When does one help man?
The distribution and meaning of dedicated impersonal pronouns
in Germanic languages
*∗rMA thesis Linguistics, University of Amsterdam
Student: Paula Fenger
Student number: 6066275
Supervisors: Suzanne Aalberse (University of Amsterdam); Ad Neeleman (University College London)
Second reader: Jan Don (University of Amsterdam)
*
∗ Next to my supervisors, I would like to thank the participants that have helped me with the questionnaire
and provided me with data from their language: Hans Broekhuis, Riny Huybregts and Fred Weerman for the Dutch data; Ger de Haan, Eric Hoekstra, Arjen Versloot and Willem Visser for the Frisian data; Klaus Abels. Eva Csipak and Roland Pfau for German; and Laura Bailey and Vikki Janke for providing me with English data.
ABSTRACT
Dedicated impersonal pronouns are found in many languages and are used to express generic and arbitrary statements. A semantic and syntactic asymmetry exists between two types of dedicated impersonal pronouns (a. o. Malamud 2004; Cabredo-Hoffher 2008, 2010; Siewierska 2008). The English type, one, on the one hand, can only express generic statements, but can occur in both subject and object position. The German type, man, on the other hand, can both have a generic and arbitrary reading, but can only occur in subject position. There is a difference of feature specification (following Ackema and Neeleman 2014, in prep.) between the German and the English type, which derives the semantic asymmetry. This study focuses on these two types of dedicated impersonal pronouns in Germanic languages in order to find why the German type is only restricted to subject position. It will be shown that this type of pronoun is restricted to occur only with nominative case, rather than it is restricted to its underlying syntactic position (as is claimed by Cinque 1988 and Egerland 2003). This follows from a proposal, following Neeleman and Weerman (1999), that nominative case is different from all other cases. This idea entails that the German pronoun type can only occur if it is in an agreement relation with the verb, because the pronoun itself is underspecified for any functional structure. The English pronoun type, on the other hand, is specified enough by itself and therefore does not need to stand in agreement relation with the verb and can thus occur with accusative case. This proposal for agreement points toward an unification approach (a.o. Gazdar et al. 1985, Shieber 1986, and Pollard and Sag 1994) rather than an asymmetry approach (Chomsky 2000, 2001) for agreement.
KEYWORDS: dedicated impersonal pronouns; Germanic; generic and arbitrary readings;
§1INTRODUCTION
Dedicated impersonal pronouns are found in many languages and are used to express generic and arbitrary statements. An example is given in (1), for the German pronoun man and for the English dedicated impersonal pronoun one.
(1) a. Wenn man in Italien ist, isst man Nudeln. [German]
b. When one is Italy, one eats pasta [English]
The sentence in (1) provides a generic statement, thus ‘people, including you and me have the habit of eating pasta while we are in Italy’. It is not the case however, that both man and one are exactly the same. With regard to their meaning, the English type can only receive a generic reading, whereas it is possible for the German type to also refer to ‘someone’, and can thus receive an arbitrary reading (a. o. Malamud 2004; Cabredo-Hoffher 2008, 2010; Siewierska 2008):
(2) a. Man hat dich angerufen, aber ich weiss nicht, worum es ging. [German] b. *One has called for you, but I don’t know what it was about [English]
In this respect, one is more restricted than man. However, with regard to the syntactic positions, the reverse pattern holds: only man can occur in subject position, whereas one is also grammatical in object position, this is shown in (3).
(3) a. *Das Pensionsalter ist im allgemeinen 65. Aber die Regierung zwingt man jetzt dazu, sich erst mit 67 pensionieren zu lassen.
[German]
one to retire at the age of 67
It is not the case that German does not have an impersonal pronoun in object position; it cannot have the man form. It can however, have the one form in object position. Thus the sentence in (3)a. would be grammatical if einen is used (Zifonun 2000). In a way, the one (i.e. einen) form complements man, since the latter cannot occur in object position, whereas the former can.
This difference between the possible readings on the one hand and the syntactic positions on the other hand will be the main focus of this paper. First of all, it will be shown in more detail what the exact similarities and differences of these two types of pronouns are. (section 2.1) After that, (in section 2.2) the discussion will focus on three explanations for the difference in meaning and syntactic usage of the dedicated impersonal pronouns (Egerland 2003; Hoekstra 2010; Ackema and Neeleman 2014, in prep.). It will be shown that these three theories derive the difference between the English type on the one hand and the German type on the other hand by assuming different feature specifications. It is generally assumed that the German type does not have any features in its functional structure, whereas there is at least some featural content present in the English pronoun type. These two different feature specifications also derive the semantic difference, following the proposal by Ackema and Neeleman (2014). Although Egerland (2003) claims that this difference in feature specification can explain why man cannot occur in object position, whereas one can, it will be shown that there are some problems with this view both theoretically and empirically (section 2.2). Therefore, one of the aims of this paper is to find out whether man is only ungrammatical in object position, or whether its syntactic distribution is even more restricted.
In order to clarify some of the empirical questions as to the nature of impersonal pronouns and what restricts their distribution, a questionnaire is developed (which will be discussed in section 3). Four Germanic languages are tested, namely Dutch, English, Frisian
and German. The main finding is that the German type of dedicated pronoun is restricted to occur only with nominative case, whereas the English type can also occur with other cases. Thus it is not the case that the underlying syntactic position restricts the occurrence of the man pronoun type. This follows from a proposal, following Neeleman and Weerman (1999), that nominative case is different from all other cases. This idea entails that the German pronoun type can only occur if it is in an agreement relation with the verb, because the pronoun itself is underspecified for any functional structure. The English pronoun type, on the other hand, is specified enough by itself and therefore does not need to stand in agreement relation with the verb and can thus occur with accusative case. The details of this proposal will be discussed in section 4. This idea has not only consequences for the nature of impersonal pronouns, but it also has consequences for how agreement works, which will be discussed in section 5.
§2EMPIRICAL AND THEORETICAL BACKGROUND
In this section some background will be given on impersonal pronouns. In the literature, the terminology varies to a large extent. In this section, a uniform terminology will be used. If other proposals deviate from the terminology used here, such will be pointed out. It will become clear that a division can be made in two types of dedicated impersonal pronouns, which will be called imp-1 and imp-2 throughout the remainder of this paper (terminology based on Ackema and Neeleman 2014, in prep.) A list of imp-1 and imp-2 pronouns in various languages is given in table 1.
imp-1 imp-2
one English men Dutch (Weerman 2006; Ackema and
Neeleman 2014) men Frisian (Hoekstra 2010) man German (Zifonun 2000) maður Icelandic (Sigurdsson and man Swedish (Egerland 2003)
Egerland 2009) on French (Coveney 2000)
si Italian (Cinque 1988)
Table 1. Overview of dedicated impersonal pronouns
As was already briefly shown in the introduction and will be elaborated on in section 2.1, the division between imp-1 and imp-2 pronouns is based on their different syntactic distribution and meaning. Imp-1 pronouns have a freer syntactic distribution and can occur in both subject and object position, whereas imp-2 pronouns can only occur in subject position. With regard to their meaning, imp-1 pronouns are more restricted and can only have a generic reading; imp-2 pronouns can both receive a generic and an arbitrary reading. In section 2.1 an overview of the properties of imp-1 and imp-2 will be given. Section 2.2 focuses on the existing theories for impersonal pronouns. The theories will be discussed and evaluated with the help of the properties discussed in section 2.1. Section 2.3 provides a short summary and the goal of this study will be explained.
§2.1 The properties of dedicated impersonal pronouns
This section discusses the most important properties of impersonal pronouns. The focus will be on the syntactic properties and the possible readings. It will be shown that on the basis of the described properties, a division can be made into two types of dedicated impersonal pronouns, imp-1 and imp-2, as is shown in table 1.. Although there are many languages that display either
imp-1 or imp-2, the discussion will focus on one representative language for imp-1, namely English, and for imp-2 the properties are discussed with respect to somewhat more languages since some properties aren’t visible in all languages. All properties discussed in this section need to be explained or derived by a theory of impersonal pronouns, which will be discussed in the next section.
There are several syntactic properties that are similar for imp-1 and imp-2. Both types of impersonal pronouns have verbal singular agreement (a.o. Egerland 2003; Siewierska 2008a); this is shown in (4) for Dutch and English.
(4) Dutch: In dit land geef-t3sg men elkaar cadeautjes met de kerst.
English: In this country, one give-s3sg each other presents during Christmas.
If there is verbal agreement in a language, it is always 3sg. Sometimes, however, it is possible to have plural agreement on the adjective or participle in some languages. The plural agreement is grammatical when the reading of the pronoun is plural (Egerland 2003). In example (5) a Swedish example is given.
(5) Om man inte är gift / gifta måste man ha skilda rum på detta hotel If IMP-2is not married.SG / married.pl must IMP-2have separate rooms in this hotel
‘If one is not married, one should have separate rooms in this hotel’
As can be seen in the example, the pronoun needs to refer to more than one person; therefore plural agreement is also possible. This type of agreement is possible in several languages, such as Swedish, French and Italian. It is not visible, however in other languages, such as Dutch, because there is no number agreement on the participle.
Although the pronoun seems to behave –most of the time - syntactically singular, the semantics of both imp-1 and imp-2 is plural. This is shown in the examples above, where in (4), the reciprocal (elkaar or each other) is plural and in (5) plural agreement on the adjective is only possible if imp-2 has a plural interpretation and refers to more than one individual.
With regard to the possible syntactic positions, the following has been observed. Imp-1 can both occur in subject and object position, whereas imp-2 can only occur in subject position. If a language has an imp-2 pronoun, it can, but may not, have a suppletive form in object postion, and that is always the imp-1 type (Egerland 2003; Hoekstra 2010; Weerman 2006; Zifonun 2000).
(6) English: The retirement age is typically 65. But now the government is obliging one-OBJ to retire at the age of 67.
German: Das Pensionsalter ist im allgemeinen 65. Aber die Regierung zwingt *man-OBJ / einen-OBJ jetzt dazu, sich erst mit 67 pensionieren zu lassen.
Next to the different syntactic distributions of both pronouns, there is also a difference between imp-1 and imp-2 in how many interpretations they allow. Imp-2 seems to be more flexible in this respect, as it allows both generic and arbitrary readings, as is shown for Dutch in the examples in (7) and (8). A generic meaning can roughly be translated as ‘people (including you and me)’; an arbitrary reading can be translated as ‘someone’ or ‘they’.1 Imp-1 on the other hand only allows generic readings, shown in the English examples (Ackema and Neeleman 2014; Malamud 2012; Siewierska 2008b).
1
Cabredo-Hofherr (2003) points out that there are more impersonal readings, at least for 3pl pronouns. This thesis will focus only on what she calls the ‘universal reading’, which is the generic reading; and ‘existential readings’, the ‘arbitrary; reading in this thesis. What she calls the ‘coporate reading’ will not be taken into account. See for a discussion on this reading and its impersonal nature Ackema & Neeleman 2014).
(7) Dutch: Als men in Italie is, eet men spaghetti [generic] English: When one is in Italy, one eats spaghetti
(intended reading: When people are in Italy, they have the habit of eating spaghetti)
(8) Dutch: kijk, men heeft hier bomen geplant. [arbitrary] English: *Oh look, one has planted trees here.
(intended reading: ‘Oh look, they have planted trees here.’)
Siewierska (2012) and Giacalone Ramat & Sansò (2007) show that there seems to be some kind of grammaticalization cline with regard to the different meanings. The more a pronoun has an arbitrary meaning, the more grammaticalized it becomes. Giacalone Ramat & Sansò (2007) also observe that the more grammaticalized the pronoun becomes, it becomes more and more a weak pronoun in the sense of Cardinalleti and Starke (1999), thus it becomes a clitic. It has also been observed by others that imp-2 pronouns seem to have the properties of weak pronouns or clitics, whereas imp-1 pronouns do not (D’allesandro & Alexiadou 2006; Weerman 2006).
From the discussion in this section, it can be concluded that imp-1 only allows one type of reading, whereas it can occur in more syntactic positions; imp-2 on the other hand allows more readings, but the syntactic distribution is limited. There is one other observation that can be made with regard to the syntactic distribution and the possible readings. If imp-2 is in subject position of a passive sentence, the only possible reading is a generic one (Cinque 1988, Zifonun 2000).
(9) Gister middag werd men ontslagen Yesterday afternoon was IMP fired
‘Yesterday afternoon somebody was fired’
The impersonal pronoun can only refer to ‘people’ and not to ‘someone’. In a passive, the subject is underlying an object. Therefore, it has been concluded that the underlying object can only receive a generic reading.
On the basis of the properties discussed in this section, a division can be made between two types of dedicated pronouns, namely imp-1 (an example is the English one) and imp-2 (the German man). The properties are summarized in table 2.
properties imp-1 imp-2
(i) verbal agreement 3sg 3sg
(ii) other agreement
?
pl (if generic)
(iii) subject position √ √
(iv) derived subject position √ √ (only generic)
(v) object position √ *
(vi) generic reading √ √
(vii) arbitrary reading * √
Table 2. properties of imp-1 and imp-2
As can be observed, there are several similarities and differences between imp-1 and imp-2. Both have 3sg verbal agreement, (i), and both allow generic readings, (vi), and can occur in subject positions (iii, iv). However, only imp-1 can occur in object position, (v), and only imp-2 also allows arbitrary readings, (vii). Thus the following general pattern seems to hold: if more syntactic positions are possible, fewer readings are available. Also, the reverse pattern seems to hold for imp-2 pronouns: this type allows more readings, but is syntactically more restricted.
Although observations have been made about the syntactic -surface and underlying- object position (iv, v), more research is needed. As will become clear in section 2.3, the observations iii - vii will be the main focus of this thesis. Moreover, it seems to be possible to have arbitrary readings if the pronoun is a derived subject, thus it seems that observation iv is not completely accurate. Before we turn to this observation, the existing theories will be discussed and it will be shown which of the properties in table 2. can be derived by different theories and which properties are problematic.
§2.2 The existing theories
In the previous section the known and most important properties of impersonal pronouns are discussed. A theory of impersonal pronouns should be able to derive all these properties. In this section, the most prominent existing theories will be discussed. It will be shown how they analyze the impersonal pronouns and where the problems arise. First, the analysis of Egerland (2003) who mainly looks at French, Italian and Swedish, will be discussed; then the approach of Hoekstra (2010) for Frisian, which is a somewhat similar approach to Egerlands. Finally, the explanation given by Ackema and Neeleman (2014, in prep.) for the two types of dedicated impersonal pronouns in general will be discussed.
Egerland (2003) discusses data from a cross-linguistic study, focusing on French, Italian, Swedish and to some extent Icelandic and Danish. The following observations that have been discussed in the previous section are the core properties that he wants to derive in his analysis (the romance numbers correspond to the numbers in table 2):
(10) (vi, vii) man (imp-2) allows both arbitrary and generic readings;
(ii) imp-2 can have different agreement patterns: it is mostly singular, but if the reading is plural, plural agreement can occur
(iv) Arbitrary readings of imp-2 are only possible if the pronoun is underlying a subject (the observation of Cinque 1988);
(v) imp-2 can never occur in surface-object position, whereas imp-1 can.
To account for the observations in (10), he argues that imp-2 and imp-1 have different features: the pronoun man or imp-2 is underspecified and doesn’t have any phi-features(11)a. One or imp-1 does have a phi-node (11)b, because historically it is derived from a numeral. This means that Egerland predicts that all imp-1 pronouns are derived from numerals, but this is not necessarily true, as can be seen in table 1: Frisian and Icelandic both have imp-1 pronouns that historically are derived from the word for ‘human’.
(11) a. [imp-1] φ ] b. [imp-2]
In his framework, there are two types of phi-features: syntactic features that need to be checked on LF and PF; and lexical features that are on the lexical item itself. These two types of features need to be matched after the syntactic derivation (Chomsky 1995; Marantz 1993, 1997). Thus the structures in (11) refer to the lexical features of the pronouns. Imp-2 does not have any lexical phi-features, whereas imp-1 has. The only -semantic- feature that imp-2 might have is [+animate], because the pronoun always needs to refer to humans. As will be shown below, this difference in feature make-up is used in somewhat similar ways in other proposals. In a sense, Egerland argues, imp-1 pronouns and pronouns that can be personal and impersonal, such as you, are the same: they both have a lexical phi-feature and it is predicted that their syntactic behavior is the same.
According to Egerland, property (10)ii, illustrated with a sentence in (12), now follows easily.
(12) Quand on est belles,…
When IMP is.3SG beautiful.FEM.PLUR
‘When one is beautiful,…
The plural adjectival agreement and the singular verbal agreement are different in nature. The plural adjectival agreement is argued to be semantic agreement: the subjectis pragmatically understood as plural and the morphology, seeing this semantic (and not the lexical!) feature, agrees with it. However, at the point where the adjective agrees, the subject is not inserted yet; this happens only after the agreement has taken place. The subject is inserted in the syntax to get into a local agreement relation with the verb, hence the subject-verb agreement relation is syntactic in nature. Because of the lack of features on the subject, the verbal agreement will be the default, 3sg agreement. The subject on in French is compatible with both the singular verbal agreement and the plural adjective agreement, because it isn’t specified for any lexical features at all. Therefore, the derivation will not crash when the syntactic features and the lexical features, which are absent on the imp-2 pronoun, are matched after the syntax. 2
The difference in the lexical feature make up can also explain the fact that imp-2 is flexible between generic and arbitrary readings, as is shown in (10)vi-vii. In the literature, it has been assumed that a generic reading is licensed via a generic operator (a.o. Krifka et al. 1995; Chierchia 1995). Egerland now argues that an arbitrary reading is not derived via another operator, but via the absence of any features. Thus ‘Arbitrary-ness’ has in fact no status in the syntax and arises because of the absence of features (Egerland 2003). Since imp-2 has no lexical features, it can receive an arbitrary reading. It depends solely on the discourse whether
2 This type of semantic agreement is in a somewhat similar way also proposed by Wechsler & Hahm
(2011). They propose that the pronoun vous in French can have flexible adjective agreement on the adjective, since the pronoun itself lacks syntactic features.
the subject is understood as one person or a group. Imp-1 pronouns on the other hand cannot receive an arbitrary reading, because it always has –at least– one feature: phi. The same holds for you, because it already has a phi feature and can therefore only have a generic reading.
Egerland argues that this lack of features does not only explain the flexible agreement and different type of readings, it also is possible to derive the fact that a imp-2 pronoun can never have an arbitrary reading when it is the subject of an ergative or passive. The surface subject of both types of sentences is argued to be an underlying object (Burzio 1986; Jaeggli 1986). An object, being an internal argument of the VP, is closer to the verb than the subject, being an external argument. The relation between the internal argument and the verb needs to be evaluated at LF, to see what the exact role is of the object. This can only happen if the object has some feature content. Imp-2 pronouns do not have any features and can therefore never be evaluated; hence they can never have an arbitrary reading in passives and ergative sentences. The fact that this type of pronouns does occur as a derived subject with a generic reading is due to the generic operator that can serve as the content that can evaluate the role of the object. This same explanation also derives the fact, according to Egerland, that the class of imp-2 pronouns can never occur as a surface object: they lack feature content and therefore the object-predicate relation cannot be evaluated. The fact that imp-1 pronouns and you do occur as object is because of the lexical phi-feature.
In the above section, Egerland ideas about the difference impersonal pronouns have been explained. He makes two main claims: 1) imp-1 and imp-2 have a different feature make-up, only imp-1 has a phi bundle, whereas imp-2 does not; 2) generic readings are derived via an operator, arbitrary readings via the absence of an operator. These two claims can derive some of the features in table 2 correctly. First, vi and vii are derived because arbitrary readings can only occur with imp-2 pronouns, since that is the only pronoun that does not have any features. The absence of features also explains observation iv: only generic readings can occur if the
pronoun is a derived subject, because a generic operator licenses the relation between object and the predicate and the absence of any functional structure means that the object-verb relation cannot be evaluated.
There are, however, in my view, some problems that will be explained presently. First of all, it is not clear to me what his view on grammar exactly is. I believe he assumes a late-insertion model (Ackema and Don 1992; Halle and Marantz 1993). Thus phonological material only is inserted after syntax. It is not clear to me, however, how he then can account of his type of semantic agreement, which occurs on the adjective or the participle. At the point where morphology ‘picks up’ the plural semantic feature, the subject is, according to Egerland, not yet inserted. This only happens when the verbal agreement takes place. However, in my view, Egerland does not mean that the –lexical– subject is inserted post-syntactically, because the morphology needs to be able to see the semantic content of the subject and it needs to get into a local relation with the verb, thus a syntactic subject needs to be present during the whole derivation.
Although the different feature specifications for imp-1 and imp-2 seem to explain the different behavior, Egerlands explanation for the similarities between imp-1 and you are a bit odd. Both imp-1, which is almost always the numeral for one, and you have lexical phi-features that explain their behavior w.r.t. the syntactic positions and readings; these lexical phi-features should refer to [singular], and in the case of 2sg also to [addressee]. Maybe that the generic one has, diachronically, lost his number feature and has a unspecified number and/or person feature, leading to two synchronic lexical items one: a number and an impersonal pronoun. The impersonal use of the 2sg pronoun could have two explanations, neither are explicitly proposed by Egerland. First of all, you could also have two instances: one that is used personal, and has [addressee] and one that has lost this feature and has only an empty phi node. However, if it is possible to have two synchronic items you, Egerland would also predict that in principle every
personal pronoun could have an impersonal counterpart, contrary to fact (Ackema and Neeleman 2014 show that 1sg and 3sg can never be used impersonally). Secondly, Egerland could propose that there is only one lexical item for 2sg and the pragmatics can ‘erase’ the [addressee] feature. Again, if this would be possible, it remains unclear why not all personal pronouns could have person features erased via pragmatics, and what this operation would look like. Moreover, it remains unclear in Egerlands proposal why it is common for languages to use one or you as an impersonal pronoun with plural readings, while they both have a clear singular reading in their number and personal use. Although in English you is number neutral, since it occurs in both 2sg and 2pl, it is not the case that all pronouns for the second person are number neutral. For example, the Dutch form only is used in 2sg and is clearly singular. Moreover, as already pointed out in the discussion of Egerlands proposal, imp-1 does not always have the form of a numeral: Frisian men (syncretic to Dutch men) and Icelandic maður both have only generic readings, but stem from the word for ‘man’, as can also be seen in table 1. This means that it is not always clear why imp-1 pronouns need to have a phi-bundle. Moreover, if it is in some cases historically true, such as one in English, it does not mean that this is still relevant in the synchronic grammar and during the acquisition of the pronoun. It is not clear why a child would assume a phi-bundle on top of a dedicated impersonal pronoun, only because the pronoun is historically derived from a numeral.
Finally, it seems problematic to account for both the Cinque-generalization (iv in table 2) and the observation that imp-2 pronouns cannot occur in object position (v in table 2). Egerland argues that imp-2 pronouns can only have a generic and not an arbitrary reading when they are the subject of a passive or ergative, because the generic operator licenses the object-predicate evaluation. If this is true, we would expect that Egerland predicts that generic readings are also possible in sentences where the pronoun surfaces as object: the generic operator, again can license the evaluation. Also, if the interpretation is dependent on the discourse and not on the
generic operator, the same problem would occur. It cannot be the case that the pronoun cannot occur as an internal argument, but can end up as a surface subject. When the derivation starts, the internal argument cannot ‘know’ if it will end up as a surface subject or not. Thus, it seems problematic that the movement of the pronoun with a generic operator makes the pronoun grammatical, whereas the absence of the movement leads to an ungrammatical result.
Summarizing, Egerland tries to account for the different behavior of imp-1 and imp-2 by arguing that imp-1 has lexical feature content, whilst imp-2 lacks features. Egerland claims that the lack of features with imp-2 pronouns can derive the flexibility of generic and arbitrary readings. This lack of features also derives the fact that imp-2 cannot occur as a surface object. However, several problems have been pointed out with regard to the feature specification, because there is no 1-to-1 mapping from imp-1 being derived from a numeral and imp-2 from the word for ‘human’. Moreover, it is not clear why you can be both personal and impersonal, whereas I cannot. Secondly, a problem has been pointed out for the analysis of the syntactic distribution of imp-2. It is not clear why that pronoun can occur as a derived subject but not as a surface object, while in both cases the pronouns starts as the underlying object.
In many ways, Hoekstra (2010) follows the analysis of Egerland (2003), but he mainly focuses on Frisian data. He observes that the Frisian men always receives a generic inclusive reading, (13), which can sometimes even be specific to 1sg,(13)c, whereas, as was already shown in the previous section, the Dutch men and the German man are not exclusively generic inclusive.
(13) a. Wannear’t men yn Italië is, yt men pasta [generic inclusive] When IMP in Italy is, eat IMP pasta
‘When one is in Italy, one eats pasta’
IMP has for you called, but I know not what it about went
‘They have called for you, but I don’t what it was about’
c. Men wit bytiden net mear, hoe’t men moat. [personal] One knows sometimes not anymore, how.that one must
‘Sometimes I don’t know what to do anymore.’ (Hoekstra 2010, p.34)
Although Hoekstra doesn’t mention it, the Frisian men seems to pattern with English one with regard to the possible readings, thus it can be classified as an imp-1 pronoun. Moreover, the Frisian impersonal pronoun can occur in object position, though a suppletion form, jin, is needed (14). Here, Frisian seems to pattern with German, because the suppletive accusative form of ein is used instead of man, see example (6).
(14) De pensjoenâldens is 65. Mar no ferplichtet de oerheid jin om mei pensjoen te The retirementage is 65. But now forces the government IMP to with retirement to gean mei it 67e libbensjier
go with the 67e life year
‘The retirement age is typically 65. But now the government is obliging one to retire at the age of 67’
From these facts, Hoekstra reasons as follows. The Frisian impersonal pronouns are obligatorily inclusive, contra the Dutch and German forms, and therefore need a different feature make up. He follows Egerland in saying that imp-2 is only specified for the feature [animate] (15), whereas imp-1 is specified for more features. He uses the feature specification of Harley and Ritter (2002), which is a feature geometry that uses privative features. This means that if a pronoun semantically doesn’t express a feature, it is absent from the geometry, rather than defined
negatively (which would be the case in a binary feature system). To account for the difference between imp-1, (16)a, and 3pl they, (16)b, which always excludes speaker and addressee, Hoekstra proposes the following geometries:
(15) optionally inclusive man : Animate (=Human)
(16) a. obligatory inclusive men (Frisian) b. obligatory exclusive they
Referring Expression Referring Expression
PARTICIIPANT INDIVIDUATION INDIVIDUATION
Speaker Addressee Group CLASS Group CLASS
Animate Animate
Both imp-1 and they refer to humans and to plural entities, thus the features [group] and [animate] are needed. The difference between imp-1 and they is the specification of the person node: since imp-1 always refers to the speaker and addressee, the features are included in the geometry. These person features are absent in the geometry of they, which is expressed by the fact that there is no participant node in the structure present. Observe that the geometry for imp-1 pronouns is exactly the same as for the personal first person plural pronoun, such as we.
These three feature specifications are able to derive the different patterns in Frisian and Dutch. With regard to the ungrammaticality of imp-2 in object position, Hoekstra follows Egerland. The fact that in Frisian both the impersonal pronoun men can occur as a personal pronoun and personal pronouns such as they can occur as impersonal pronoun, follow from the feature geometries. The features are exactly the same for impersonal and personal use, and the only difference is if they are used, in Hoekstra’s terms, specific or not. With this analysis,
Hoekstra predicts that some impersonal pronouns seems to receive a more personal meaning over time, such as Frisian men, in example (13)c and French on, which refers to 1pl.
Within this approach, it is expected that impersonal pronouns can be used personal and vice versa, since the feature geometries are the same for both types. However, it seems insufficient to derive the difference between impersonal and personal pronouns by saying that this is due to specificity. The fact that the geometry of men is exactly the same as 1pl pronouns, doesn’t explain the subtle differences between both pronouns. Moreover, ‘specificity’ only describes the difference between personal and impersonal uses, but doesn’t derive it.
A second problem with regard to the feature geometry is the fact that the feature [group] needs to be included. This is necessary to explain the fact that the pronoun refers to a plurality of people. However, this would also predict plural agreement on the verb, but this is not the case as was shown in section 2.1: in almost every language 3sg agreement is used. However, Hoekstra (2010) does point out that it is predicted within his approach that plural agreement on the adjective can occur (point ii in table 2).
Finally, all the problems that arise in Egerland (2003) with regard to the ungrammaticality of imp-2 in object position also hold for Hoekstra’s analysis.
Recently, Ackema and Neeleman (A&N, 2014, in prep.), give an account for the difference between imp-1 and imp-2. Their starting point for the analysis is the subtle differences between imp-1 and imp-2 on the one hand, and the use of impersonally used personal pronouns on the other hand. Their analysis has the following assumptions, which will be explained in more detail: 1) the syntactic person and number features operate on an input set; 2) if phi features are present in the syntax, the interpretation of this phi node needs to be preserved within the DP (‘Preservation of Phi’).
There are only two person features, [prox] and [dist], that are functions that operate on an input set (following Harbour 2011a,b). The semantic input set Si+u+o consists of speaker(s)
(i); addressee(s) , (u); and others that can neither be associates of the speaker and addressee (o) (example (17) taken from Ackema and Neeleman 2014).3
(17)
If we now want to select a subset of the input set, the two person features can operate on the whole set. For example, [dist] is a feature that selects the outer layer of the set: leading to So, thus leading to a 3sg reading. [prox] on the other hand ‘deletes’ the outer layer and gives the subset: Si+u..
If a phi-feature is used in the syntax, it needs to stay interpretable in the DP. It can never be the case that a phi-feature is used to select a part of the input set and is then overridden during the derivation. A&N call this the ‘Preservation of Phi’ and this will be important to derive the difference between imp-1 and imp-2. Thus, for example, [dist] selects the outer layer of the input set and this will lead to a 3sg reading. It is not possible anymore to also apply [prox] to the input set. Already (o) is selected by [dist] and [prox] cannot select anything anymore. This entails that the first selected phi feature needs to be interpretable during the derivation. If this is not the case, it would be possible for [prox] to override [dist], which would yield an ungrammatical result.
3
The analysis presented here is a simplified version. The semantic input also consists of associates of either the speaker and / or the addressee. This is however, not important for the idea presented here. See for an extensive discussion Ackema and Neeleman (2014, in prep.)
i u o Si Si+u Si+u+o
Now, lets turn to impersonal readings. In order to get a generic reading, a generic operator, GEN is put on top of the person and number nodes, similar to Egerlands (2003) proposal. This operator selects an undefined group from the semantic set. Next to this operator, A&N assume, contra to Egerland, that arbitrary readings also arise because of another operator, ARB (similar to a proposal made by Cabredo Hoffherr 2010). This operator selects one individual from the semantic set. These two operators can only be put on top of nouns that are structurally specified enough. For example, the features on 1pl are [prox, plural] and therefore select a group: a plurality of Si+u. If ARB would select this structure, the result would be ungrammatical since ARB needs to select one individual, while the phi features have selected a group, (18)b. A Generic operator leads to a grammatical result, since both the person and number features and the operator select a group (18)a.
(18) a. Generic 1pl b. *Arbitrary 1pl
GEN Number ARB Number
Number Person Number Person
Plural Person N Plural Person N
Prox Prox
(undefined) ≈ plural group 1 ≠ plural group
group individual
Here, it is important to see that the syntax has a phi-node, and this node needs to be persevered during the derivation. If phi is preserved, arbitrary readings with 1pl pronouns are
ruled out: it is not possible to select one individual if person and number already selected a group. This analysis thus correctly predicts that some personal pronouns can have an impersonal reading. In the case of the example, only generic readings are predicted.
The analysis of imp-1 and imp-2 pronouns is as follows. Just as in the previous proposals, A&N also argue that the feature make-up of 1 consists of a phi-node, while imp-2 pronouns lack features. The difference with the proposal of Egerland is that the difference between the two types is not lexical, but syntactic.
(19) a. imp-1 b. imp-2
φ N
φ N
Since imp-2 does not have any phi-features, the input set is not selected, thus the output set is the same as the input: Si+u+o.. Nothing is selected by the syntax, and therefore the whole input set is available. Both GEN and ARB can select this structure, without violating the principle that phi needs to be preserved, since there is no phi node. Therefore, ARB can select a single individual from the output set. The same holds for GEN: it can select an undefined group. The structure of imp-1 on the other hand, does not allow the ARB operator. The underspecified phi feature selects the whole input set, leading to an output set Si+u+o. The crucial difference with the output set of imp-2 pronouns is that nothing from the set can be eliminated, because everything is selected, whereas you can select a subset with imp-2 pronouns. Therefore, if you have ARB selecting phi + N, the result will be ungrammatical (20)a, whereas GEN will be grammatical (20)b.
(20) a. Generic one b. *Arbitrary one
GEN Person ARB Person
Person N Person N
(undefined) ≈ Si+u+o 1 ≠ Si+o+u
group individual
This analysis correctly predicts that the semantics of the pronouns can be plural, whereas the syntax is singular. From the underspecified phi-node, it is not clear what type of agreement will occur. However, there is not a plural feature present, thus plural agreement is not likely. A&N argue that 3sg agreement is the default agreement.
The analysis presented by A&N uses two assumptions: syntactic person features operate on a semantic input set; 2) if phi features are present in the syntax, the interpretation of this phi node needs to be preserved within the DP (‘Preservation of Phi’). With this two assumptions A&N argue that the different feature specifications derive the differences for the two types of dedicated impersonal pronouns imp-1 and imp-2 on the one hand and personal pronouns that can be used impersonal on the other hand. Imp-1 and imp-2 differ in their feature specification, since imp-2 is underspecified and imp-1 only has a phi node. Although the point about the impersonal used personal pronouns is not discussed in detail here, their proposal is that some specific feature combinations of [prox], [dist] and [plural] can lead to impersonal readings. Therefore, there are differences between you and one, because one is not specified for those features. There is one point that is not explained by their proposal, namely the fact that imp-2 pronouns cannot occur in object position. They do, however, not make any claims about this observation.
§2.3 The goal of this thesis
In the previous sections, it has been shown that a difference can be made between two types of dedicated impersonal pronouns, called imp-1 and imp-2 (see table 1). The main differences between imp-1 and imp-2 have been discussed in section 2.1 and are the following: imp-1 has a less restricted syntactic distribution and can occur in both subject and object position, whereas imp-2 can only occur in subject position (points iii and v in table 2). The reverse pattern holds for the different possible meanings: imp-1 can only have a generic reading, whereas imp-2 can also receive an arbitrary reading (points vi and vii in table 2). Section 2.2 discussed three main theoretical approaches to account for these differences. All three theories (Egerland 2003; Hoekstra 2010 and Ackema and Neeleman 2014, in prep.) seemed to agree on at least one point: the difference between imp-1 and imp-2 pronouns is derived by a difference in feature specification. Although the features of imp-1 differ per approach, all have in common that imp-2 pronouns are radically underspecified. This entails that an imp-2 pronoun does not have any features in his functional structure, as is shown in (11)b for Egerlands approach, in (16)a for Hoekstra’s analysis and in (19)b for Ackema and Neeleman. This underspecificity means that both generic readings and arbitrary readings can occur with the imp-2 type. For all three approaches a generic reading occurs when a generic operator selects the pronoun. This operator can select both imp-1 and imp-2 and the result will be grammatical. Arbitrary readings are derived differently in Egerlands approach on the one hand, and A&N on the other hand (Hoekstra does not claim anything w.r.t. arbitrary readings). For Egerland, arbitrary readings can occur if there are no features or operators present. This can only be the case with imp-2, since there are no features. For A&N an arbitrary reading is derived with a different operator in combination with the preservation of phi principle. The principle states that if a phi node is used in the syntax, it needs to be interpretable in the DP and it cannot be overridden during the rest
of the derivation. Recall that only imp-1 has an empty phi-node, which thus selects the complete input set. The ARB operator can only select imp-2 and not imp-1, since the phi-feature on imp-1 has selected the whole semantic input and thus a plurality of individuals are selected, whereas the ARB operator only selects one individual.
The similar feature make-up in all approaches can derive the different possible readings of imp-1 and imp-2. However, it is not clear how the syntactic properties that distinguish imp-1 and imp-2 are derived and this will be the main topic for this thesis. These observations, iv and v in table two, are repeated in (21).
(21) iv. Imp-2 cannot have an arbitrary reading when it is the subject of a passive or ergative sentence (Cinque’s 1988 generalization)
v. imp-2 can never occur in object position
All theories do not have a clear answer for these two generalizations. Ackema & Neeleman do not make any claims about these generalizations; Egerland and Hoekstra on the other hand, do, but several problems have been pointed out. Both argue that the absence of any functional features on top of imp-2 is the reason that imp-2 cannot be an underlying object, since there is no feature to evaluate the predicate-object relation. Imp-1 on the other hand, can start as an underlying object and surface as an object, because it has a phi-node. The reason that imp-2 can have a generic reading as a derived subject, and thus can be the subject of a passive or ergative sentence, is because of the generic operator on top of the pronoun. This operator makes the evaluation possible between the object and the verb. However, if the generic operator licenses the generic reading when the pronoun is a derived subject, it is not clear why this operator doesn’t license the pronoun to be grammatical as a surface object. Thus the question is why the movement of the pronoun to subject position makes a generic reading
grammatical. Moreover, it is not clear whether the claim made by Cinque (1988) is correct, namely the fact that an imp-2 pronoun can never have an arbitrary reading when it occurs as a derived subject. At least in Dutch, it is possible for some people to have an arbitrary reading when the impersonal pronoun is in subject position of a passive, see example (22).4
(22) (Context: Het heeft gevroren en het meer in het bos is bevroren, maar er is een wak Context: The has frozen and the lake in the forest is frozen, but there is a hole in het ijs.) Gisteren is men hier door het ijs gezakt.
In the ice.) Yesterday is IMP here through the ice fallen.
(Context: It has been freezing and the lake in the forest is frozen. However, there is a hole in the ice.) Apparently, yesterday somebody has fallen through the ice.
If it is true that, at least for Dutch, arbitrary readings are possible in subject position, it is worthwhile to see how the pattern is for other languages. Moreover, if the pattern holds, the analysis that imp-2 pronouns cannot receive an arbitrary reading if they are an internal argument needs to be revised. This leads to the question why imp-2 can start as an underlying object, but cannot end as a surface object, only as a surface subject. This also relates to the question mentioned above, why the movement of the pronoun out of object position makes the occurrence of the pronoun grammatical. Syntactically speaking, the pronoun starts in the same position. However, it does receive different case in both surface positions: nominative if it is a
4
It has been pointed out to me by Hans Broekhuis and Eric Hoekstra that an arbitrary reading is only available if the pragmatic context is clear enough and therefore it is not clear if this reading is due to the grammar or pragmatics. However, on the view that pragmatic readings can only result from all available options of the grammar, the arbitrary reading should be grammatical. It seems that all impersonal
readings should have some pragmatic context: generic readings are most of the time allowed by a modal etc. The question remains why some impersonal readings need more contexts than others. Moreover, if the context would give the arbitrary reading, we would also expect this reading to be grammatical with languages that do not allow generic readings in subject position at all, such as English and Frisian. However, the results of this study show that there is a division between grammaticality of this sentence with imp-1 and imp-2 pronouns.
subject and accusative when it is an object. Therefore, it might be that the generalization about the syntactic distribution of imp-2 is that it can only occur with nominative case, whereas imp-1 can also occur with accusative, or other cases. To see if this generalization works better than the observation made by Cinque (1988) and Egerland (2003), it might be useful to look at sentences where the pronoun receives its theta-role from the embedded clause, but case from another syntactic position, such as him in (23).
(23) I see him walk to the library every day
In the literature this type of construction is either called Raising-To-Object (Postal 1974) or Exceptional-Case-Marking (Chomsky 1986). In the analysis of the first construction only the subject of the embedded clause moves into the object position of the main clause and receives accusative there; in the second analysis the whole embedded clause receives case from the main clause. Which analysis is on the right track is not important for this paper. The crucial point is that the subject of the embedded clause receives its theta-role from the embedded clause, while it receives case from the main clause. Thus him in (23) receives its theta-role from the embedded clause and is a subject, but receives accusative case from the verb in the main clause. Now, it is interesting to see which of the dedicated impersonal pronouns are grammatical in this position and what types of readings are grammatical. If the generalization is correct that imp-2 are grammatical in subject position, we would expect them to occur as an ECM-subject, since it is syntactically a subject. However, if case restricts the distribution of imp-2, we would expect imp-2 to be ungrammatical in an ECM-sentence. This type of sentences therefore can disentangle between the two generalizations.
Summarizing, there are several empirical questions that can help decide what the theoretical status is of dedicated impersonal pronouns. The previous discussion led to the following -empirical- questions:
1) Are arbitrary readings possible if the imp-2 pronoun is a derived subject?
2) If a language with imp-2 has no suppletive form, are ECM constructions with a dedicated impersonal pronoun grammatical?
3) What readings are possible with the suppletive pronoun in an imp-2 language?
First of all, it is interesting to see whether arbitrary readings are possible if the pronoun is the subject of a passive or ergative sentence, since that would shed light on the claims made by Cinque (1988) and Egerland (2003). Secondly, it is worthwhile to see what happens if impersonal pronouns are an ECM-subject, because the pronoun will be syntactically a subject, but will have a different case. To this end, a questionnaire is developed and four languages are tested. Two languages seem to have imp-1 pronouns, namely English and Frisian (see table 1). Two other languages have imp-2 pronouns: German and Dutch. Although there is one group of languages with imp-1 and another with imp-2, there are differences within both groups with respect to (the richness of the) agreement and the use of suppletive forms in object position. As was already shown in section 2.1 and 2.2 both German (an imp-2 language) and Frisian (an imp-1 language) have suppletive forms in object position: einen and jin respectively. Therefore, it is interesting to see what will happen in all these languages and what readings are possible with those suppletive forms.
The data from the questionnaire can help us understand the nature of impersonal pronouns better. Moreover, the data can help shed light on how case interacts with impersonal pronouns. The next section provides the set-up of the questionnaire in detail. In section 4 the
results will be given. It will be shown that imp-2 pronouns can only occur with nominative agreement, whereas imp-1 pronouns can occur with accusative too.
§3 METHOD
The previous section showed that it might be useful to look at several sentence types, namely those where the dedicated impersonal pronoun is a derived subject and the subject of an ECM or Raising-To-Object construction, in order to see what restricts the syntactic distribution of imp-2 pronouns. Since the research is more a qualitative research rather than a quantitative research, because it is about the use and distribution of the pronoun and not the frequency, databases, such as the ones used by Siewierska (2012) and Giacalone Ramat & Sansò (2007), are not used and a questionnaire is developed. Moreover, as Siewierska (2008) and Weerman (2006) show, the pronoun is not used that often anymore, thus it is more difficult to see where the pronoun is grammatical in a database than with a questionnaire. Therefore, In order to find those generalizations a questionnaire is developed in which linguistic participants need to judge the grammaticality of the sentences. Since the sentences, as will be shown below, ask for subtle judgments, the research will be small and the participants will be linguists. The properties that are tested are summarized in table 3.
Pronoun Reading Context Impersonal Generic Subject 3pl Arbitrary Ergative
Passive
ECM
Object
Table 3. tested properties
First of all, the questionnaire looks at two pronoun types: the dedicated impersonal pronoun, such as one and man and the form for the –normally used – 3pl personal pronoun, such as they. The most important difference that needs to be tested is the behavior of imp-1 and imp-2. Therefore, the questionnaire contains sentences with these pronouns. However, an important question of this research is whether arbitrary readings are possible if the pronoun is a derived subject. If a participant judges a pronoun in that sentence ungrammatical with an arbitrary reading this might have different reasons. It might be due to the sentence and the position of the pronoun, but it might also be the case that the participant has problems with arbitrary readings in general. Therefore, sentences with a 3pl pronoun are also included. This pronoun can be used for generic readings and arbitrary readings (Cabredo-Hofherr 2003) and is thus a good pronoun to control for the use of arbitrary readings. If a participant is not able to use arbitrary readings with 3pl pronouns, it is not clear what the reason might be for the ungrammatical judgments with arbitrary readings in derived subjects. Those participants will therefore be excluded from the dataset. All sentences are minimal pairs: the same sentence is given with an impersonal pronoun and a 3pl pronoun.
The next property that is tested, are two possible readings: generic and arbitrary readings. For the languages that have imp-2 pronouns, it is necessary to see where arbitrary readings are possible. For imp-1 languages, that only have generic readings, it is also necessary to see in which syntactic position the pronoun is grammatical. Both pronouns will be tested in the questionnaire with both readings.
The third property that is tested is the position of the pronoun, because it is important to see in which positions the dedicated impersonal pronouns are grammatical and which readings they receive. The first position is the subject position of non-passive sentences. This is not a crucial test-condition, but it is included in the questionnaire to (re-)confirm the claims made in the literature, concerning the observations that imp-2 and imp-1 are both grammatical in that position, but only imp-2 can have both an arbitrary and generic reading (Egerland 2003). Thus the sentences can also serve as a control condition: if a participant is not able to use impersonal pronouns at all in simple sentences where it is a subject, the results will be excluded. An example sentence is given in (24).
(24) When < one is / they are > in Italy, <one eats / they eat > pasta
The second and third conditions are those where the pronoun is the derived subject of an ergative and passive sentence, such as the sentence given in (22). These sentences are included to see if arbitrary readings are possible with imp-2 pronouns. For imp-1 pronouns it is predicted that only the sentences with a generic reading are grammatical.
In the fourth condition sentences where the pronoun is the ECM-subject are tested, an example is given in (25).
This condition is necessary to see what will happen if the pronoun is a subject and receives accusative case. Again, imp-1 pronouns are predicted to be grammatical if the pronoun has a generic reading and the suppletive forms in languages with imp-2 are predicted to be grammatical, whereas it remains unclear what readings are grammatical with imp-2 pronouns without a suppletive form, such as Dutch. Moreover, it is interesting to see what readings are possible with the suppletive form in imp-2 type languages.
The final condition is where the pronoun is the object of the predicate, such as the sentence presented in (6). These sentences will also have a controlling function, since the literature has suggested that imp-2 pronouns can never occur in object position (a.o. Egerland 2003; Weerman 2006).
If all three properties are combined, the questionnaire would contain of 2(pronouns) x 2(readings) x 5 (conditions) = 20 test sentences. However, some test sentences, such as the arbitrary reading of imp-2 in derived subject position, is more crucial than other sentences. Therefore, of those conditions more than one test-sentence is included, leading to a total of 31 test sentences. Moreover, for those languages that have a suppletive form, such as German, both the imp-1 form as the imp-2 form are included to see what pronoun is grammatical.
After each test sentence, it is asked which of the readings is possible with this pronoun and which one is preferred. This strategy is chosen, because now a participant needs to look at all possible readings, and does not need to come up with a reading themselves.
Since one of the goals of this research is to see what the difference between imp-1 and 2 pronouns is, each pronoun will be tested in two languages. Frisian and English have imp-1 pronouns and Dutch and German have imp-2 pronouns.
As already mentioned above, the dedicated impersonal pronouns are not very frequently used in most languages, and because the sentences ask for subtle judgments, the participants
are linguists. One participant translates the questionnaire into one of the four languages. Preferably, two other participants also judge the -translated- sentences. This is done in order to make sure that every participant of the same language judges the same sentences.
§4 RESULTS
To summarize, there are three empirical questions that this research aims to answer, repeated below.
1) Are arbitrary readings possible if the imp-2 pronoun is a derived subject?
2) If a language with imp-2 has no suppletive form, are ECM constructions with a dedicated impersonal pronoun grammatical?
3) What readings are possible with the suppletive pronoun in an imp-2 language?
Although the questions are about languages that have imp-2 pronouns, this research also investigated languages with imp-1 pronouns, to see where the differences arise. It will be shown that imp-2 pronouns can never occur with any other case than nominative, whereas imp-1 pronouns can. This distribution follows if we propose a syntactic requirement that at least one phi-node is needed during the derivation. Imp-2 pronouns do not have a phi-node in their structure, but can receive one from the verb with which it agrees. Imp-1 pronouns on the other hand, do have a phi-node and can therefore occur with other cases than nominative. As will be discussed later, this analysis has some consequences for the view on agreement.
Now, lets turn to the results of the questionnaire. In (26) and (27) the differences are shown if the pronoun is the subject of a non-passive sentence. All participants judged the test sentence in (26) grammatical with a generic reading. The other sentence, (27), with an arbitrary reading, is only possible in Dutch and German.
(26) D(utch) Wanneer men in Italie is, eet men pasta [generic] G(erman) Wenn man in Italien ist, isst man Nudeln. [generic] E(nglish) When one is in Italy, one eats pasta [generic] F(risian Wannear’t men yn Italië is, yt men pasta [generic]
(27) D. Men heeft voor je gebeld, maar ik weet niet waar het over ging [arbitrary] G. Man hat dich angerufen, aber ich weiss nicht, worum es ging. [arbitrary] E. One has called for you, but I don’t know what it was about *[arbitarry] F. Men hat foar Jo skille, mar ik wyt net wêr’t it oer gie *[arbitrary]
These two sentences already show the difference between imp-1 and imp-2: imp-1 only allows generic readings, whereas imp-2 is flexible (recall that this are points vi and vii in table 2 in section 2.1). Although it might be superfluous, but again, it is shown that it is not possible to make an exact division between the two types of pronouns and their origin. Frisian seems to pattern with English, but both have a different pronoun: Frisian the one stemming from ‘human’, whereas the English pronoun stems from a numeral. This is also the case for the Icelandic form maður, which also stems from ‘human’ and only allows generic readings, as is shown in (28) (examples taken from Egerland 2003, p. 81).
(28) a. Maður vinnur til 65 ára aldurs [generic]
IMP works until 65 years age
‘one works until the age of 65’
b. Maður hefur unnið að því í tvo mánuði að leysa vandamáliðn *[arbitrary]
‘someone has worked for two moths to solve the problem’
If we now turn to the first question, ‘can imp-2 pronouns have an arbitrary reading if it is a derived subject?’, we see that it is possible for languages that have imp-2 pronouns to have arbitrary readings if the subject is underlying an object. This is shown in (29) for an ergative sentence, and it means that there is some individual who has made a hole in the ice. In (30) the intended meaning is that there one individual who went to the hairdresser.
(29) D. (Context: Het heeft gevroren en het meer in het bos is bevroren, maar
er is een wak in het ijs.) Gisteren is men hier door het ijs gezakt.
[arbitrary]
G. Endlich ist das Eis auf dem See dick genug um Schlittschuh zu laufen. Letzte Woche ist man hier noch eingebrochen.
[arbitrary]
E. (Context: It has been freezing and the lake in the forest is frozen. However, there is a hole in the ice.) Apparently, yesterday one has fallen through the ice.
*[arbitrary]
F. (Context: It hat ferzen en de mar yn de bosk is beferzen, mar der sit in wek yn it iis.) Juster is men hjir troch it iis sake.
*[arbitrary]
(30) D. (Context: Er ligt haar op de grond bij de kapper) Men is hier goed te grazen genomen (door de kapper)
[arbitrary]
G. (Kontext: Auf dem Fussboden vom Friseursalon liegt abgeschnittenes Haar.)
Man ist hier (vom Friseur) kahlrasiert worden
[arbitrary]
E. (Context: At the barber, there is hair on the floor) One is shaved bald (by the barber).
F. (Context: Der leit hier op ‘e grûn by de kapper.) Men is hjir goed beetnommen (troch de kapper)
*[arbitrary]
Although there is more context needed to trigger the arbitrary reading, it cannot be the case that the context is what gives the sentence an arbitrary reading. If that would be the case, the sentence should be grammatical in Frisian and English, where an imp-1 pronoun is used. As can be observed in the above examples, there is a difference between the grammaticality of imp-1 and imp-2 pronouns.
Concluding, the following generalization seems to hold:
A. Languages that have imp-2 pronouns do allow arbitrary readings if the pronoun is the subject of a passive or ergative sentence
This generalization already shows that it is possible for imp-2 pronouns to occur in an underlying object position either with an arbitrary and a generic reading. Therefore, the claims made by Cinque (1988) and Egerland (2003) are falsified.
The second question, asking if it is possible to have an imp-2 pronoun as the subject of an ECM construction, is answered negative, as can be seen in (31) for sentences where a generic reading is triggered and (32) where an arbitrary reading is triggered. Imp-2 pronouns cannot occur as ECM subjects if a generic reading and if an arbitrary reading is triggered. Imp-1 pronouns can occur as the subject of an ECM construction, but are only grammatical with a generic reading.
(31) D. Omdat ik treinconducteur ben, zie ik men altijd op vakantie gaan *[generic] G. a. Der Bahnhofswärter sieht man immer in Ferien fahren. *[generic]
b. Der Bahnhofswärter sieht einen immer in Ferien fahren. [generic] E. The train conductor sees one always leave for the holidays. [generic] F. a. Omdat ik treinkondukteur bin, sjoch ik men altiten mei fakânsje
gean5
b. Omdat ik treinkondukteur bin, sjoch ik jin altiten mei fakânsje gean
*[generic]
[generic]
(32) D. Ik heb de hele nacht wakker gelegen. Ik hoorde men aan de weg werken
*[arbitrary]
G. a. Ich lag die ganze Nacht wach. Ich habe die ganze Zeit man auf/an der Straße arbeiten hören.
b. Ich lag die ganze Nacht wach. Ich habe die ganze Zeit einen auf/an der Straße arbeiten hören.
*[arbitrary]
*[arbitrary]
E. I lay awake all night. I hear one work on the road. *[arbitrary]
F. a. Ik ha de hiele nacht wekker lein. Ik hearde men oan de dyk
wurkjen
b. Ik ha de hiele nacht wekker lein. Ik hearde jin oan de dyk wurkjen
*[arbitrary]
*[arbitrary]
Moreover, it can be seen that for German, the imp-2 form is not grammatical, but the imp-1 form is. The same holds for Frisian, where men is not grammatical, but the suppletive form jin is. These suppletive forms have a generic reading and not an arbitrary one, thus answering question 3, which asks what possible readings are allowed. Therefore, Frisian seems to pattern
5
All Frisian informants judged all sentences with the pronoun as an ECM subject as ungrammatical. This was due to the fact that the pronoun always has a reading that includes the speaker. In the main clause the 1sg pronoun was used, and therefore the impersonal pronoun could not have a generic inclusive reading. The pronoun jin, however, is grammatical as ECM subject, for example in the following sentence:
