmer (1993:133) notes that emotionally charged speech acts might belong to the middle,

but it seems as though all types of emotionally charged actions pattern with the middle in

Arabic (Ajer 2014:24). Emotive verbs also seem to fit conceptually with the idea of the

middle. As the emotional or cognitive capacities of the subject are tied up in the event, some

effect in the subject is associated with it. Thus, it might be construed as being affected by the

event as well as initiating it. It can therefore be argued that the subject plays more than one

role in emotive acts, giving rise to low event elaboration (Ajer 2014:15).

A last type of involved verbs is verbs with benefactive meanings, such as ibtaʕa (“buy”) or

istama (“bargain, haggle”). As these actions are normally construed as being carried out for

the subject’s benefit, the subject is benefactor as well as initiator, giving rise to low event

elaboration (Ajer 2014:20; Kemmer 1993:74,78). Further evidence that self-benefactive

readings might belong to the middle comes from Ugaritic, which is another Semitic language

with a verbal system very similar to that of Arabic. In addition to reflexivity and reciprocity,

the middle Forms in Ugaritic are reported to express actions that benefit the initiator (Pardee

1997:138).


of 95

The subject might be construed as being particularly motivated for carrying out self-

benefactive acts, and these events might therefore also be associated with mental involvement

on the part of the subject. For this reason, I class them as involved verbs (Ajer 2014:20).

As we have seen, the unifying property of involved verbs is that the subject’s mental

involvement is associated with the event. The idea that this could be syntactically relevant

might seem unusual. However, one of the theta system primitives proposed by Reinhart

(2002:231) is a feature specifying that a participant is in a mental state which might motivate

the event. I am therefore not the first to propose such an idea.

We have now seen the typical readings for Forms V, VII and VII, as analysed by Ajer (2014).

The I- and U-patterns were also identified as belonging to the middle, and similarly to Form

VII, they normally express events which lack initiators. Spontaneous verbs are common, but

states are predominant (Ajer 014:33-35).

As shown by Table 1, the I- and U-patterns are sometimes said to encode temporary and

permanent states respectively. This corresponds to the distinction between stage-level and

individual-level predicates, which goes back to Carlson (1977). While it is true that the U-

pattern is mainly used for individual-level states such as hasuba (“be highborn”), it also

contains some stage-level states such as batuʾa (“be slow”). Furthermore, both stage-level

states such as kariya (“be asleep”) and individual-level ones such as lahina (“be intelligent”)

are common in the I-pattern. It therefore does not seem as though the notions of stage- and

individual-level states can help us distinguish between the two patterns.

My analysis in Ajer (2014) indicates that there are good reasons to assume that Forms V, VII


of 95

and VIII and the I- and U-patterns of Form I encode the middle voice. Even though these

Forms express many of the same middle readings, it varies from Form to Form which

readings seem to be preferred.

What remains to be seen, however, is what the syntactically relevant property of the middle is

in Arabic. Which subset of verbs with low event elaboration do they pick out? Furthermore,

what are the syntactically relevant differences between the Forms? In the current study, I will

therefore continue investigating the middle Forms identified in Ajer (2014), and address these

questions.

From the discussion above, it is clear that the property of low event elaboration in the Arabic

middle arises as the verbs do not have prototypical initiators as subjects – either their subjects

are not initiators at all, or they are both initiating the action and experiencing some change or

mental involvement in relation to it. Crucially, our theory of the Arabic middle needs to

capture this observation. Establishing that the subjects of middle verbs are not prototypical

initiators is the first step towards an analysis. The next step is to figure out what these

subjects have in common, so that we can describe them in positive terms. Is there any

property which holds of all of these subjects? This is what we will look at next.

2.2.5 An extended notion of subject-affectedness in Arabic

A prototypical initiator is characterised by bringing about the event, and is not construed as

experiencing its effects. Middle subjects do not share this property of non-affectedness.

Whether they initiate the eventualities in question or not, we will see that they all seem to be

affected by it in some way. I will therefore argue that the crucial property of the Arabic

middle is that it must be able to have an affected argument as subject.


of 95

However, as has already been alluded to, I will propose that an extended notion of

affectedness is the relevant one for our analysis. I have identified three subclasses of affected

subjects in Arabic, based on how they are construed to be affected. I term these classes

‘directly influenced’, ‘mentally influenced’ and ‘lastingly influenced’ respectively. The

reason I use the term ‘influenced’ rather than ‘affected’ in the names of these classes, is that

languages might differ as to which of them pattern as affected. We will come back to this

point in 2.5. We will now look at how the different types of middle subjects in Arabic fall

into these classes, and try to define each class.

Many of the middle readings involve undergoer subjects. These conform to the typical idea of

affected arguments, as they undergo change as a result of the event. Reflexive, resultative and

spontaneous readings normally have such undergoer subjects. The subject arguments in these

readings undergo change as a direct result of the eventuality, which is why I term them

‘directly influenced’. They can be defined as follows:

(3) A directly influenced argument is one which undergoes change as a result of the

eventuality.

The change in a directly influenced argument might be a change in property, such as that

which is predicated of the subject of the Form VII verb insalaʕa (“split, break open”). The

subject of this spontaneous verb clearly changes in the event, as it goes from being whole to

being split open.


of 95

Movement might also constitute such a change. The subject of the spontaneous Form VII

verb insaha (“pour forth”) is directly influenced, as the event leads it to undergo movement.

The change in a directly influenced argument might also relate to it being created or

destroyed in the event, which can be called a change in its ‘physical extent’ (Ramchand

1997:76,117). It is such a change in physical extent which makes the subject of the resultative

Form VIII verb imtahaqa (“be destroyed”) a directly influenced argument.

There are certain subjects I analyse as directly influenced, which undergo a change that is not

as readily apparent. For instance, it might be hard to see why the subject of the reflexive

Form V verb tarawwaha (“fan o.s.”) should pattern as directly influenced. However, the

fanning event leads the subject to feel the air touching its skin, and this in itself might be

construed as a change in property. In addition, the normal objective of fanning oneself is the

more obvious change in property of cooling oneself down, and it could be that this effect and

the direct effect of the fanning itself are conflated. I therefore view the subject of this verb as

directly influenced, as the subject feels an impact from the movement of the air itself, and this

impact might lead to a change in the subject’s body temperature.

Now that we have reviewed directly influenced subjects, it is time to look at the mentally

influenced ones. From the discussion of the Arabic middle in 2.2.4, it is clear that some effect

in the subject might be associated with an eventuality, without necessarily being a direct

result of it. The emotive, self-benefactive, emotion and cognition verbs that together make up

the class of involved verbs, display this subtler type of affectedness. The events described by

these verbs are associated with the mental involvement of the subject, leading the subject to

be construed as more affected than it would have otherwise. It is this type of affected


of 95

argument I term ‘mentally influenced’, and it can be defined as follows:

(4) A mentally influenced argument is one whose mental involvement is associated with

the eventuality.

We have already seen that the emotive Form VIII verb ihtala (“strive”) entails commitment

and effort on the part of the subject, meaning that the subject is really putting its mind to the

task. It is this mental involvement which makes it a mentally influenced argument.

It is similarly obvious that the subject of a self-benefactive verb such as the Form VIII istama

(“bargain, haggle”) has particular goals in mind it is trying to achieve. Thus, it is clear that it

has the property of mental involvement which defines a mentally influenced argument.

However, it might not be as easy to see that mental involvement is implied by all other self-

benefactive verbs. This can be demonstrated by the Form VIII ibtaʕa (“buy”), which does not

seem to be associated with a particular mental state in the subject. Nevertheless, self-

benefactive verbs do seem to pattern as though their subjects are mentally influenced,

regardless of whether their mental involvement is made clear or not. This might be attributed

to the fact that self-serving motives might be expected when the initiator itself benefits from

the action, leading this argument to be seen as mentally invested.

The Form VIII verbs ihtasaba (“take into account”) and ihtadda (“be furious, agitated”) have

already been used to exemplify cognition and emotion verbs. As these events take place in

the minds of the subjects, there is no doubt that they are mentally influenced. I will also argue

that perception verbs such as the I-pattern verb našiqa (“smell”) have mentally influenced


of 95

subjects, as such sensory experiences clearly influence the mind.

It is possible for a subject to be both directly influenced and mentally influenced at the same

time. This is the case for the subject of the Form VIII verb irtajafa (“tremble, shudder”). The

movement predicated of the subject makes it directly influenced, and it is in addition

associated with mental involvement, as trembling is often brought on by fear. Thus, the

subject can also be construed as being mentally influenced.

The last class of affected subjects in Arabic are the lastingly influenced ones. This property

holds of the subjects of stative verbs. It might seem hard to reconcile stative readings with the

idea of affectedness, as affectedness is often tied to the notion of change. Stative readings, on

the other hand, by definition express unchanging situations. However, it is clear that stative

readings belong to the middle, as they are common in all the middle Forms and rare in the

active Forms (Ajer 2014:28–30). The question is thus how stative subjects can be tied to the

notion of affectedness.

When looking more closely, stateholders are not as different from undergoers as they may

seem. Whereas the main effect predicated by a dynamic event is a change in the undergoer

argument, the main effect predicated by a stative situation is a lack of change in the

stateholder. Whatever it was that brought the state about, it has had a continued effect on the

stateholder argument, and the stative verb predicates that this effect still holds of its subject.

Interpreted in this way, it seems natural to analyse also the lack of change predicated of an

argument as a type of affectedness. I have termed the subjects of stative verbs ‘lastingly

influenced’, as these verbs predicate that the effect of some previous event on the subject is


of 95

still lasting. They can thus be captured by the following definition:

(5) A lastingly influenced argument is an argument of which it is predicated that the

effect of some previous event, whether obvious or not, still holds.

It is certainly true that stage-level verbs predicate that the subject is still in a state brought

about by a previous event. This can be demonstrated by the I-pattern verb kariya (“be

asleep”). Clearly, the state of being asleep is the result of a previous event of falling asleep,

and the stative verb predicates that the effect of this event still holds of its subject. It should

thus be uncontroversial to class the subject of such a verb as lastingly influenced.

It might seem as though a distinction should be made between stage- and individual-level

states, as we normally do not think of the latter as having been caused by any particular

event. Thus, the subject of an individual-level verb such as the U-pattern verb hasuba (“be

highborn”) might not appear to be lastingly influenced. However, it seems as though

individual- and stage-level states display the same patterning in my data. It is also worth

noting that every state holding in the world will have been brought about by something. For

instance, it is the event of being born into a noble family which leads to the state of being

highborn. Thus, we have seen that stage- and individual-level verbs pattern alike, and that

also individual-level states can be reconciled with the idea of predicating that the effect of a

previous event is still holding. I will therefore treat the subjects of both types of stative verbs

as lastingly influenced, and will not make a distinction between them.

An argument cannot be both directly influenced and lastingly influenced at the same time. As

only predicates expressing change give rise to directly influenced arguments, and only those


of 95

expressing a lack of change can give rise to lastingly influenced ones, it is a logical

impossibility for these two types of arguments to arise in the same reading.

However, a verb may have one reading which requires a directly influenced argument, and

another requiring a lastingly influenced one. This is the case for the Form VII verb inbagata

(“be taken by surprise, be aghast”), whose subject is a directly influenced argument in the

former, eventive reading, whereas it is a lastingly influenced argument in the second, stative

reading. It is worth noting that both readings imply the mental involvement of the subject,

which is thus also mentally influenced in each case. It is therefore clear that mentally

influenced arguments are compatible with both the other types of affectedness.

The three types of arguments defined so far can all be construed to be affected in some way.

As we will see in chapter 4, the subjects of the middle verbs in my sample belong to one of

these classes, with very few exceptions. I therefore argue that all arguments belonging to any

of these three classes pattern as affected in Arabic.

Affected arguments contrast with arguments which are not construed as affected. In Arabic,

subjects falling under this class are the ones which bear the role of initiator, without being

construed as undergoing change or as being particularly mentally involved. I will call these

non-influenced arguments. This last class subsumes all arguments which do not belong to any

of the previously defined categories. The subject of the Form IV verb ʾanʕasa (“make

sleepy”) is a typical non-influenced argument, as it initiates change in another argument,

without any implication that it is particularly mentally invested in the event.

We have now extended our notion of affectedness to subsume the three classes of directly


of 95

influenced, mentally influenced, and lastingly influenced arguments. Using this notion of

affectedness, it seems as though middle verbs in Arabic can be characterised by having

affected subjects. My findings, which will be presented in chapter 4, support this position.

The idea of the three different types of subject-affectedness will be explored more in that

chapter, as well as in 2.5.

In my analysis in chapter 4, I will not only seek to establish what unifies the middle Forms,

but also what differentiates them. It has already been argued that the middle in Arabic can be

unified by the overarching property of subject-affectedness. As this unifying property can be

divided into three subcategories, it seems plausible that the differences between the Forms

might relate to differences in type of affectedness. This idea will be pursued in my analysis.

By investigating how the middle voice is instantiated in Arabic, and using empirical data

from Arabic to inform an extended theory of subject-affectedness relevant to the middle, this

study seeks to increase our understanding of the middle voice in general.

I will also investigate whether lexical aspect might play a role in the middle Forms. The

analysis of Danks (2011), which we will look at in the next section, indicates that this

phenomenon might be relevant to Arabic verb Forms. It seems a worthwhile pursuit to look

into the interaction between aspect on the one hand, and the middle and affectedness on the

other, as these phenomena have been tied to each other in the literature (Kemmer 1993:x;

Roberts 1987:213; Tenny 1987:2).

In addition, I will look at how the middle meanings and aspectual properties of the different

Forms affect their valency, as there are indications that the three phenomena might be linked


of 95

(Borer 2005:53; Dixon and Aikhenvald 2000:7,12; Kemmer 1993:210; Ramchand 1997:114-

115,128-130; Tenny 1987:69-72). As will be seen in section 2.4.1, my analysis in Ajer (2014)

also points to a connection between the middle and valency.

A more thorough treatment of the middle verb Forms in Arabic might therefore increase our

knowledge of the phenomena of voice, lexical aspect and valency, as well as the interplay

between them. It might also elucidate our understanding of the function of derivational verbal

morphology in Arabic. This will of course be of particular interest to students and researchers

of Arabic, but should also be of interest to other linguists, as it is widely assumed that

morphemes are meaning-bearing (Danks 2011:23). One would therefore expect that when a

root can enter into so many different Forms, the choice of Form should signal some

difference in meaning.

Before we get to the analysis of the middle Forms, however, it is necessary to review the

relevant literature on aspect and valency.

2.3 Lexical aspect

As it is the contributions of derivational verbal morphology we are investigating, it is only

lexical and not grammatical aspect which will concern us here. For simplicity’s sake, I refer

to lexical aspect simply as ‘aspect’.

Lexical aspect, or ‘aktionsart’, refers to how inherent properties of a predicate and its

arguments influence the way the event is viewed as unfolding in time. What this means will

become clear when we go through the aspectual classification famously argued for in Vendler


of 95

(1957). In this classification, a distinction is made between the four classes of ‘states’,

‘activities’, ‘accomplishments’ and ‘achievements’.

Eventualities such as knowing and loving are states. Verbs of this class are characterised by

the fact that they describe unchanging situations. Due to the lack of change, they are ‘stative’

rather than ‘dynamic’. Furthermore, these eventualities are not instantaneous but hold over

time, and are therefore said to be ‘durative’. Durative adverbials such as “for many years” are

often used to test for durativity, and (6) shows that this adverbial can indeed be combined

with the state “know”. States can further be distinguished from durative, dynamic

eventualities by the fact that they are largely incompatible with continuous tenses, as can be

seen in (7) (Vendler 1957:144,146-149).

(6) I knew him for many years.

(7) *I was knowing him for many years.

The other three classes all involve change, and thus express dynamic eventualities, or

‘events’ (Vendler 1957:144,149). Activities are durative events, which are also characterised

by being atelic, meaning that they have no set terminal point. An event such as running in the

park is atelic, as we are given no indication as to when it will come to an end. Its atelicity is

demonstrated by the fact that it cannot be combined with the completive adverbial “in an

hour”, as shown by (8). We also see that it is compatible with the durative adverbial “for an

hour”, as well as with continuous tenses, showing that it is durative and dynamic (Vendler

1957:144-146,149). Thus, it is clear that this event is an activity.


of 95

(8) He ran/was running in the park (for an hour/*in an hour).

Accomplishments are similar to activities in that they are durative events, but they differ in

that they have a set terminal point. They are thus ‘telic’. An accomplishment such as running

to the store will necessarily be complete once the store has been reached. Its dynamicity and

durativity is shown by its compatibility with continuous tenses, as can be observed from (9).

(Vendler 1957:145–148). It is clear from (10) that it is telic, as it can occur with the

completive adverbial “in an hour”. However, it is incompatible with the durative adverbial

“for an hour”, as it is non-completive.

(9) He was running to the store.

(10) He ran to the store (*for an hour/in an hour).

Achievements, which are the last of Vendler’s classes, are also telic. Contrary to

accomplishments, however, they are not portrayed as unfolding over time. Rather, they are

seen to be instantaneous. They are therefore often called ‘punctual’ events. An event of

recognising someone is an achievement, as the subject changes from not having to having

recognised the person from one moment to the next. Achievements often sound strange with

continuous tenses, but are compatible with punctual adverbials such as “at what moment”, as

(11) and (12) demonstrate. (Vendler 1957:144,146–147,149)

(11) ? I was recognising her.

(12) At what moment did you recognise her?


of 95

Achievements are sometimes contrasted with punctual events which do not lead to new result

states, such as sneezing. These are called ‘semelfactives’, and they are normally analysed as

being unbounded (Smith 1997:29). However, my data does not seem to support an opposition

between punctual events which lead to a new result state and those which do not. As the

event is completed in either case, I will treat all punctual events as bounded, and will not

distinguish between them. Thus, I will refer to ‘punctual events’ rather than achievements

and semelfactives.

As Ramchand (1997) points out, the above classification is concerned with the aspectual

properties emerging from the VP or sentence as a whole, rather than from the properties of

the verb itself (Ramchand 1997:4-5). This can be demonstrated by the fact that even though

the same verb is employed in (8) and (10) above, (8) is an activity and (10) an

accomplishment.

For this reason, Ramchand seeks to develop a theory of aspect which looks at the contribution

the verb itself makes to the overall aspectual classification (Ramchand 1997:4,108). As we

are concerned with the contribution of derivational verbal morphology, Ramchand’s

approach seems appealing.

In her system, one of the most important aspectual properties of a verb is its internal temporal

structure, which refers to how many distinct conceptual moments the eventuality it expresses

contains (Ramchand 1997:126–7). On these grounds, there is reason to uphold the distinction

between states and punctual events, as the two types of eventualities clearly involve different

timelines. Whereas a state consists of a single undifferentiated time interval, the timeline of a


of 95

punctual event is broken up by an abrupt change (Ramchand 1997:120–4,127,131).

Ramchand dispenses with Vendler’s classes of activities and accomplishments, as they are

distinguished by whether other elements in the sentence give rise to telicity. It is information

contributed by the PP which leads (8) above to be classed as an activity and (10) as an

achievement, whereas the verb is the same in both instances. Thus, such a distinction has no

place in Ramchand’s theory, and she proposes that the two classes should rather be viewed as

one. She classes them together, as the timelines of both activities and accomplishments

contain several distinct conceptual moments (Ramchand 1997:120,123-4,131). As they

therefore encode gradual change, I will term them ‘gradual events’. In my analysis, I will rely

on the three different classes of states, punctual events and gradual events, as identified by

Ramchand.

One obvious way in which the properties of the verb’s temporal structure interact with the

middle, is that a stative verb will always have a lastingly influenced subject. Thus, verbs with

this property must necessarily have an affected subject and must therefore belong to the

middle.

The question is whether there are other, less obvious ways in which the aspectual properties

of the verb interact with the roles it assigns to its arguments. Are there any correlations

between the aspectual properties of a verb and the roles it assigns to its subject? If so, does

the connection between aspectual properties and argument roles stem from separate

requirements imposed by the verbal morphology, or is there a causal relation between the

two? These issues will be explored in the analysis in chapter 4.


of 95

Ramchand (1997:126-128,191,213) also takes into account the fact that the aspectual

properties of the verb interact with the argument roles it assigns, and proposes a set of

argument roles she views as relevant to the overall aspectual classification of the VP. Her

argument roles and their relevance to my analysis of the Arabic verb Forms will be discussed

in chapter 2.4.

Ramchand’s theory is not only concerned with aspectual properties related to how many

conceptual moments an event has, but also with whether it has a set endpoint or not. If it has

a set endpoint at sentence level, she terms the event ‘telic’, whereas she calls it ‘bounded’ if

the verb and its arguments provide it with an endpoint (Ramchand 1997:136-137). Only the

latter is relevant to her theory, and as this is indeed the relevant notion to our investigation as

well, I will follow her in using the term ‘boundedness’ rather than the term ‘telicity’.

The temporal structure of the event partly determines its properties for boundedness. States

are necessarily unbounded, as an endpoint would lead to the time interval becoming

differentiated. Punctual events, on the other hand, will all be treated as bounded, as the

change they describe is construed to take place from one moment to the next, providing them

with an inherent endpoint (Ramchand 1997:123,131).

The situation is less straightforward for gradual events, as their boundedness depends on the

properties of the verb’s arguments (Ramchand 1997:128-30). If such verbs are most

commonly used with arguments that give rise to boundedness, I will classify them as

bounded in my sample, whereas I will classify them as unbounded if the opposite is the case.

In order to see the overall patterning of the different Forms and middle meanings for

boundedness, all the verbs needed to be classified in this way. However, as will become clear


of 95

in chapter 4, my analysis does not rest heavily on whether these gradual verbs are interpreted

as bounded or unbounded.

As we have seen, Ramchand’s system allows for boundedness to arise as a result of the

verb’s temporal structure, or through the interaction of the verb and its arguments. However,

she also postulates that the verb itself might impose boundedness restrictions. She argues that

what we think of as a verb can be decomposed into one unit containing lexical information,

and one introducing aspectual requirements. In her analysis, Scottish Gaelic verbs contain

such an aspectual head, which serves to impose boundedness requirements. However, it is

possible to imagine that an aspectual head might impose restrictions on the verb’s temporal

structure as well (Ramchand 1997:142–3,146,158,165).

This begs the question of whether the Arabic verb Forms might introduce aspectual

requirements like the aspectual heads in Scottish Gaelic. If this is the case for the middle

Forms, a further question is whether the affectedness requirements and aspectual

requirements associated with them are separate from each other, or whether one follows from

the other.

These questions presuppose that there is reason to believe that the Arabic verb Forms might

have bearing on aspect. The findings of Danks (2011), which we will turn to next, seem to

indicate that this is indeed the case.

2.3.1 Lexical aspect in Arabic

The topic of aktionsart in Arabic is discussed by Danks (2011), who appeals to it in his

analysis of Forms III and VI. As the Form III pattern is faʕala and the Form VI pattern


of 95

tafaʕala, the two Forms are clearly morphologically related. Danks therefore sets out to

investigate their syntactic and semantic relationship (Danks 2011:23,83). He argues that the

unifying property for Forms III and VI is unboundedness, as nearly all verbs of these Forms

that are in actual usage allow unbounded readings (Danks 2011:191,211,255).

That Forms III and VI might be characterised by a compatibility with unboundedness is an

interesting finding, as it suggests that the verb Forms might encode aspectual properties.

However, I fear that the property of unboundedness might be too general to set these two

Forms apart from all the other Forms, and that one might also need to appeal to other

properties in their description. I will briefly touch on this in chapter 4.

Danks investigates not only what Forms III and VI have in common, but also how they can

be distinguished from each other. He argues that the difference between them is that Form VI

normally has a reduced valency as compared to Form III (Danks 2011:129). The concept of

valency and its relevance to the study at hand will be discussed in the next section.

2.4 Valency

The term ‘valency’ is used in syntax to refer to the number of direct arguments a verb

requires, or ‘s-selects’. A verb such as “sleep”, which only s-selects for one argument, is

called ‘monovalent’. In the same way, a verb like “hit” which s-selects for two arguments is

called ‘divalent’, and so forth.

However, valency oppositions might not be as clear-cut as they seem, as there are indications

that some objects have a closer relationship to the verb than others. Building on Krifka (1989;

1992), Ramchand (1997:114-115) proposes an opposition between ‘real’ objects and


of 95

‘pseudo’-objects, to which we will now turn.

Real objects are characterised by having ‘aspectual’ roles, meaning that the properties of the

roles they are assigned contribute to the overall aspectual classification of the VP. For

instance, if a change in physical extent is predicated of the object, boundedness only arises if

it is quantised (Ramchand 1997:128-130). Thus, the quantised object “a house” makes (13)

bounded, whereas (14) is unbounded due to the unquantised object “houses”. This can be

seen from the fact that (13), but not (14), is compatible with the completive adverbial “in a

week”. The opposite is the case for the durative adverbial “for a week”, which implies

unboundedness (Vendler 1957:145).

(13) Danny built a house (in a week/*for a week).

(14) Danny built houses (*in a week/for a week).

Real objects correspond to directly influenced arguments in our terms. They are characterised

by the fact that they undergo a change which is mapped to the timeline of the event

(Ramchand 1997:115,127). In the above examples, this is demonstrated by the fact that the

house or houses in question are created as the event proceeds in time. The object of the Form

VIII verb ibtaʕa (“buy”) is a real object, as it goes from being in the possession of one person

to being in the possession of another as the event unfolds. Ramchand (1997:127)

distinguishes between different aspectual roles for canonical objects, but these do not seem

relevant to the Arabic middle. I will therefore not distinguish between different types of real

objects in my analysis.

Real objects contrast with pseudo-objects, which are assigned a non-aspectual role. Such


of 95

objects undergo no change which is relevant to the verb’s timeline, and have no bearing on

the aspectual classification of the VP. They merely serve to modify the event (Ramchand

1997:124-128). The difference in how a real object and a pseudo-object is perceived can be

demonstrated clearly by the I-pattern verb taʕima. The verb can express the event of eating,

in which case there will be a change in the physical extent of the object which is impossible

to overlook. In this reading, the verb requires a real object. However, it can also have the

meaning of tasting or enjoying, in which case no change in the object is clearly and

necessarily implied. This means that the object is a pseudo-object in the second reading.

Ramchand (1997:47,88,111-112,220) identifies different properties for real objects and

pseudo-objects in Scottish Gaelic which indicate that the opposition between them might be

syntactically relevant. If not all objects are equal, this suggests that determining a verb’s

valency might not be as simple as adding up the number of arguments it has, but that also the

types of arguments might play a role. The behaviour of these objects in Scottish Gaelic point

to the fact that a verb might be more strongly transitive if it has a real object than if it has a

pseudo-object (Ramchand 1997:168,220). As will be seen in chapter 4, this seems to hold in

Arabic, as pseudo-objects appear to be more compatible than real objects with Forms which

are normally intransitive.

Ramchand (1997:191) also postulates different roles for canonical subjects. The distinctions

she makes do not seem to fit the distinctions that are relevant to the Arabic middle Forms,

and they will therefore not be discussed here. However, I wish to discuss the terms ‘external’

and ‘internal’, which she applies to her canonical subject roles and object roles respectively

(Ramchand 1997:127,191). I will now outline why I do not think these notions are useful to

the analysis of the Arabic middle.


of 95

Whereas the roles assigned to initiators and stateholders are classified as external, the roles

assigned to directly influenced canonical object arguments are seen as internal (Ramchand

1997:127,191). However, neither the class of external nor internal roles can subsume all the

roles assigned to the subjects of the middle Forms. As we touched on in 2.2.5, and as will be

seen more fully in 2.5 and chapter 4, the defining property of the Arabic middle seems to be

that the subject must be affected. Not only directly influenced subjects pattern as affected, but

also lastingly and indirectly influenced subjects seem to belong to this class. Stateholders are

necessarily lastingly influenced arguments, whereas even initiators can be indirectly

influenced. Thus, both these supposedly external roles are compatible with affectedness, a

property which has often been linked to internal argumenthood (Roberts 1987:210; Tenny

1987:69–70).

This might lead the reader to wonder whether I should expand the notion of an internal role,

as I have expanded that of affectedness. However, not all affected subjects could be classified

as internal arguments. This can be demonstrated by the Form VIII verb iddaʕa (“accuse

s.o.”), which takes a directly influenced object as well as an indirectly influenced initiator

subject. As this verb has an affected object, it is hard to argue that the affected initiator could

really be an internal argument. Thus, it would be more natural to classify the subject as

external, showing that its affectedness cannot arise from internal argumenthood. It would be

just as strange to postulate that all affected arguments are external, as we would then have to

argue that both the subject and the object of the above verb are really external arguments. It is

therefore clear that the property of affectedness is independent of internal or external

argumenthood, and that middle subjects in Arabic cannot be unified by appealing to such

notions.


of 95

We have thus seen that the notions of external and internal roles do not seem to be relevant in

analysing the Arabic middle, and that they are independent of the property of affectedness,

which unifies the argument roles assigned to middle subjects. In light of this, the use of the

terms ‘external’ and ‘internal’ roles would be especially problematic, as the terms seem to

imply that the properties of the roles derive from their structural position. For that reason,

these terms will not be used here.

This is not to deny the existence of any structural asymmetries, however. Whereas I argue

that there are three types of affected subject arguments – directly influenced, indirectly

influenced and lastingly influenced – the object position seems to be more restricted in the

types of affected arguments it allows. This will be discussed in 2.5.

As already mentioned, Ramchand’s research indicates that there might be a syntactic reality

to the difference between real objects and pseudo-objects, and that a verb which takes a real

object might be more strongly transitive than one which only allows a pseudo-object. A more

complex notion of valency, which takes into account both the type and the number of

arguments a verb s-selects for, might be relevant to studies of the middle voice.

The middle voice has been claimed to be valency-reducing as compared to the active voice,

and is often linked to intransitivity (Croft, Shyldkrot, and Kemmer 1987:184; Dixon and

Aikhenvald 2000:1,12). The difference between real objects and pseudo-objects raises

questions regarding whether the supposed valency-reduction in the middle voice only relates

to the number of arguments, or also the types of arguments. Could it be that middle verbs

which have the same numerical valency as their active counterparts might still differ from


of 95

them in not allowing real objects? Do middle verbs resist transitivity per se, or only real

objects?

If the middle is sensitive to a distinction between real objects and pseudo-objects, which as

we have seen differ in their aspectual properties, this would highlight an interesting

connection between the middle voice, aspect, and valency. However, future research will

have to determine whether this more complex idea of valency might elucidate our

understanding of the valency-reducing properties of the middle. A thorough comparison of

middle and active verbs would be necessary to answer such a question, which is beyond the

scope of this study.

Nevertheless, my analysis in chapter 4 might have bearing on the question, as I will appeal to

the difference between real objects and pseudo-objects in my investigation of the valency of

the different middle Forms. As we will see, it seems as though it is the different readings,

rather than the middle itself, which to varying degrees resist transitivity. These readings,

however, do indeed seem to be sensitive to the distinction between real objects and pseudo-

objects.

I also highlighted a possible connection between a Form’s valency and the types of readings

it gives rise to in Ajer (2014), where I explored the sheer numerical valency-reduction and

intransitivity associated with the middle Forms. We will now move on to a discussion of this.

2.4.1 Valency in Arabic

Whereas Forms V, VII and the U-pattern are described as intransitive in the literature, it is

claimed that Form VIII and the I-pattern are compatible with both transitivity and


of 95

intransitivity (Holes 2004:101; Ryding 2005:114,565).

My analysis in Ajer (2014:37-38) shows that all the middle Forms have a tendency to yield

intransitive verbs, and that they normally have a reduced valency as compared to the active

Forms. However, there is considerable variation in how strong these tendencies are, which

largely supports the claims made in the Arabic grammars. In order to explain these

differences, I noted that each Form’s likelihood of being transitive corresponds with their

likelihood of having an initiator argument which does not overlap with the undergoer.

According to my analysis, Form VII is the least likely to allow an initiator and is also the

most strongly intransitive Form, whereas Form VIII is the most compatible with initiator

subjects and is fairly often transitive. The other Forms fall somewhere in between (Ajer

2014:39).

It should be noted that the data for the I- and U-patterns in Ajer (2014) might be obscured by

the fact that verbs which occur in several of the three Form I patterns with no change in

meaning were included in the analysis for each of the patterns they occur in. As we will see

in chapter 3, such verbs will be excluded from analysis in the current study, as they seem to

pattern differently than other verbs of these patterns.

We have thus seen that the findings from Ajer (2014) indicate that the Arabic middle seems

to favour intransitivity, and that the strength of this tendency in each Form might be

explained by the types of middle readings they commonly display. In my analysis in chapter

4, I will explore whether the extended subject-affectedness approach to the middle might

inform our understanding of the link between middle readings and valency.


of 95

Before we turn to that, however, it is necessary to develop the theory of affectedness further.

That is what the next section will deal with.

2.5 A hierarchy of affectedness

So far, I have extended the notion of affectedness, and argued that the subject of an Arabic

verb patterns as affected if it is directly influenced, mentally influenced or lastingly

influenced. The definitions of these three subclasses of affected arguments, given in (3), (4)

and (5), are repeated below for your convenience.

(3) A directly influenced argument is one which undergoes change as a result of the

eventuality.

(4) A mentally influenced argument is one whose mental involvement is associated with

the eventuality.

(5) A lastingly influenced argument is an argument of which it is predicated that the

effect of some previous event, whether obvious or not, still holds.

Any argument which does not fall into any of these three classes is a non-influenced

argument, and is not construed to be affected by the eventuality.

We have also seen that Ramchand makes a distinction between pseudo-objects and real

objects. The former seem to be non-influenced arguments in our terms, and therefore pattern

as non-affected, whereas the latter seem to be directly influenced and thus affected. This

raises the question of whether it is possible for an object to belong to any of the other classes

of affected arguments.


of 95

In my Arabic data, lastingly influenced objects do not seem to exist. On the other hand, there

are objects which might be construed as being mentally influenced. This is the case for the

object of the Form IV verb ʾazʕaja (“alarm s.o.”), of which the event predicates a change

from being calm to being alarmed. As this change is in the object’s mental state, it is mentally

influenced. However, the object is clearly directly influenced as well, due to the change in

property. In fact, all the objects which might be construed as mentally influenced are also

directly influenced. The question is therefore whether the notion of being mentally influenced

is useful to the analysis of objects. In order for it to be relevant, one would need to show that

objects which are both directly and mentally influenced pattern differently from those which

are only directly influenced. I will leave this issue for future research.

Either way, the most crucial property of real objects is that they are directly influenced. The

fact that this notion is relevant to both subjects and objects, whereas the notions of mentally

and lastingly influenced arguments mainly seem to be relevant to subjects, point to an

asymmetry between directly influenced arguments and the other classes.

When taking a closer look, it seems as though there is an asymmetry between mentally and

lastingly influenced arguments as well. As the referent of a mentally influenced argument

must be construed as being mentally involved in the event, it must be an animate entity.

Furthermore, the arguments of the Arabic verbs in my data are never merely mentally

influenced, they are always either initiators, or directly or lastingly influenced arguments as

well. Similar restrictions do not hold of lastingly influenced arguments, revealing an

asymmetry between lastingly and mentally influenced arguments.


of 95

As already mentioned, the reason the terms for these three subclasses contain the word

‘influenced’ rather than ‘affected’, is that there might be cross-linguistic differences as to

whether all these classes of arguments pattern as affected. The analysis Roberts (1987:187–

192,210) proposes for the English middle constructions makes it seem as though this

prediction might hold, as these constructions appear to only allow directly influenced

subjects. This is yet another indication that directly influenced arguments are less restricted

than mentally and lastingly influenced ones.

I therefore postulate that there is a hierarchy of influenced arguments, based on the

asymmetries observed above. It ranges from the types of arguments which are the most likely

to pattern as affected cross-linguistically, to those which are the least likely to pattern as

affected. I also predict that the higher an argument is in the hierarchy, the less restricted

should its behaviour be in phenomena where affectedness is relevant.

(15) Hierarchy of affectedness

directly influenced > lastingly influenced > mentally influenced > non-

influenced

We have also seen that I have identified the following classes as affected and non-affected in

Arabic:

(16) Affected arguments

Directly influenced

Lastingly influenced


of 95

Mentally influenced

(17) Non-affected arguments

Non-influenced

I chose the more neutral term ‘influenced’ rather than ‘affected’ to describe the subtypes of

affected and non-affected arguments, so as not to prejudice cross-linguistic analyses of

affectedness. However, as I will henceforth only discuss how these classes pattern in Arabic,

I will call the types of affected arguments ‘directly affected’, ‘lastingly affected’ and

‘mentally affected’, and the non-influenced arguments ‘non-affected’.

We have already seen that direct affectedness is a property of the undergoer role, lasting

affectedness a property of the stateholder role and mental affectedness a property of the role

of involved argument. As it is the affectedness properties of the roles which seem relevant to

my analysis, I use the terms for the types of affectedness as a shorthand for the roles

themselves.

We have now seen that there seem to be three different types of affectedness relevant to

Arabic, and that they can be arranged in a hierarchy. All three properties seem to be relevant

to the analysis of subjects, whereas it is the property of direct affectedness which seems

crucial to the definition of a real object. Direct affectedness holds the highest position in the

affectedness hierarchy, and this property should therefore be the least restricted. The

hierarchy also makes predictions about how these types of arguments might pattern cross-

linguistically.


of 95

The fact that the notion of direct affectedness seems important to both subjects and objects,

might lead the reader to wonder whether a verb might have a directly affected subject at the

same time as a directly affected object. A directly affected argument undergoes change as a

direct result of the event, so the question is whether an event can predicate change of two

different arguments at the same time. This is indeed possible. For instance, the I-pattern verb

taʕima has both a directly affected subject and object when it expresses the event of eating. In

this event, the object undergoes a change in property related to its physical extent, whereas

the subject undergoes a change in property related to its fullness. However, even though it is

not disallowed for a verb to have two directly affected arguments for the same reading, it

seems to be dispreferred. This can be seen from the fact that out of the 367 middle verbs in

my sample which allow a directly affected argument, only 21 allow there to be two for the

same reading.

Even though we can construe an event as leading to change in two different arguments at

once, we tend to construe them as mainly affecting one argument at a time. For instance, in

the event of a man building a house, we see the main result of the event to be that the house

comes into existence. The event might result in a change in the man as well, as the hard work

might for instance make him stronger, but this is not the main outcome we associate with the

event. Thus, I argue that most events are seen as directly effecting change in only one

argument, and I think this is where the dispreference for having more than one directly

affected arguments stems from.

The first step towards an extended theory of affectedness is to motivate the need for

extending it, by identifying grammatically relevant categories and show why and how they

should be unified with the notion of affectedness. That is what I am trying to achieve in this


of 95

paper. The second step is formalising the definitions of these categories, and coming up with

diagnostics for reliably identifying them. I will have to leave this second step for future

research.

In my analysis in chapter 4, I will motivate the different types of affectedness I have

postulated by showing that they are relevant to the analysis of the Arabic middle. First,

however, it is time to look at how I undertook my analysis.

3 Methodology

In my analysis of the Arabic middle Forms, I relied on a data sample I developed. This

sample built on the one used for Ajer (2014), which we will examine, before turning to how

the current sample was created.

For the reasons outlined in 2.2.4, I included verbs of Forms I, II, IV, V, VII and VIII in my

sample for Ajer (2014). The sample was organised in a spreadsheet, and contains 1247 verbs.

This amounts to 19% of the verbs Wehr lists for these Forms. It was intended to be small

enough for me to manage, but still big enough that I might be able to reveal trends in these

verb Forms. In order to get a reasonably random sample, I included every 20th verb in Wehr,

along with other verbs from the same root which belong to one of the relevant Forms (Ajer

2014:16-17).

For each verb, I compiled information from Wehr regarding its meaning and valency,

comparing its properties to those of other verbs from the same root. The verbs were then

classed as active, middle or passive, and categorised according to the types of readings they


of 95

give rise to, particularly drawing on Kemmer’s middle categories (Ajer 2014:13-18). These

classifications will not concern us here, however, as a different approach to the middle is

used.

When undertaking the current study, it seemed sensible to focus on the Forms which had

been analysed as belonging to the middle in Ajer (2014). This was the case for Forms V, VII,

VIII and the I- and U-patterns of Form I. Looking at Forms which had already been identified

as middle Forms meant that I could take the analysis a step further, and look at what unifies

and distinguishes them. Due to the limited time available, I also chose to build on the sample

used in Ajer (2014), instead of constructing a completely new sample.

The current sample can be found in appendix D, and includes all the information found in the

sample of Ajer (2014). In addition, I have classified the verbs of the middle Forms according

to argument roles and aspectual properties.

As already mentioned, the middle Forms investigated in this study are Forms V, VII and VIII,

and the I- and U-patterns. Before we proceed, it is worth noting that not all verbs displaying

the I- and U-pattern will be analysed as belonging to these Forms. When the same root occurs

in different Form I patterns with no change in meaning, I will consider there to be one verb

which belongs to a ‘mixed pattern’, rather than to the different patterns it displays. Such

mixed-pattern verbs contrast with ‘single-pattern’ verbs, which cannot be used synonymously

in different patterns. As the different patterns of mixed-pattern verbs are non-contrastive, we

would predict that their behaviour might differ from that of single-pattern verbs. This does

indeed seem to be the case, and mixed-pattern verbs will therefore be excluded from the

analysis. Henceforth, when referring to I- and U-pattern verbs, I am thus only referring to


of 95

single-pattern verbs.

All verbs belonging to one of the middle Forms will be termed ‘middle verbs’. In order to get

an idea of the number of middle verbs in my sample, the reader can refer to Table 2 below. In

total, my sample includes 479 middle verbs.

Table 2. Number of verbs in sample compared to total of verbs in Wehr, as reported by

Danks (2011:28).

My sample Wehr Proportion of verbs in Wehr

included in my sample

I-pattern 68 - -

U-pattern 17 - -

Form V 198 953 21%

Form VII 55 267 21%

Form VIII 141 606 23%

In the rest of the paper, the numbers and percentages given relate to the total number of verbs

of the relevant Form(s) in my sample, unless otherwise specified. For convenience, all

percentages have been rounded up or down to the closest whole number.

The middle verbs in my sample were categorised according to their aspectual properties and

the argument roles they assign. I relied on the aspectual classes of states, punctual events and

gradual events, as well as on a distinction between verbs allowing bounded and unbounded

readings. For the definitions of these categories, the reader is referred to 2.3. The subject

roles employed were those of initiator, directly affected, mentally affected and lastingly

affected, as discussed in 2.2.1 and 2.5. When a verb allows several of these roles to hold of its


of 95

subject, whether they all hold at the same time or different roles hold in different readings, I

will say that it allows an ‘overlap’ of roles. Furthermore, transitive verbs were classed

according to whether they allow real objects or pseudo-objects, as defined in 2.4.

Many verbs have different readings with different properties, which might make them or their

arguments fall under several of the different categories above. We also know that the subject

roles are not all mutually exclusive. This is important to bear in mind when interpreting the

percentages given. As many verbs belong to several different categories for the same

property, in that they for instance can have different readings belonging to different aspectual

classes, the percentages for the different categories will not add up to 100%.

In addition to assigning the verbs in my sample to the different classes discussed, I also

constructed a questionnaire to further investigate the aspectual properties and valency of

some of the verbs. The questionnaire used, as well as details about how it was administered,

can be found in appendix E. The methodology used in making and administering the

questionnaire is based on that of Danks (2011).

Modern Standard Arabic has no native speakers, and it is mainly a language of education and

formal communication. One would therefore expect that the speakers who have the firmest

grasp of its grammar are native speakers of an Arabic colloquial who have undertaken their

education in Modern Standard Arabic. I approached several people who fit these criteria, but

due to the limited time available, only two people returned the completed questionnaires to

me.

Due to the low number of participants, and the low number of verbs and constructions tested,


of 95

the questionnaire responses should not be used to make generalisations about the data.

However, they might still provide us with some indication of which constructions seem to be

more or less acceptable for the verbs in question. Sentences from the questionnaire will

therefore be used in my analysis to exemplify the trends observed in my data, but no

conclusions will be made solely on the basis of these sentences.

The translations of questionnaire sentences are my own. The key to the abbreviations used in

the glosses can be consulted in appendix B. A sentence will be treated as acceptable only if

both my participants found it to be fully acceptable, whereas sentences that either of my

participants had doubts about will be treated as less acceptable.

From the comments, it seemed as though many sentences were rejected due to the verbs

being uncommon. This seemed to be the case for three verbs, which occurred in thirteen

sentences. In the time available, it was not possible to rectify this, and the sentences thus had

to be discarded from the analysis. To avoid this problem in future studies, it should be made

sure that the verbs tested are in common usage. It would also be useful to be able to ask the

participants more follow-up questions.

In future research, the usage of the different verbs should be tested with more participants, so

that the trends seen in the responses can give a better indication of the way these verbs are

generally used. If the responses are to be used to generalise about the Forms themselves, a

higher number of verbs and constructions need to be tested. In the current study, I could only

test a limited number of verbs in a limited number of constructions, in order to make it

manageable for my participants to fill in the questionnaires.


of 95

This concludes the discussion of the background to this study, as we have now looked at both

the theoretical concepts underpinning it and the methodology used. It is therefore time to turn

to the analysis itself.

4 Analysis

4.1 Subject-affectedness in the middle Forms

In section 2.2.5, I proposed that the defining property of the Arabic middle is that it must

allow an affected subject. It is clear from Table 3 that this generalisation fits the middle verbs

in my sample well.

Table 3. Subject-affectedness.

Affected

Nr. %

I-pattern 68 100%

U-pattern 17 100%

Form V 196 99%

Form VII 55 100%

Form VIII 136 96%

The table shows that nearly all the verbs of each Form have at least one reading where the

subject is affected. Whether they also allow other readings seems irrelevant. Thus, even

though a verb such as the Form VIII iqtada (“lead, be led”) has a non-affected subject in the

first reading, the affected subject in the second reading ensures that it complies with the

restriction on the middle.


of 95

All the verbs of the I- and U-patterns and Form VII obey this restriction, and nearly all Form

V and VIII verbs do. As 99% of the verbs of the middle Forms in my sample allow affected

subjects, the few verbs that do not satisfy this requirement will be treated as lexical

exceptions.

Thus, it seems that what unifies middle Forms is the requirement that they allow affected

subjects. However, we must also figure out what distinguishes them. As we have established

that this defining property of affectedness consists of several categories, it would be

conceptually pleasing if we could postulate different requirements for each Form related to

type of affectedness.

We will see if this is possible when we try to come up with requirements particular to each

Form in sections 4.3-4.7. Before we turn to that, it might be useful if we look at how such

requirements might be formalised, as understanding how the requirements could be imposed

might make them clearer. However, the reader should note that what I mainly wish to argue

for in this paper are the requirements themselves, rather than any particular formalisation.

4.2 A feature approach to the Arabic middle

We will start by looking at how the general requirement on the middle might be formalised. I

propose that the middle in Arabic arises when there is a middle voice head present, which is a

voice head introducing a [+affected] feature. This feature is assigned to the subject.

(18) Middle voice head: Voice [+affected]


of 95

I postulate that such a middle voice head is introduced by the morphology of each of the

middle Forms. Furthermore, I will argue that in addition to the [+affected] feature, the middle

voice heads in the different Forms introduce other features particular to them.

It does not make a great difference to the argument whether the features that will be proposed

here are independent of each other, or whether some are diacritics of others. I do not wish to

take a strong view on this point, but will for simplicity’s sake treat all features as

independent. All features proposed will be positive and privative, as postulating binary

features would allow for more possibilities than seems to be needed.

In the following sections, we will identify properties for the I- and U-patterns and Form VII

which unify the class of verbs belonging to each Form. However, we will not arrive at any

property which can be used to capture the whole class of Form V or VIII verbs. The question

is therefore how one can account for these Forms.

The answer might of course be that they are defined by properties not considered here. It

would therefore be desirable for these two Forms to be investigated in more detail in future

research. If no defining property can be found, however, the answer might lie in diachrony. It

might be that the middle voice heads of these Forms imposed absolute requirements at some

earlier stage in the language, but that these requirements have later been relaxed and are thus

not as active synchronically.

This hypothesis might be supported by the fact that the Arabic verb Forms do not seem to be

fully productive, as you cannot necessarily use the verb Forms to derive new verbs from any


of 95

root you wish. However, they do not seem to be completely fossilised either, as new verbs

and loanwords can be made to fit into the different Forms (Cuvalay-Haak 1997:89). If the

morphology of the middle Forms introduce a middle voice head, one would thus expect that

this head might not be completely active anymore.

It therefore might be that Forms V and VIII verbs have originally constituted unified classes,

but that they have started allowing new readings as the requirements have been relaxed,

eventually obscuring their properties. This is what I will tentatively assume here.

A diachronic view might inform our understanding of the middle Forms in general. As we

have seen, the middle verbs must at least have one reading which complies with the subject-

affectedness requirement imposed by the middle voice head, but they also allow other

readings. Similarly, we will see that the verbs of the I- and U-patterns and Form VII can have

other readings in addition to those that fit their respective requirements. This can be

explained if we assume that only readings adhering to these requirements were allowed at the

outset, but that the verbs have gradually taken on more readings through lexical extension.

One might wonder why it should be of interest to investigate requirements which might not

be syntactically active and which might not capture the different verb Forms fully. However,

I argue that postulating possible original requirements for each Form might shed light on the

issues explored in this study, even if they do not hold perfectly synchronically. Firstly, an

investigation into these requirements might still bring us closer to explaining the distribution

of verbs across these Forms. Furthermore, the different kinds of middle heads proposed might

contribute equally much to the discussion of the middle voice, regardless of whether the

requirements they impose are synchronic or diachronic. Lastly, the effects of these


of 95

requirements might still be observed in the semantic and syntactic behaviour typical of each

Form, as will be seen in the following sections.

We will now go through each Form, and look at their properties regarding type of

affectedness, aspect and valency. After identifying the key characteristics of each Form, I will

try to postulate what the original requirements might have been, and assess how well they

seem to hold synchronically. I will further look at how these requirements might explain the

rest of each Form’s properties.

We will start from the Forms which can be captured most easily. In the tables given,

particularly interesting results will be highlighted in blue.

4.3 The U-pattern

Table 4. Argument roles assigned by U-pattern verbs

Direct Mental Lasting Real object Pseudo-object

Nr. 8 2 17 0 0

% 47% 12% 100% 0% 0%

Table 5. Aspectual properties of U-pattern verbs

Unbounded Bounded State Punctual event Gradual event

Nr. 17 2 17 0 10

% 100% 12% 100% 0% 59%

As the key property of the middle is subject-affectedness, it would be preferable if the

specific requirements we identify for each Form relate to type of affectedness. It does indeed


of 95

seem like such a requirement holds of the U-pattern, as all the verbs of this pattern in my

sample allow lastingly affected subjects, as is shown by Table 4.

We see that verbs such as saʕuba (“be difficult”), garuba (“be a stranger, be strange”) and

katura (“be numerous, increase”) obey this requirement, as all of them have at least one

reading where the subject is lastingly affected.

Thus, I postulate that the following middle voice head is introduced by the U-pattern

morphology:

(19) The middle voice head of the U-pattern: Voice [+affected][+lasting]

However, it is clear from Table 4 that synchronically, the U-pattern allows other readings in

addition to stative ones. After stative readings, change of state readings are the most

common. The verb katura (“be numerous, increase”) exemplifies such a reading. It is easy to

see how these readings might have developed, as it seems logical for a verb expressing a state

to be extended to express the change resulting in that state as well.

The information collected from Wehr and from the different grammars indicate that the U-

pattern is indeed defined by the stative readings to which it gives rise (Badawi et al. 2004:60;

Holes 2004:101; Ryding 2005:457; Wright 1962, vol. 1:pt. 2,§38). However, one of my

participants notes that the stative readings of many of the U-pattern verbs seem archaic, and

that these verbs are now more commonly used to express changes of state. This might

indicate that the Form’s focus is shifting away from lastingly affected subjects.


of 95

Even if it is correct that the stative readings of U-pattern verbs are becoming less common, I

will still focus on the proposed original requirement of allowing lastingly affected subjects.

This is because it seems as though it is this requirement which can give us the most insight

into what might unify U-pattern verbs, as well as into the semantic and syntactic properties

they typically display. Furthermore, if the middle voice heads in Arabic were only fully

active at a previous stage, investigating their properties at that point in time is what might

contribute to our understanding of the middle.

As Table 5 shows, all U-pattern verbs in my sample must allow unboundedness. This can

easily be explained by a requirement of allowing lastingly affected subjects, as this means all

verb must have stative readings, which are by definition unbounded. The Form’s focus on

lasting affectedness might lead it to be associated with unboundedness overall. It is therefore

unsurprising that boundedness should be dispreferred also for the change of state readings,

and that none of the U-pattern verbs in my sample encode punctual events. The verb katura

displays behaviour typical of a U-pattern verb in this respect, as the change of state it can

denote is the unbounded event of increasing.

However, even though boundedness is rare, some U-pattern verbs do allow bounded

readings. This is expected if the strong tendency for unboundedness is an effect of the

requirement of lasting affectedness, rather than being a requirement in itself. Example (20)

shows that the verb samura (“be brown, turn brown”) can give rise to a bounded reading, as it

can co-occur with the completive adverbial fi saʕat-in (“in an hour”).


of 95

(20) samura al-haql-u fi saʕat-in

turn.brown.PF.3MSG DEF-field-NOM in hour-NOM.INDEF

“The field turned brown in an hour.” (Questionnaire: 3)

We have thus seen that a requirement of lasting affectedness can account for the aspectual

properties of the U-pattern. The last property which warrants explanation, is the fact that the

Form seems to disallow objects.

It would of course not be possible for the verbs to have real objects in the stative readings, as

no change could be predicated of the object. The fact that the core readings of the U-pattern

are incompatible with real objects, might be expected to lead to an association with

intransitivity in general. I will argue that this is the reason none of the other readings seem to

allow real objects either, and the reason why pseudo-objects seem to be rejected as well.

Thus, it seems as though a very strong preference for intransitivity has developed in the U-

pattern due to the relative intransitivity imposed by the requirement of lasting affectedness. I

see the Form’s intransitivity as a result of this preference, rather than as a requirement in

itself.

In this section, we have seen that the U-pattern’s preference for unboundedness, its lack of

punctual readings, and its intransitivity, might all be seen to follow from a requirement of

lastingly affected subjects. I take this as support for my proposal that the morphology of the

U-pattern introduces such a requirement. It also indicates that the type of middle voice head

present in a middle verb might strongly influence its valency and aspectual properties,


of 95

without separate restrictions of this kind being imposed on it.

4.4 Form VII

Table 6. Argument roles assigned by Form VII verbs


Nr. 53 15 26 0 0

% 96% 27% 47% 0% 0%

Table 7. Aspectual properties of Form VII verbs


Nr. 39 40 26 37 27

% 71% 73% 47% 67% 49%

It seems as though Form VII might also be characterised by a requirement on type of

affectedness, as Table 6 shows that 96% give rise to directly affected subjects.

Form VII verbs such as insaha (“pour forth”), inzaʕaja (“be stirred up, feel uneasy”) and

infajara (“go off, explode”) fit this generalisation, as they can all have directly affected

subjects. This is clearly displayed in (21), where the nominative subject dimaʾun (“blood”)

undergoes movement as a result of the event, meaning that it is directly affected.

(21) insah-at dimaʾ-un min il-jurh-i

pour.forth.PF-3FSG blood-NOM.INDEF from DEF-wound-GEN

“Blood gushed from the wound.” (Questionnaire: 10)


of 95

However, the verbs inkataba (“subscribe”) and inkaffa (“abstain”) do not seem to give rise to

directly affected subjects. As these seem to be the only exceptions in my sample, they will be

treated as lexical exceptions, and disregarded in the further discussion of the Form.

I therefore postulate the following voice head for Form VII:

(22) The middle voice head of Form VII: Voice [+affected][+direct]

The fact that Form VII verbs have directly affected subjects means that they predicate some

change of them. It seems natural that verbs expressing change might be extended to express

the result of that change as well. This might explain why Form VII verbs commonly convey

stative meanings as well, as shown in Table 7. The verb inzaʕaja (“be stirred up, feel

uneasy”) is an example of a verb which can describe both a change in the subject and its

resultant state. The existence of such verbs demonstrates that Form VII allows other types of

affectedness to hold of its subjects synchronically, as long as all verbs have at least one

reading where the subject is directly affected.

The changes in the subject argument which are the core readings of Form VII, might as easily

be construed as bounded or unbounded. It is therefore unsurprising that both bounded and

unbounded readings are common in this Form, as Table 7 shows. We see that whereas insaha

(“pour forth”) expresses an unbounded event, and infajara (“go off, explode”) a bounded one,

inzaʕaja (“be stirred up, feel uneasy”) allows both.


of 95

We further see that it is particularly common for Form VII verbs to express punctual events,

an aspectual class which we will see is much rarer in the other middle Forms. The events

encoded by both inzaʕaja (“be stirred up, feel uneasy”) and infajara (“go off, explode”) are

punctual, as the event of being stirred up and the event of exploding are construed as

instantaneous.

The reason punctual readings are dominant in Form VII might have to do with the fact that

most of the other Forms seem unsuitable for predicating a punctual change in the subject, as

will be clear from the discussion of each Form. Several of them might readily express a

gradual change in the subject, however. Even though both types of change should be equally

compatible with Form VII, the fact that only gradual change in the subject is commonly

expressed by other Forms, might have lead punctual readings to become prevalent in Form

VII.

If one wishes to emphasise the fact that the subject undergoes change, having an object which

also undergoes change might obscure this focus. Moreover, we saw in section 2.5 that it is

rare for a verb to have two directly affected arguments. As the core property of Form VII is

the direct affectedness of its subjects, it is therefore unsurprising that it seems to disallow real

objects. This resistance to real objects seems lead to an overall association with intransitivity,

leading the Form to reject pseudo-objects as well. Thus, the focus on directly affected

subjects in Form VII seems to be the cause of its intransitivity.

We have therefore seen that the valency of Form VII, and to some extent its aspectual

properties, follow from its requirement of allowing directly affected subjects.


of 95

4.5 The I-pattern

Table 8. Argument roles assigned by I-pattern verbs


Nr. 32 26 54 5 11

% 47% 38% 79% 7% 16%

Table 9. Aspectual properties of I-pattern verbs


Nr. 66 15 54 11 34

% 97% 22% 79% 16% 50%

Unlike verbs of the U-pattern and Form VII, it is clear from Table 8 that the class of I-pattern

verbs cannot be captured by a requirement on type of affectedness. It seems as though the

property of unboundedness is the one which comes closest to unifying all verbs of this Form.

Only the verbs rahiqa (“overtake”) and naqida (“escape”) seem to resist an unbounded

interpretation, and might thus be viewed as lexical exceptions. I therefore propose that the I-

pattern is characterised by requiring all verbs to have unbounded readings.

Postulating an unboundedness requirement might seem strange, as the requirements proposed

so far relate to the notion of affectedness. As there are many middle Forms, however, it is not

surprising that they might not all be distinguished from each other based on type of

affectedness alone. Thus, it seems plausible that the different Forms might also introduce

other types of requirements.

It might also be objected that a [+unbounded] feature appears somewhat unnatural. However,


of 95

I view this as an artefact of the terminology used. The last question raised by an

unboundedness restriction on the I-pattern, is whether it is imposed by the middle voice head

or an aspectual head. I will not take a strong position on this issue, but the interactions seen

between aspect and affectedness in 2.3 make it seem plausible that the same functional head

might introduce restrictions related to both phenomena. For convenience, I will therefore

assume that all the features I propose are introduced by the middle voice head.

I therefore argue that the I-pattern is underspecified for type of affectedness, and that

unboundedness is its defining property. Thus, I postulate the following voice head for the I-

pattern:

(23) The middle voice head of the I-pattern: Voice [+affected][+unbounded]

It is important to note that whereas the [+affected] feature is assigned to the subject, the

[+unbounded] feature pertains to the verb itself. However, as this technicality is not central to

the argument at hand, it is not indicated in the notation.

The fact that the I-pattern is underspecified for type of affectedness but unified through

unboundedness, can be demonstrated by the verbs tariya (“become wealthy”), salifa

(“boast”) and naziha (“be respectable”). The verbs differ from each other in that the subject

of tariya is directly affected, that of salifa is mentally affected and that of naziha is lastingly

affected. However, we see that they all display the unbounded readings required by the I-

pattern. Examples (24) and (25) indicate that tariya is indeed unbounded, as my participants


of 95

found it to be acceptable with the durative adverbial tiwala sanatin (“for a year”), whereas it

was less acceptable with the completive adverbial fi sanatin (“in a year”).

(24) tariya ar-rajul-u tiwala sanat-in

become.wealthy.PF.3MSG DEF-man-NOM throughout year-NOM.INDEF

“The man got wealthier for a year.” (Questionnaire: 34)

(25) ? tariya ar-rajul-u fi sanat-in

become.wealthy.PF.3MSG DEF-man-NOM in year-NOM.INDEF

? “The man got wealthy in a year.” (Questionnaire: 8)

As already mentioned, the requirement of unbounded readings does not prevent bounded

ones from occurring as well. For instance, the verb darima can both express the unbounded

event of burning, or the bounded, punctual event of catching fire.

Though neither bounded nor punctual events are disallowed in the I-pattern, it is clear from

Table 9 that they are not very common. It seems as though boundedness in general, and

punctuality in particular, are dispreferred in the I-pattern. This is natural, as one would expect

the requirement of allowing unbounded readings to lead to an association with

unboundedness, making the Form more resistant to readings that go against this association.

It is therefore also unsurprising that gradual verbs such as tariya (“become wealthy”) occur

more commonly.

As stative eventualities are the only ones which are necessarily unbounded, they might be


of 95

seen as the most strongly unbounded eventualities. A Form such as the I-pattern, which has

an unboundedness focus, should therefore be particularly compatible with states. Tables 8

and 9 show that this is indeed the case. The verb naziha (“be respectable”) is an example of a

stative I-pattern verb.

We would not expect the unboundedness requirement of the I-pattern to have any direct

effect on its valency. However, we have seen that this property is the reason for the

prevalence of stative readings, which are not compatible with real objects. Furthermore, there

are also many I-pattern verbs which have directly affected subjects, and we know that such

verbs tend to resist real objects as well. This seems to associate the I-pattern with

intransitivity, and it is therefore unsurprising that intransitive verbs such as tariya (“become

wealthy”) are predominant, as evidenced by Table 8.

However, as there is no direct connection between intransitivity and the Form’s core property

of unboundedness, it is natural that the association with intransitivity should be weaker than

for the U-pattern and Form VII. This is evidenced by the fact that objects are allowed in the I-

pattern, even though they are rare. As the main readings of the Form are mainly resistant to

real objects, it is also expected that pseudo-objects should be more common. The verb hawiya

(“love s.o./s.th.”) is an example of the former, whereas qabila (“accept s.o./sth., receive s.o.)

is an example of the latter.

It therefore seems as though a requirement of allowing unbounded readings can unify the

verbs of the I-pattern, in addition to shedding light on its valency and aspectual properties.

However, in order to determine whether such an unboundedness requirement is enough to


of 95

distinguish the I-pattern from all other Arabic middle Forms, the potentially middle Forms

not investigated in this paper must be researched further. As we saw in 2.3.1, Danks

(2011:191,211) uses unboundedness to capture Form VI, a Form which seems as though it

might be connected to the middle. Thus, it might be that Form VI is also a middle Form

characterised by unboundedness. However, as seen in 2.1 and 2.2.4, reciprocity is often seen

as a core reading of Form VI. One might speculate that there could be some property related

to affectedness underlying reciprocity, which might be used to unify the class of Form VI

verbs, distinguishing them from I-pattern verbs. Thus, until further research into this matter

has been undertaken, I will tentatively assume that the middle voice head of the I-pattern

contributes an unboundedness requirement.

4.6 Form VIII

Table 10. Argument roles assigned by Form VIII verbs


Nr. 82 95 56 51 20

% 58% 67% 40% 36% 14%

Table 11. Aspectual properties of Form VIII verbs


Nr. 105 87 56 75 87

% 74% 62% 40% 53% 62%

It is clear from Tables 10 and 11 that Form VIII cannot be unified by type of affectedness, or

by any aspectual property. The question is thus what might characterise this Form.


of 95

As already mentioned, there do not seem to be enough types of affectedness for each middle

Form to be distinguished on these grounds. We have also seen that it is common for the verbs

of the other middle Forms to allow several different subject roles. It might be that such an

overlap is more common in the Forms which do not have a strong preference for one type of

affectedness, and that this could be what distinguishes Form VIII. If the Form does have a

preference for overlapping roles, another question is whether it makes any difference which

roles it is that overlap. Table 12 therefore presents information on overlapping roles in Form

VIII.

Table 12. Overlapping roles in Form VIII

Overlapping roles Direct + other role Mental + other role Lasting + other role

Nr. 126 77 95 51

% 89% 55% 67% 36%

We see that 89% of Form VIII verbs assign several roles to their subjects. Though this means

that there are too many exceptions for all the Form’s verbs to be unified by this property, it is

the closest we have come to capturing Form VIII. I therefore hypothesise that there might

have been an original requirement of overlapping roles on the Form, which has later been

obscured as new readings have arisen.

However, in the next section, I propose that this is also the case for Form V. As both these

Forms seem to be characterised by favouring overlapping roles, I will term them ‘overlapping

Forms’. The two overlapping Forms seem to be distinguished by which roles they prefer to be

part of the overlap. As we see from Table 12, 67% of Form VIII verbs have a mentally

affected subject which is also assigned another role. Thus, it seems as though Form VIII does


of 95

not only prefer that its verbs assign several subject roles, but also that one of them is mentally

affected. This might be the result of an original requirement such as that in (26). The [+role]

feature is an underspecified feature which can stand in for any argument role, affected or not.

(26) The middle voice head of Form VIII: Voice [+affected][+mental][+role]

If it is indeed correct that there was originally a requirement of overlapping subject roles on

Forms V and VIII, it is easy to see why these Forms should be the ones to develop the most

readings that do not satisfy the original requirements. If they always required at least two

roles, of which one was underspecified, they would have given rise to a greater variety of

readings from the outset than what the less flexible requirements of the other Forms would

allow. New verbs might have been derived by analogy with the different kinds of readings

they allowed, and in this way, the semantic area they encoded might have been gradually

expanded. It seems natural that eventually some of the new verbs derived would not have any

of the readings initially permitted. This could therefore be the reason why the verbs of these

Forms seem harder to unify.

If they are overlapping Forms, this might also explain why Forms V and VIII give rise to a

much higher number of verbs than the other middle Forms. As the overlapping Forms would

allow for more flexibility than the other Forms in deriving new verbs, they would also expand

more rapidly.

I will therefore assume that the restrictions imposed by the middle voice head in (26) initially

held in Form VIII, but that they have later been obscured due to lexical extension. Even so,


of 95

the original requirement has led to a preference for verbs with mentally affected subjects

which are also assigned another role, a preference which can be observed synchronically.

This can be demonstrated by the verbs ihtawaja (“need s.th.”), intaqama (“take vengeance

for”) and iddaraʕa (“arm o.s.”). The subjects of these verbs are all mentally affected. In

addition, the subject of ihtawaja is lastingly affected, that of intaqama is the initiator, and that

of iddaraʕa is directly affected as well as being the initiator.

However, we also see that some Form VIII verbs have readings which are not at all related to

the original core readings, as we speculated might happen in overlapping Forms. The verb

igtassa (“be overcrowded”) is such a verb, as it only assigns one role, which is that of

stateholder.

It is now time to turn to the aspectual properties of Form VIII. Whereas it seems to favour

gradual verbs such as iddaraʕa (“arm o.s.”), states and punctual events are common as well,

as shown by Table 11. The fact that it allows verbs belonging to such a variety of aspectual

classes might be explained by its property of being an overlapping Form, as this results in it

giving rise to a variety of readings. However, its preference for gradual verbs is unexplained.

It also seems that it has a slight preference for unboundedness, without the reason for this

being obvious.

Having aspectual properties which are hard to account for, might actually be another result of

the overlap favoured by Form VIII. As overlapping Forms pattern less clearly for type of

affectedness and thus have less clear core properties, it might also be less clear how the rest

of their properties arise.


of 95

It is easier to understand why Form VIII should be the only middle Form which commonly

gives rise to transitive verbs. We have seen that the other middle Forms predominantly have

lastingly and directly affected subjects, whose resistance to real objects leads to an

association with intransitivity. The mentally affected subjects preferred by Form VIII,

however, do not cause any such association. This is because the mental involvement

characteristic of them can hold equally well of an initiator, as of a stateholder or undergoer.

Thus, Form VIII is the middle Form which can most easily express an event in which an

initiator acts on another entity, giving rise to more active-like readings.

It is also worth noting that real objects such as that required by intahara (“drive s.o. away”)

or ibtalaʕa (“swallow s.th.”), are favoured over pseudo-objects such as that required by

ištamma (“smell s.th.”). Example (27) demonstrates that ibtalaʕa takes an object, whereas

(28) shows that this object can become the passivised subject, indicating that it is indeed a

real object (Danks 2011:154-155,239).

(27) ibtalaʕa ar-rajul-u al- ʾakl-a

swallow.PF.3MSG DEF-man-NOM DEF-food-ACC

“The man swallowed the food.” (Questionnaire: 14)

(28) ubtuliʕa al- ʾakl-u

swallow.PASS.PF.3MSG DEF-food-NOM

“The food was swallowed.” (Questionnaire: 17)

In Ajer (2014:37), it was noted that an overwhelming majority of the verbs of the active

Forms II and IV have objects, and these mainly seem to be real objects. The fact that Form


of 95

VIII favours real objects over pseudo-objects is thus another trait which makes it reminiscent

of the active Forms.

However, Form VIII clearly differs from the active Forms in that many of its subjects are also

directly or lastingly affected. As we have seen, readings with these types of subjects often

show a relative resistance to transitivity. It is therefore unsurprising that Form VIII should

favour intransitive verbs overall, as we see from Table 10. The verbs intaqama (“take

vengeance for”) and iddaraʕa (“arm o.s.”) are examples of intransitive Form VIII verbs.

We have thus seen that a possible original requirement of mentally affected subjects might

explain the valency of Form VIII. However, we have not arrived at a property which can

unify the whole class of Form VIII verbs, as well as explain the aspectual properties of the

Form. Further research is needed to determine whether such a property exists. If no such

property can be found, this might be due to the fact that Form VIII seems to be an

overlapping Form. As such, it easily lends itself to lexical extension, which might have

obscured the core properties of the Form.

4.7 Form V

Table 13. Argument roles assigned by Form V verbs


Nr. 153 79 97 14 26

% 77% 40% 49% 7% 13%

Table 14. Overlapping roles in Form V


of 95

Overlapping roles Direct + other role Mental + other role Lasting + other role

Nr. 159 126 79 87

% 80% 64% 40% 44%

Table 15. Aspectual properties of Form V verbs


Nr. 175 93 97 78 132

% 88% 47% 49% 39% 67%

Table 13 shows that Form V cannot be unified by a requirement on type of affectedness,

whereas Table 14 indicates that it seems to favour overlapping roles, as mentioned in 4.6.

This tendency is weaker than for Form VIII, but it still seems to be the most characteristic

property of the Form. Rather than preferring one of the roles to be mentally affected, Form V

seems to prefer that one of them is directly affected.

Thus, I propose that the middle voice head of Form V might originally have imposed the

following requirements:

(29) The middle voice head of Form V: Voice [+affected][+direct][+role]

The verb tabarrara (“justify o.s., be justified”) displays the core properties of Form V, as it

cannot only have a directly affected subject, but can also assign the roles of initiator, mentally

affected and lastingly affected.

As for Form VIII, the fact that Form V is an overlapping Form seems to have led it to take on


of 95

readings which are further and further from the core readings, making it hard to unify the

whole class of Form V verbs. The core readings seem particularly distant from verbs whose

subjects are neither directly affected nor have overlapping roles, as is the case for tawajjaba

(“be necessary”).

As Table 15 shows, Form V is also similar to Form VIII in having a preference for

unboundedness and gradual events. The aspectual properties of the verb taqassa

(“investigate”) is thus typical, as it expresses a gradual, unbounded event. It is not evident

why Form V should display such a preference. However, as we saw for Form VIII, the less

clear core properties of the overlapping Forms might make it hard to account for their

aspectual properties. This does indeed seem to be the case for Form V.

For valency, on the other hand, Forms V and VIII do not pattern alike. As Form V has a focus

on directly affected subjects, we would expect it to have a stronger association with

intransitivity, and thus a stronger resistance to objects. As Table 13 shows, this prediction

seems to hold.

The resistance is not as strong as for Form VII, however. This might be because the focus on

directly affected subjects is somewhat obscured by the fact that it is an overlapping Form.

Thus, transitive Form V verbs occur, but they are not common.

As the focus on directly affected subjects mainly leads to real objects being disfavoured, it is

unsurprising that more Form V verbs give rise to pseudo-objects than real objects. The verb

tahassa (“drink s.th. noisily”) has a real object, whereas tajannaba (“avoid”) requires a

pseudo-object. The fact that tajannaba takes an object is demonstrated in (30), whereas (31)


of 95

shows that this object cannot become the passivised subject. This points to the fact that it is

indeed a pseudo-object.

(30) tajannaba al-walad-u mudarris-a-hu

avoid.PF.3MSG DEF-boy-NOM teacher-ACC-3MSG.POSS

“The boy avoided his teacher.” (Questionnaire: 24)

(31) * tujunniba al-mudarris-u

avoid.PASS.PF.3MSG DEF-teacher-NOM

* “The teacher was avoided.” (Questionnaire: 31)

We have thus seen that the active Forms and Form VIII seem to favour real objects over

pseudo-objects, whereas the opposite is the case for the I-pattern and Form V. This means

that the Forms which more commonly give rise to transitive verbs have a preference for real

objects, while the Forms more strongly associated with intransitivity are less resistant to

pseudo-objects. It therefore seems as though real objects are more strongly transitive than

pseudo-objects. This indicates a need for a more complex notion of valency, which accounts

for both the number and type of the verb’s arguments, an idea which was discussed in 2.4.

However, further developing the idea of complex valency is beyond the scope of this paper. I

merely wish to highlight that my findings seem to support a distinction between real objects

and pseudo-objects, of which the former pattern as more transitive.

We have seen that the valency of Form V seems to follow from its focus on directly affected

subjects, which is somewhat weakened by the fact that it seems to be an overlapping Form. If

my hypothesis about it being an overlapping Form is correct, it is unsurprising that it seems


of 95

hard to find a property which can unify all the verbs of the Form and account for its aspectual

properties. However, further research must determine whether this is indeed the case.

5 Conclusion

In this paper, I have provided considerations that motivate an extension of the notion of

affectedness. The traditional sense of affectedness relates to a change in an argument which

happens as a direct result of the event. I have termed this type of argument ‘directly

influenced’. In addition to directly influenced arguments, I argue that there are two other

types of arguments which might pattern as affected. These are ‘lastingly influenced’

arguments, of which a state is predicated to hold, and ‘mentally influenced’ arguments, which

experience a particular mental involvement in the event. Future research should formalise

these types of affectedness and identify diagnostics which pick out each type, as well as

investigate whether relevant subtypes exist.

I have proposed that these possible types of affectedness might be organised in a hierarchy, in

which directly influenced arguments are the most likely to pattern as affected, followed by

lastingly influenced and then mentally influenced ones. Cross-linguistic research is needed to

determine whether such a hierarchy of affectedness can be observed in other languages, and

which asymmetries there are between the different types.

The extension of affectedness is motivated by my analysis of five middle Forms in Arabic:

the I- and U-patterns and Forms V, VII and VIII. I propose that the Arabic middle is unified

by requiring all middle verbs to allow affected subjects. All the Forms discussed satisfy this

requirement, with only a handful of verbs violating it.


of 95

I have further shown that directly, lastingly and mentally influenced subjects all pattern as

affected in Arabic. The middle Forms seem largely to be distinguished by the type of

affectedness they require of their subjects. Whereas the U-pattern must allow lastingly

influenced subjects, Form VII must allow directly influenced ones. I hypothesise that Forms

V and VIII might originally have required their subjects to have overlapping roles. For Form

V, one of these roles must have been directly influenced, whereas it must have been mentally

influenced for Form VIII. Only the I-pattern seems to be underspecified for the type of

affectedness required. It rather seems to be characterised by the fact that its verbs must allow

unbounded readings.

Furthermore, I tentatively propose that these requirements might be imposed by different

middle voice heads introduced by the morphology of each Form. The requirements mainly

arise as the voice heads assign features to the subject. Whereas they might have been absolute

requirements in the beginning, they seem to have been relaxed over time, allowing a certain

amount of lexical extension in all Forms. The lexical extension seems to have been taken

particularly far in Forms V and VIII due to their affinity for overlapping subject roles, which

might be the reason why it is hard to unify all the verbs of each of these Forms. However, it

might also be the case that a more extensive investigation of these two Forms will reveal

other properties which will be more successful in capturing them. I will have to leave this

question for future research.

There are other Arabic verb Forms which might also encode the middle voice. Further

research is needed to determine whether these are characterised by always allowing affected

subjects as well, and whether the types of affectedness identified here might be relevant to


of 95

these Forms too. It should also seek to establish whether other properties are needed to

account for the Arabic middle Forms. Furthermore, the system of middle voice heads

introducing different types of requirements should be extended to the other middle Forms if

possible. Ultimately, it would be desirable if all the verb Forms in Arabic could be captured

by postulating different sorts of voice heads.

As is evidenced by the unboundedness requirement of the I-pattern, aspectual requirements

might be imposed on the middle Forms. A certain aspectual property might also arise as a

necessary consequence of the type of affected subject required. This is demonstrated by the

U-pattern, which must always allow stative and unbounded readings, as it must always be

compatible with lastingly influenced subjects. Other aspectual properties seem to be favoured

due to an association with the core readings of the Form in question. Thus, though it appears

to be possible for aspectual requirements to hold of middle Forms, the aspectual properties of

the middle mainly seem to be effects of the requirements related to affectedness.

Contrary to what is often claimed of the middle voice, it does not seem to place any

restrictions on valency in Arabic. Rather, the varying degrees of intransitivity of the middle

Forms seem to follow from the affectedness requirements imposed on them. This means that

the function of the middle in Arabic is not to reduce the verb’s valency, but that valency

reduction is instead often a by-product of different middle readings. I leave open the question

of whether the middle might serve to reduce valency in other languages. However, as this

does not seem to be the case in Arabic, it seems doubtful that valency reduction can be the

main function of the middle.

My analysis finds support for distinguishing between non-affected objects or ‘pseudo-


of 95

objects’, and affected objects or ‘real objects’. Real objects seem to be more strongly

transitive in Arabic, and are therefore rarer in the middle Forms. It thus seems that a more

complex notion of valency is useful to the description of the Arabic middle. Future research

should look into whether the middle makes a distinction between the two types of objects

cross-linguistically.

By showing how valency and aspectual properties seem to follow from the affectedness

properties of the Arabic middle Forms, I have in this paper sought to elucidate the interaction

between aspect, voice and valency. Furthermore, the extended notion of affectedness

proposed might contribute to our understanding of the middle voice in itself, and possibly

other phenomena which have been linked to affectedness as well.

Lastly, my aim has been to increase our knowledge of the properties of the Arabic verb

Forms. It is my hope that the requirements identified for the Forms here will be a first step

towards a better characterisation of them.


of 95

6 Bibliography

Ajer, Hanna Danbolt. 2014. “The Middle Voice in Arabic - Towards an Understanding of

Voice, Valency and Morphological Form.” BA, London: School of Oriental and

African Studies, University of London.

Alexiadou, Artemis, and Edit Doron. 2012. “The Syntactic Construction of Two Non-Active

Voices: Passive and Middle.” Journal of Linguistics.

Anderson, M. 1979. “Noun Phrase Structure.” Ph.D., University of Connecticut.

Badawi, El-Said M., M. G Carter, and Adrian Gully. 2004. Modern Written Arabic: A

Comprehensive Grammar. Routledge Comprehensive Grammars. London/New York:

Routledge.

http://search.ebscohost.com/login.aspx?direct=true&scope=site&db=nlebk&db=nlabk

&AN=602113.

Borer, Hagit. 2005. Structuring Sense, Vol. 2: The Normal Course of Events. Oxford

Linguistics. Oxford: Oxford University Press.

Carlson, Gregory N. 1977. “Reference to Kinds in English.” Ph.D. dissertation, University of

Massachusetts, Amherst.

Carnie, Andrew. 2013. Syntax: A Generative Introduction. 3rd ed. Introducing Linguistics 16.

Hoboken, New Jersey: Wiley-Blackwell.

Croft, William, Hava Bat-Zeev Shyldkrot, and Suzanne Kemmer. 1987. “Diachronic

Semantic Processes in the Middle Voice.” In Papers from the 7th International

Conference on Historical Linguistics, edited by Anna Giacalone Ramat, Onofrio

Carruba, and Giuliano Bernini, 179–92. Current Issues in Linguistic Theory 48. John

Benjamins Publishing.

Cuvalay-Haak, Martine. 1997. The Verb in Literary and Colloquial Arabic. Functional

Grammar 19. Berlin/New York: Walter de Gruyter.

Danks, Warwick. 2011. The Arabic Verb: Form and Meaning in the Vowel-Lengthening

Patterns. John Benjamins Publishing.

Dixon, R.M.W., and Alexandra Y. Aikhenvald. 2000. “Introduction.” In Changing Valency,

edited by R.M.W. Dixon and Alexandra Y. Aikhenvald, 1–29. Cambridge: Cambridge

University Press.

Fehri, Abdelkader Fassi. 2012. Key Features and Parameters in Arabic Grammar. Linguistik

Aktuell/Linguistics Today 182. Amsterdam/Philadelphia: John Benjamins Publishing

Company.

Holes, Clive. 2004. Modern Arabic: Structures, Functions, and Varieties. Georgetown

Classics in Arabic Language and Linguistics. Washington, D.C.: Georgetown

University Press.

Jaeggli, Osvaldo A. 1986. “Passive.” Linguistic Inquiry 17 (4): 587–622.

Kaufmann, Ingrid. 2007. “Middle Voice.” Lingua 117 (10): 1677–1714.

doi:10.1016/j.lingua.2006.10.001.

Kemmer, Suzanne. 1993. The Middle Voice. Typological Studies in Language 23.

Amsterdam/Philadelphia: John Benjamins Publishing Company.

Klaiman, M. H. 1991. Grammatical Voice. Vol. 59. Cambridge Studies in Linguistics.

Cambridge: Cambridge University Press.

Kratzer, Angelika. 1996. “Severing the External Argument from Its Verb.” In Phrase

Structure and the Lexicon, edited by J. Rooryck and Laurie Zaring, 109–37. Studies in

Natural Language and Linguistic Theory 33. Dordrecht: Kluwer Academic

Publishers.


of 95

Krifka, M. 1989. “Nominal Reference, Temporal Constitution and Quantification in Event

Semantics.” In Semantics and Contextual Expression, edited by Renate Bartsch, J.

van Benthem, and P. van Emde Boas. Groningen-Amsterdam Studies in Semantics

11. Dordrecht: Foris Publications.

———. 1992. “Thematic Relations as Links Between Nominal Reference and Temporal

Constitution.” In Lexical Matters, edited by Ivan A. Sag and Anna Szabolcsi, 1

edition, 29–83. Stanford, CA: Center for the Study of Language and Information

CSLI.

Pardee, Dennis. 1997. “Ugaritic.” In The Semitic Languages, edited by Robert Hetzron, 131–

44. Routledge Language Family Descriptions. London: Routledge.

Ramchand, Gillian. 1997. Aspect and Predication: The Semantics of Argument Structure.

Oxford: Clarendon Press.

———. 2008. Verb Meaning and the Lexicon: A First Phase Syntax. Cambridge Studies in

Linguistics 116. Cambridge: Cambridge University Press.

Reinhart, Tanya. 2002. “The Theta System – An Overview.” Theoretical Linguistics 28 (3):

229–90.

Roberts, Ian. 1987. The Representation of Implicit and Dethematized Subjects. Linguistic

Models 10. Dordrecht: Foris Publications.

Ryding, Karin C. 2005. A Reference Grammar of Modern Standard Arabic. Cambridge:

Cambridge University Press.

Smith, C. S. 1997. The Parameter of Aspect. 2nd ed. Studies in Linguistics and Philosophy

43. Dordrecht: Kluwer Academic Publishers.

Tenny, C. 1987. “Grammaticalizing Aspect and Affectedness.” Ph.D. dissertation, Oberlin

College.

Vendler, Zeno. 1957. “Verbs and Times.” The Philosophical Review 66 (2): 143–60.

doi:10.2307/2182371.

Wehr, Hans. 1979. A Dictionary of Modern Written Arabic: Arabic-English. Edited by J.

Milton Cowan. 4th ed. Wiesbaden: Harrassowitz.

http://archive.org/details/Dict_Wehr.pdf.

Wright, William. 1962. A Grammar of the Arabic Language. Edited by William Robertson

Smith and Michael Jan de Goeje. 3rd ed. Vol. 1. 2 vols. Cambridge: Cambridge

University Press.


of 95

7 Appendix A – Transliteration chart

The Arabic letters are listed alphabetically.

Arabic letter Transliteration Notes

ʾ My transliteration differs from that of Wehr in that I do not ء

omit ʾ at the beginning of words

a ا

b ب

t ت

t ث

j ج

h ح

k خ

d د

d ذ

r ر

z ز

s س

š ش

s ص

d ض

t ط

z ظ

ʕ The symbol ʕ is used instead of Wehr’s ʻ in order to avoid ع

confusion with ʾ

g غ

f ف

q ق

k ك

l ل

m م

n ن

h ه

w, u or u و

y, i or i ي


of 95

8 Appendix B – Abbreviations used in example glosses

3 = 3rd person M = masculine

ACC = accusative NOM = nominative

DEF = definite PASS = passive

F = feminine PF = perfect

GEN = genitive POSS = possessive

INDEF = indefinite SG = singular


of 95

9 Appendix C – Key to data sample

General notes: all verbs are listed alphabetically after root according to the Arabic

alphabet. The only exception is the “Form I – mixed” sheet, where

verbs displaying the same patterns are grouped together. Within each

group, the verbs are listed alphabetically.

see chapter 2.2-2.5 and chapter 3 for definitions of the aspectual

properties and argument roles appealed to. For the full details of what

the different labels used in the other columns mean and how they were

applied, see Ajer (2014:12-18,47-49). The “Semantic class”, “Voice”,

“Transitivity” and “Valency change” columns are not relevant to the

present study.

Columns

Root: gives the verb’s radicals.

Form: gives the number of the morphological Form of the verb.

Verb: gives the transliteration of the Arabic verb.

Meaning: gives the meaning of the verb in English.

Relation: shows how the meaning of the verb relates to that of other verbs from

the same root. If a verb is synonymous or almost synonymous with

another verb from the same root, this will always be listed. Apart from

this, each verb is generally compared to its counterpart. This means

that verbs of Forms II, IV, VII and VIII are all compared to Form I

verbs. Form VII verbs might be compared to Form IV verbs when this

is more suitable, whereas Form V verbs is compared to Form II verbs.

When a verb has a counterpart, its relation to it will always be given. If

it has no counterpart or an unrelated counterpart, I will list its relation

to other verbs when relevant. Form I verbs are only compared to other

Form I verbs from the same root, if there are any.

Semantic class: lists the applicable semantic labels I found useful when categorising

the verbs.

Aspect: lists which aspectual classes the verb’s readings belong to, and which

boundedness properties they have. Appears in all sheets, except for the

“All Forms” sheet.

Roles: lists the subject and object roles a verb assigns. Appears in all sheets,

except for the “All Forms” sheet.

Voice: lists whether a verb’s meaning is associated with the active, middle or

passive voice. It can be associated with more than one voice.


of 95

Transitivity: encodes the number of complements each verb has. If it can have

prepositional complements, the respective prepositions are given in

brackets.

Valency change: encodes the difference in valency between the verb and its counterpart.

If the verb has a counterpart, it is always compared to this. If not, it

will be compared to other verbs from the same root where relevant.

See “Relation” for more details.

Nr.: gives the value 1 for each verb, in order to make the number of verbs

easy to count when analysing.

Abbreviations and symbols used

Column it is used in Abbreviation or symbol Explanation

Meaning, Transitivity,

Aspect and Roles - Separates senses of a verb which have

different transitivity values in the “Meaning”

column, as well as their respective values in

the “Transitivity” column. Used to separate

aspectual class from boundedness properties

in the “Aspect” column, and subject roles

from object roles in the “Roles” column.

Relation, Voice,

Transitivity and

Valency change

/ When the verb in question has the same

relation to different verbs in terms of meaning

and valency change, the slash separates the

numbers indicating their Forms. It also

separates the different voices the verb is

associated with. When the same sense or

group of senses of a verb can have different

transitivity values or be used with different

prepositions, a slash is used to separate these.

Transitivity + When a verb takes more than one

prepositional complement, a plus sign is used

to separate them.

Relation and Valency

change

A space followed by

roman numerals

The numerals indicate the Form of the verb

which the verb in question is compared to.

Relation S Synonymous

Relation AS Almost synonymous

Relation R Related

Relation VR Vaguely related

Relation UR Unrelated

Semantic class Reflexive Reflexive middle

Semantic class Indirect Indirect middle

Semantic class Caus. Causative. Within brackets the type of event

or situation caused is indicated.

Semantic class Anticaus. Anticausative

Aspect s State

Aspect p Punctual event

Aspect g Gradual event


of 95

Aspect u Unbounded

Aspect b Bounded

Roles i Initiator

Roles l Lastingly affected

Roles d Directly affected

Roles m Mentally affected

Roles r Real object

Roles p Pseudo-object

Voice A Active voice

Voice M Middle voice

Voice P Passive voice

Voice P (HW) Used when the verb is indicated as passive by

Wehr.

Transitivity I Intransitive. The verb has no complements.

Transitivity T Transitive. The verb takes a direct object.

Transitivity 2T Doubly transitive. The verb takes two direct

objects.

Transitivity P Prepositional. The verb takes a prepositional

complement.

Transitivity 2P Doubly prepositional. The verb takes two

prepositional complements.

Transitivity TP Transitive and prepositional. The verb takes a

direct object and a prepositional complement.

Transitivity 2TP Doubly transitive and prepositional. The verb

takes two direct objects and a prepositional

complement.

Valency change S Same. The verb has the same valency as the

verb it is compared to.

Valency change H Higher. The verb takes one direct object more

than the verb it is compared to.

Valency change L Lower. The verb takes one direct object less


Valency change 2H Double increase. The verb takes two direct

objects more than the verb it is compared to.

Valency change 2L Double decrease. The verb takes two direct

objects less than the verb it is compared to.

Valency change SP Same with preposition. The verb has the same

valency as the verb it is compared to, but it

takes a prepositional complement more.

Valency change SmP Same minus preposition. The verb has the

same valency as the verb it is compared to,

but it takes a prepositional complement less.

Valency change S2P Same with two prepositions. The verb has the

same valency as the verb it is compared to,

but it takes two prepositional complements

more.

Valency change Sm2P Same minus two prepositions. The verb has

the same valency as the verb it is compared


of 95

to, but it takes two prepositional complements

less.

Valency change HP Higher from preposition. The verb takes a

direct object where the verb it is compared to

takes a prepositional complement.

Valency change LP Lower to preposition. The verb takes a

prepositional complement where the verb it is

compared to takes a direct object.

Valency change HaP Higher and preposition. The verb takes one

object and one prepositional complement

more than the verb it is compared to.

Valency change LaP Lower and preposition. The verb takes one

object and one prepositional complement less


Valency change H2P Higher from two prepositions. The verb takes

one direct object where the verb it is

compared to takes two prepositional

complements.

Valency change L2P Lower to two prepositions. The verb takes

two prepositions where the verb it is

compared to takes one direct object.

Valency change 2LP Double decrease to preposition. The verb

takes one prepositional complement where the

verb it is compared to takes two direct

objects.


of 95

10 Appendix D – Data sample

The data sample can be found by following this link:

https://www.dropbox.com/s/gipduncjjijv70r/Appendix%20D%20for%20Hanna%20Danbolt

%20Ajer%27s%20thesis%20-%20Data%20sample.xlsx?dl=0

https://www.dropbox.com/s/gipduncjjijv70r/Appendix%20D%20for%20Hanna%20Danbolt%20Ajer%27s%20thesis%20-%20Data%20sample.xlsx?dl=0

https://www.dropbox.com/s/gipduncjjijv70r/Appendix%20D%20for%20Hanna%20Danbolt%20Ajer%27s%20thesis%20-%20Data%20sample.xlsx?dl=0


of 95

11 Appendix E – Questionnaire

11.1 Procedure

The questionnaire, consent form and procedure were all based on those of Danks (2011).

Two participants took part in the study. Both were male and had undertaken their whole

education or part of their education in Modern Standard Arabic, and had also taught it at

university. They were native speakers of different Arabic colloquials.

The participants were sent the consent form, the background information form and the

questionnaire, which can be found in 11.2, 11.3 and 11.4 respectively. They were encouraged

to let me know if they had any questions or comments. After completing the forms and the

questionnaires, they sent them back to me, and one participant was asked some follow-up

questions regarding one of the judgments. In accordance with the information given in the

consent form, the background information form and the original questionnaires have been

destroyed.

The questionnaire given in 11.4 contains a summary of the acceptability judgments, with ‘1’

indicating that the sentence is acceptable, ‘2’ that it is doubtful and ‘3’ that it is unacceptable.

It does not contain any of the comments made by the participants. Any sentence which

seemed to be rejected for other reasons than the fact that the verb could not occur in the

construction in question, have been discarded from the analysis, and marked with an ‘X’.

Some of the sentences were fillers, which were included so that it would not be too obvious

which constructions I wanted to test.


of 95

11.2 Consent form

Information and consent form for the questionnaire on Arabic verb forms

I am currently conducting research related to the verb forms of Modern Standard Arabic for

my MPhil thesis. In order to assess the contemporary usage of some of the verbs, I need

native Arabic speakers with a high level of proficiency in Modern Standard Arabic to judge

the acceptability of the 40 example sentences in the questionnaire.

Please put a number from 1 to 3 next to every example to indicate its acceptability. 1 means

that the example is completely acceptable, 2 that the example might not be fully acceptable or

might sound a bit strange, and 3 that the example is unacceptable. If you have any comments

as to why you find certain sentences to be less acceptable, please write them next to each

example. I am especially interested in which constructions the different verbs can and can’t

be used in.

The questionnaire will help me determine which usages are acceptable for the different verbs.

It is not a test of your personal intelligence or ability.

It would also be useful to me if you could fill in a background information form about your

age, gender, which Arabic colloquial you are a native speaker of, and your educational level

in Modern Standard Arabic.

Your responses to the questionnaire and background information form will only be seen by

me and my supervisor. None of the responses you provide me with will be identifiable by

name in any research papers and presentations that result from this work. Your background

information forms and the original questionnaires will be destroyed once I have finished my


of 95

thesis research.

Your participation is voluntary, and you may refuse to answer certain questions on the

questionnaire or the background information form. You can withdraw from the study at any

time with no penalty. This does not waive your legal rights. Please keep a copy of this form

for your own record. If you have further questions related to this research, please email me at

[email protected]. You can also contact me via this email address after 01.07.2015 if you are

interested in finding out about the results of the research.

Please sign below if you are willing to take part in the research, and return the signed form to

me via email if possible. If you prefer not to submit via email, please contact me to agree on

an alternative arrangement.

Thank you for your participation!

Hanna Danbolt Ajer,

Department of Theoretical and Applied Linguistics, University of Cambridge

Your signature below indicates that you are over 18 years of age and have understood the

information about this questionnaire and consent to your participation.

Signed:

Name:

Date:

mailto:[email protected]


of 95

11.3 Background information form

1. Please state your gender:

2. Please underline the age range you are in:

Under 25 25-50 Over 50

3. Please state where the Arabic colloquial you are a native speaker of is from (i.e.

Lebanon, Morocco, etc.):

4. Please state your educational level in Modern Standard Arabic, or the professional

background in which you use or have used it (i.e. completed your primary education

in it, studied it at university, have taught it at university etc.):


of 95

11.4 Questionnaire

Questionnaire on Arabic verb forms

1 = acceptable 2 = doubtful 3 = unacceptable

Comment 1/2/3

جل 1 1 1 أدب الر

ب المطر التربة 1 1 خص طوال ساعة

2

3 سمر الحقل في ساعة 1 1

X انسرق الكتاب بعد الدرس

4

جل طوال 1 1 أدب الر سنة

5

X جل طوال ح الر ترو ساعة

6

جل بسبب 1 2 ثري الر الحرب

7

جل في سنة 1 2 8 ثري الر

جل األكل 1 2 ابتلع الرظهرا ١٢الساعة بط بالض

9

انساحت دماء من 1 1 الجرح

10

X انسرق الكتاب في دقيقتين

11

جل 1 3 12 أفقرت الر

X جل في ح الر ترو ساعة

13


of 95

جل األكل 1 1 14 ابتلع الر

X جل ح الر 15 ترو

X الساعة انسرق الكتاببط ١٢ ظهرا بالض

16

17 ابتلع األكل 1 1

X جل على استام الر الخبز طوال ساعة

18

ب المطر التربة 1 3 خص في ساعة

19

X جل الساعة ح الر تروبط ١٢ ظهرا بالض

20

انساحت دماء من 1 1 الجرح طوال ساعة

21

سمر الحقل كامله في 1 1 ساعة

22

جل األكل 1 1 ابتلع الر طوال دقيقتين

23

سه 1 1 24 تجنب الولد مدر

جل يأدب 2 2 25 كان الر

سمر الحقل طوال 1 2 ساعة

26

X الكتاب انسرق 27

رقة ممنوعة 1 1 28 الس

X جل على استام الر الخبز في ساعة

29

X جل على استام الر ١٢الخبز الساعة بط ظهرا بالض

30

س 2 3 31 تجنب المدر


of 95

32 سمر الحقل 1 1

جل أفقرت األزمة 1 1 الرظهرا ١٢الساعة بط بالض

33

جل طوال 1 1 ثري الر سنة

34

X جل على استام الر الخبز

35

X انسرق الكتاب طوال دقيقتين

36

جل 1 1 37 ثري الر

سمر الحقل بسبب 1 1 الشمس القوية

38

جل الساعة 1 2 أدب الربط ١٢ ظهرا بالض

39

جل بعد 1 1 ثري الر الحرب

40

Aspects of the Arabic middle – an exploration of voice, aspect and valency

Documents

Transcript of Aspects of the Arabic middle – an exploration of voice, aspect and valency