From an analysis of Fula, this article provides straightforward evidence that the hearer feature is binary. The analysis counts on Local Dislocation to handle mixed placement of subject agreement markers. A novel aspect of the analysis is that a natural class made possible by impoverishment defines the environment in which Local Dislocation applies. Impoverishment has mainly been used to capture syncretism in the past, but the present study shows that it has further potential, especially in the area of mixed placement of agreement markers. The impoverishment operation used is motivated by cooccurrence restrictions that arise in the interaction of person features with number features, falling squarely within the research agenda that tries to ground impoverishment in the logical properties of the feature system.
The significance of φ-features cannot be overemphasized in recent theorizing. They feed semantic interpretation and are reflected in morphological realization. These two aspects are connected by agreement, which takes place in narrow syntax. Given their pervasive role in the entirety of linguistic computation (see Harbour, Adger, and Béjar 2008 for showcase studies), it is very important to identify the exact inventory of φ-features provided by Universal Grammar (UG). I would like to focus on person features in this article.
A major issue in the treatment of features is whether they are binary or privative. Person features are no exception in this regard. Harley and Ritter (2002) advocate a privative system covering person and number. In the domain of person, [participant], [speaker], and [hearer]1 are organized in a hierarchical fashion under their proposal. These person features appear in the representation only if they are active. Proposals of a binary system are fairly common, on the other hand, and include Noyer 1997 and Silverstein 1976, to mention just two. See Bobaljik 2008 for an overview.
This simplistic opposition of privative and binary systems needs more elaboration in relation to the architecture of the entire computational system. For example, Harbour (2011b) argues that number features are bivalent both in syntax and in morphology, whereas Nevins (2011b) claims that binarity of [±singular] should be limited to postsyntactic morphological computation. Person features allow the same range of theoretical possibilities. To avoid proliferation of names for various theoretical options, I will apply the term binary to features for which both the plus and minus values are posited in some way or another, regardless of the architectural question, and privative otherwise.2 I will also make clear for which part of the computational system the choice between privative and binary systems is relevant.
There is another respect in which this choice is not all-or-nothing: some of the person features can be privative while others are binary. Harbour (2006) and Nevins (2007) put forth a concrete proposal—namely, that the hearer feature is privative whereas the speaker and participant features are binary. Logically, there are other possible partially privative systems, but this is the only one actually proposed, as far as I am aware. We thus have the following options to consider:
I will take up the nature of the hearer feature as the empirical problem to be addressed in this article. My goal is to present compelling evidence that the hearer feature is binary during the postsyntactic morphological computation.
As an empirical basis for my demonstration, I will work on a system with the inclusive/ exclusive distinction, for which the characterization in terms of the speaker and hearer features is sufficient, as indicated in (2).
Labels such as 1 + 2(+ 3) are used here as the metalanguage for describing available person categories. The traditional terms are 1st person exclusive for 1(+ 3), 1st person inclusive for 1 + 2(+ 3), 2nd person for 2(+ 3), and 3rd person for 3. The optional (+ 3) in 1st and 2nd persons is related to the number dimension, which is independent of the person distinction.
When we turn to the interaction between person and number, it becomes highly significant to be explicit about how person features are to be interpreted in semantics. The correct way is stated in (3).
(3) Semantic interpretation of person features
The positive value ( plus or presence) is interpreted as inclusion of the relevant discourse participant.
For example, the speaker is included in the entity denoted by the expression containing [+ speaker]. The speaker is not included in the case of [− speaker]. In the privative parlance, the speaker is not included unless [speaker] is present.3 The idea goes back to Burling 1970 and Zwicky 1977 and is formalized in Heim 2008. This interpretation enables a straightforward account of person-number interaction, as shown in Harbour 2011a. First person plural includes the speaker, plus an additional set of individuals, period.
If [+ speaker] were read as simple identification with the speaker, a special stipulation would be needed for inclusive, which is [+ speaker, + hearer], for inclusive does not mean ‘someone who is the speaker and the hearer at the same time’.4 Kratzer (2009:220–221), for instance, has to invoke the sum operation to ensure the correct reading of inclusive, because she adopts identification with the speaker/hearer as the mode of person feature interpretation. No such complication arises if the positive value signifies inclusion.
In arguing for the binary character of the hearer feature, I will examine a type of phenomenon that has not been highlighted before in discussions of the morphological behavior of φ-features. The novelty of the proposed analysis lies in making crucial use of impoverishment, which has mainly been employed as a means of affecting the way Vocabulary items are inserted (Bobaljik 2002, Bonet 1991, Calabrese 2011, Frampton 2002, Halle 1997, Harley 2008, Nevins 2011a, Noyer 1997, 1998), to provide an analysis of a phenomenon not related to Vocabulary Insertion. Impoverishment, which deletes features, is one of the morphological operations posited in Distributed Morphology (Halle and Marantz 1993), where the output of narrow syntax computation is allowed to undergo various modifications before phonological exponents of functional morphemes (Embick 2010) such as the Tense node are provided. I will show that impoverishment can affect conditioning factors of another morphological operation called Local Dislocation (Embick and Noyer 2001), which modifies the placement of morphological entities as part of the process that linearizes the output of narrow syntax.
The research carried out in this article advances the program of grounding impoverishment in the logical properties of the feature system. Following Noyer (1998), Nevins (2011a) pursues the idea that impoverishment is motivated by markedness considerations (see also Calabrese 2011). I will suggest that there is another aspect of the feature system that drives impoverishment, namely, cooccurrence restrictions. The idea is that when only one value of feature F is possible in combination with the rest of feature specification, that feature can be deleted. I will explore cooccurrence restrictions in the interaction of person features with number features.
Section 2 introduces the main empirical challenge tackled in this article: placement of subject markers in Fula, a member of the West Atlantic branch of the Niger-Congo family. Section 3 gives an analysis in terms of impoverishment and Local Dislocation. Section 4 turns to difficulties in motivating natural feature classes on the basis of syncretism. Section 5 discusses the typology of impoverishment. Section 6 takes up the reasons why the hearer feature has been regarded as privative. Section 7 concludes.
2 Fula Subject Markers
Placement of subject markers exhibits a curious pattern in the so-called relative tenses of Fula. Consider the past tense paradigm of the active voice illustrated in (4).
The puzzle comes from the split behavior of the [+speaker, −hearer] category in interaction with the number distinction, to phrase it in terms of the binary feature system. The subject marker is placed postverbally in the singular, whereas it is placed preverbally in the plural. The same split is found in the relative future, as illustrated for the active voice in (5).
(5) Fula relative future active ‘will wash’ (Arnott 1970:192)
The relative tenses are used in wh-movement contexts, such as interrogatives and relative clauses, as shown in (6).
’O-’yamet-e kò shoodu-mi.
he-will.ask-you what bought-I
‘He will ask you what I bought.’
Stewart and Stump (2007:385) remark that it is impossible to define a natural class that accounts for the behavior of the Fula subject marker in the relative tenses. To be sure, the split behavior of an identical person category appears to defy a principled explanation.
The postverbal placement of 1st person singular is not found in tenses other than the two relative tenses. In the subjunctive, whose pattern is otherwise very similar to that of the relative tenses, the subject marker for 1st person singular appears preverbally, as (7) illustrates.
(7) Fula subjunctive active ‘wash’ (Arnott 1970:192)
In all the other tenses, the subject marker uniformly appears preverbally, as exemplified by the general past tense in (8).
(8) Fula general past active ‘got lost’ (Arnott 1970:180)
To summarize, the two relative tenses form a natural class with respect to the location of the subject marker. The data pattern looks recalcitrant at first sight, but I claim that it requires a principled account in terms of features, despite Stewart and Stump’s pessimistic remarks. Note that the pattern found in the subjunctive is extremely close to that of the relative tenses, differing only in the behavior of 1st person singular. In other words, there is something special about 1st person singular. We need to understand what it is in order to provide an account of the relative tenses.
The puzzling behavior of the subject marker in the relative tenses becomes understandable once the logic of person-number interaction is closely examined. I spell out the analysis in this section.
3.1 Underspecification of the Morphosyntactic Representation
The key observation to be made is that some feature values are predictable in person-number interaction. In the fully binary representation in (9), the parenthesized features can be omitted without loss of information, since they are predictable from the rest of the specification.
Significantly, 1st person singular is one of the combinations that allow one of the features to be omitted. The semantic properties entailed by the combination [+ speaker, + singular] allow only the minus value of [± hearer], since the singular individual cannot include the speaker and the hearer at the same time. On the other hand, no such restriction applies to [+ speaker, − singular]. First person exclusive plural must be specified as [− hearer] to be distinguished from 1st person inclusive. Now, after omission of the predictable features, it is possible to define a natural class that is correlated with the preverbal placement of the subject marker, namely, [− hearer]. This feature groups together 1st person exclusive plural, 3rd person singular, and 3rd person plural.
Important in our investigation of the nature of the hearer feature is the fact that the crucial natural class must refer to [− hearer]. It cannot be defined in a system with the privative [hearer]. At this point, we have rather straightforward evidence that the hearer feature is binary for operations that are responsible for the placement of the subject marker. These operations must be able to mention the minus value of [±hearer]. The proposed analysis also enables us to capture the similarity between the subjunctive and the relative tenses at the same time. The subject marker placement in (7) can be characterized in terms of [− hearer] straightforwardly. Thus, the single statement in (10) suffices.
(10) Subject marker placement in relative tense and subjunctive clauses
The subject marker appears preverbally if it is [− hearer]. Otherwise, it appears postverbally.
We are entitled to say that the proposed analysis has uncovered a deeper generalization about Fula subject marker placement.
Of course, the predictable features should not be omitted in the case of the subjunctive. Such omission must be confined to the relative tenses. In this sense, the phenomenon in question is morphological in nature. It is not something that occurs across the board in narrow syntax. A formal device that can achieve the kind of feature omission just proposed is impoverishment in the Distributed Morphology framework (Halle and Marantz 1993). My concrete formulation of the rule is as follows:
(11) Impoverishment of [− hearer] in Fula
Delete [− hearer] from the feature matrix of the subject marker in the relative tenses when its value is predictable from the rest of the feature specification.
Deletion is limited to [− hearer], because there is no need to eliminate [− singular] from 1st person inclusive or [− speaker] from 2nd person singular as far as the account of Fula is concerned. Impoverishment is a rule that does something extra to the output of narrow syntax in the postsyntactic morphological computation. Usual considerations of parsimony allow only a minimal addition. A superfluous degree of complexity must be avoided. What is required in Fula is therefore the rule in (11), which yields the underspecified representation in (12), with [− hearer] boxed.
The next task is to give a formal account of the generalization stated in (10).
3.2 Morphological Preposing of the [−hearer] Subject Marker
So far, I have not been explicit about the exact morphosyntactic status of the Fula subject marker. Now I would like to claim that it is an ordinary agreement marker, in view of the fact that an independent pronominal subject can appear side by side with the subject marker as in (13).
’O-’anndaa ko miin ’anndu-mi.
he-know.neg what I know-I
‘He doesn’t know what I know.’
Fula is an SVO language. The word order in (13) suggests that the inflected verb is not raised to the CP domain when wh-movement takes place.
A first question to be answered when we try to capture the generalization in (10) in formal terms is where the underlying basic location of the subject marker is. As an initial step, consider the verb endings in the relative past active (4) and relative future active (5), summarized in (14).
Relative past active endings: -u, -i
Relative future active endings: -ay, -et, -ot, -at(a)
When postverbal, the subject marker follows these endings, which encode voice and tense/aspect information together.6McIntosh (1984) argues that what Arnott (1970) calls a tense distinction is in fact an aspectual one. If McIntosh’s analysis is correct, the endings in (14) should not be treated as carrying tense information, and we need to know whether Fula has tense morphology.7 The only candidate is the suffixal preterite element -no(o), which places the event/state ‘‘one stage back in time’’ (Arnott 1970:216). Thus, the realization of the tense morpheme is postverbal in Fula, whether we adopt Arnott’s analysis or McIntosh’s. The subject marker follows the preterite element when it appears postverbally, as shown in (15).
‘I’m going to help’
‘I was going to help’
‘they are going to help’
‘they were going to help’
In order to be clear about the structure of the verbal complex, we need to know whether the verb is raised to T. In Fula, negation is expressed together with voice and tense/aspect information, encoded as part of the suffixal complex (Arnott 1970:chap. 53).8 (17b) and (18b) are the negative counterparts of the general past active and general future active endings, respectively.
General past active: -ii
Negative past active: -aayi
General future active: -ay
Negative future active: -(a)taa
The portmanteau expression of negation suggests that the verb is raised to T, on the way picking up the Neg head, which appears as -aa before and after the tense ending in past and future, respectively.
Given that the φ-features for subject agreement are contained in the T head (Chomsky 1995: chap. 4), their precise location within the complex T head resulting from head movement of V to T needs to be specified in one way or another.9 When the exponent of T is postverbal as in Fula, the three possibilities in (19) exist. (The presence of a light verb is suppressed here for the sake of brevity.)
The node designated as Φ in (19) does not exist in narrow syntax. It is added during postsyntactic morphological computation to provide a position for the exponent of φ-features carried by the T node. Embick and Noyer (2001:558) call such heads dissociated morphemes. Oltra-Massuet and Arregi (2005) place the agreement features right next to the terminal node of T as in (19a) on the grounds that Spanish fuses tense and agreement morphology in the present tense, the conditional, and so on. Fused expression of tense and agreement morphology is a fairly common pattern crosslinguistically (see Julien 2002 for quantitative data). Fusion, as proposed in Halle and Marantz 1993, is an operation that combines two sister terminal nodes into one prior to Vocabulary Insertion, allowing fused realization of tense and agreement, for example. One might question the assumption that fusion requires sisterhood, however, as noted by a reviewer, who suggests that it can be formulated under adjacency. In that case, (19b)—essentially the structure adopted in Embick’s (2010) account of Latin—is also a viable option. Now, it so happens that there is no fusion of tense and agreement morphology in Fula. If (19b) is an option allowed by UG, (19c) should be, too, since (19c) differs from (19b) only in whether Φ follows or precedes its sister.
Which is the correct structure for Fula, then? As far as I can see, there is no evidence internal to Fula that tips the scale one way or another.10 As I will show below, each option yields a reasonable account, though adopting (19c) requires an extra assumption. This situation creates a serious learnability problem, leaving children in limbo, since they cannot settle on the correct structure. The verbal morphology of Fula thus presents a very strong argument that the theory of morphology must be tightly constrained. The simplest solution to this problem is to require that the Φ node be generated as a sister to the terminal T node as in (19a), for ruling out the hierarchical organization in (19a) still leaves the two orders (19b) and (19c), demanding an additional principle to resolve the choice. Of course, phenomena in other languages may lead us to different conclusions. In any case, I will work out the analysis adopting the postverbal placement of Φ as in (19a) and (19b) in this section. The discussion of (19c) is postponed till section 5, where the additional assumption is introduced.
(20) Local Dislocation analysis of Φin Fula subjunctive and relative tense clauses
Preposing of the [−hearer] subject marker takes place as an instance of Local Dislocation after the deletion of the predictable [−hearer] value.
Local Dislocation is defined by Embick and Noyer (2001) as a morphological operation that applies after or concomitant with Vocabulary Insertion. It necessarily follows impoverishment, since the latter comes before Vocabulary Insertion. We thus have the following ordering:
(21) Ordering of morphological operations
Impoverishment > Vocabulary Insertion ≥ Local Dislocation
This ordering ensures that Local Dislocation will pick out the correct natural class for the relative tenses. I will discuss the relation between impoverishment and Vocabulary Insertion in sections 4–5.
Another important characteristic of Local Dislocation is that it can mention specific properties of Vocabulary items. The other type of movement-like operation posited by Embick and Noyer (2001), namely, Lowering, cannot. We are justified in invoking Local Dislocation to handle the Fula phenomenon in this respect as well, since the operation must be restricted to the Φ node that contains [−hearer] in subjunctive and relative tense clauses.
[X * [Y * Z]]
[[Y ⊕ X] * Z]
[[X ⊕ Y] * Z]
In the case of (22c), there is no change in the morpheme order. But in both cases, the adjacency between X and Y is converted into adjunction of Y to X. The two cases differ in the direction of adjunction. Since it relies on adjacency, Local Dislocation cannot skip an intervening terminal. In (19a–b), however, the result needed is adjunction of Φ to V, apparently skipping the terminal T. In order to solve a similar problem in their treatment of the Lithuanian reflexive, Embick and Noyer (2001) propose that V and T undergo string-vacuous Local Dislocation first, creating a single impenetrable sublexical unit.12 Local Dislocation of Φ can then operate on the result of this earlier operation. The sequence of this process is shown in (23).
In this case, the constituency of the initial string (23a) does not matter (see footnote 11). Local Dislocation of Φ cannot skip T even when the bracketing is [[V * T] * Φ] as in (19b). Embick and Noyer’s account of Lithuanian deals with essentially the same situation, where prior to stringvacuous Local Dislocation of V and T, the reflexive appears outside the constituent made up of V and T. String-vacuous Local Dislocation of V and T is still needed to allow the reflexive to change its position around the unit consisting of V and T.
The step from (23b) to (23c) is conditioned by the variety of T. If T is subjunctive or relative, Φ must be [−hearer]. So the rule says, ‘‘Convert the adjacency between Φ and [V ⊕ T] into left-adjunction of Φ to [V ⊕ T] when Φ is [−hearer] and T is subjunctive or relative.’’ This amounts to (20). If T is of some other kind such as general past (recall (8)), there is no restriction on Φ.13
As noted above, impoverishment has been conceived in Distributed Morphology as a means of modifying the morphosyntactic representation before Vocabulary Insertion takes place. One important manifestation of such modification is syncretism. One may then wonder whether it is possible to rely on syncretism to provide evidence bearing on whether person features are binary, or on feature classes in general. As it turns out, the task is rather difficult. In this section, I will explain why. The difficulty comes from the availability of a default elsewhere form for Vocabulary Insertion.
To start with, let us consider whether we can find the pattern of syncretism made possible by deletion of the predictable value of [±hearer]. The obvious candidate that comes to mind immediately is a syncretism involving 1st person exclusive plural, 3rd person singular, and 3rd person plural in a system with the inclusive/exclusive distinction. As we saw in section 3, these three person-number combinations are the only ones that share [−hearer] after the impoverishment operation in (11). And indeed, concrete cases of such a syncretism exist. Cysouw (2003) mentions Shuswap, which exhibits the pattern we are looking for, shown in (24).14
(24) Shuswap intransitive suffixes (based on Cysouw 2003:158)
One might take this pattern as evidence that 1st person exclusive plural, 3rd person singular, and 3rd person plural are turned into a natural class by deletion of [−hearer] from 1st person singular. Things are not that simple, however.
Consider how φ-features are converted into Vocabulary items in the framework of Distributed Morphology. The process, called Vocabulary Insertion, is governed by the Subset Principle in (25).
(25) The Subset Principle (Halle 1997:428)
The phonological exponent of a Vocabulary item is inserted into a morpheme in the terminal string if the item matches all or a subset of the grammatical features specified in the terminal morpheme. Insertion does not take place if the Vocabulary item contains features not present in the morpheme. Where several Vocabulary items meet the conditions for insertion, the item matching the greatest number of features specified in the terminal morpheme must be chosen.
For the Shuswap data in (24), the following list of Vocabulary items for the Φ node suffices:
-wn aa [+speaker, +singular]
-әxo [+hearer, +singular]
-әt t [+speaker, +hearer]
-әp p [−speaker, +hearer, −singular]
-әs s elsewhere
After four person-number combinations are taken care of by (26a–d), the remaining three categories are realized by the default elsewhere form specified in (26e). Vocabulary items need not be fully specified. I have omitted the predictable feature values in (26).
Note that it is not necessary to assume an item that mentions [−hearer] to account for the pattern in (24). The class that shares [−hearer] can also be captured by an analysis in terms of the default form. Of course, once we conclude on independent grounds that the hearer feature is binary, the list in (27), which makes use of [−hearer], becomes a serious alternative, and the question of which is correct can be raised as an empirical issue.
-wn a [+speaker, +singular]
-әxo [+hearer, +singular]
-әt t [+speaker, +hearer]
-әp p [−speaker, +hearer, −singular]
-әs s [−hearer]
When we are trying to figure out whether the hearer feature is binary solely on the basis of syncretic patterns like (24), however, the very possibility of an analysis along the lines of (26) prevents us from reaching a conclusion.
Let me add that it is not necessary to invoke impoverishment to ensure that the items in (26) or (27) will be inserted properly. No further comments are needed for (26), where four out of the seven categories are unambiguously picked out by (26a–d). In the case of (27), both (27a) and (27e) mention a subset of the features that characterize 1st person singular. Crucially, however, (27a) matches more features than (27e) and will be chosen for that reason by the Subset Principle, yielding the correct result.
Thus, the existence of a single syncretism within the feature space defined by person-number combinations is not as informative as we want with regard to what counts as a natural class, insofar as a default form is available. Besides, we need to guard against accidental homophony (e.g., Bobaljik, Nevins, and Sauerland 2011, Harbour 2008).
To suppress the possibility of the default form analysis, we need to look for patterns of multiple syncretisms where an elsewhere specification is no longer available because it must be used to cover an irrelevant group of person-number categories. Continuing to focus on potential evidence bearing on the binary nature of the hearer feature, consider the pattern in (28).
There is not a single feature value that is shared by 1st person singular, 2nd person singular, and 2nd person plural. C must therefore be treated as the default elsewhere form. But then, B cannot be a second default form. One is driven to analyze B as corresponding to [−hearer]. The Vocabulary list looks like (29).
A a [+speaker, +hearer]
B b [−hearer]
C c elsewhere
I do not know whether the multiply syncretic pattern in (28) is attested or not. No corresponding example is found in Cysouw 2003. One can imagine that instances of multiple syncretisms within a single feature space that need to appeal to postulation of the default form are not so common.16 It then follows that the natural class defined by [−hearer] is not easy to come by in the empirical domain of syncretism. There are also multiply syncretic patterns other than (28) that can provide evidence for the binary nature of the hearer feature, but I leave exploration of those possibilities to more serious future research on syncretism. As I showed in section 3 and will confirm in section 5, Fula provides solid evidence for that conclusion, and that suffices for the purposes of this article.
Now, it may be interesting to ask why phenomena handled by Local Dislocation and patterns of syncretism differ in evidential value concerning the inventory of features supplied by UG. Syncretism arises from the way features are realized by phonological material. The system of such realization can contain an elsewhere specification, which can lump together heterogeneous categories that do not form a natural class.17 Thus, we need to find a way to get around the possibility of an elsewhere specification to identify natural classes defined in terms of features, as in the case of (28). Local Dislocation, on the other hand, has to mention a positively specified condition under which it applies, when picking up two adjacent terminal nodes as its targets. In other words, there is no role for elsewhere environments in the functioning of Local Dislocation. Thus, reasoning from data to a relevant theoretical conclusion is relatively straightforward when Local Dislocation is involved. Of course, the way Local Dislocation works can be affected by modifications in the feature makeup of the morphosyntactic representation introduced through impoverishment. The same is true with Vocabulary Insertion. Let us next examine more closely the nature of impoverishment, another key player in the proposed account of Fula. We will then be ready to consider what analysis is possible if the structure in (19c) turns out to be correct, as promised above.
5 Types of Impoverishment
Impoverishment has been used as a means of capturing syncretism (Bobaljik 2002, Bonet 1991, Calabrese 2011, Frampton 2002, Halle 1997, Harley 2008, Nevins 2011a, Noyer 1997, 1998). By deleting features, impoverishment prevents Vocabulary items that mention those features from appearing, thus neutralizing certain featural oppositions. An obvious theoretical question is whether there is any constraint on the working of impoverishment. As noted in section 3.1, impoverishment adds complexity to the mapping from the output of syntax to PF. There must be a limit on such costs.
Calabrese (2011) and Noyer (1998) suggest that impoverishment targets marked features. Nevins (2011a) proposes that we must recognize a new type of markedness-related impoverishment that deletes features that are not necessarily marked, when they are found in the context of marked features. He calls the former markedness-targeted and the newly added type markedness-triggered. These two types of impoverishment are formulated in the following way for a marked feature mF:
Delete mF in an environment E.
Delete [ ±G], G ≠ F, in an environment E that contains mF.
In both cases, prior specification of markedness is needed.18
In addition to markedness considerations, I am suggesting in this article that predictability of feature values from cooccurrence restrictions also lies behind impoverishment. Let us call it impoverishment of the predictable. It takes the following general form:
Impoverishment of the predictable
Delete a feature pF that is predictable in an environment E.
The analysis of Fula proposed in section 3 relies on this type of impoverishment, which deletes [−hearer] in the context of [+speaker, +singular] in the relative tenses.
It is important to notice that removal of [−hearer] from 1st person singular does not have a neutralizing effect. This is because deletion of predictable features does not lead to loss of information. A Vocabulary item specified as [+speaker, +singular] does not match any other person-number combination in (9), repeated here.
On the contrary, it has a differentiating function. Recall that impoverishment of [−hearer] is vital in removing 1st person singular from the class consisting of 1st person exclusive plural, 3rd person singular, and 3rd person plural in the analysis of Fula. This ‘‘differentiating’’ function of impoverishment is not surprising in the multidimensional space of person-number interaction. Loss of a distinction in one dimension can mean addition of a distinction in another dimension, because deletion of a feature bleeds an operation that specifically mentions it. Thus, the differentiating effect is not limited to impoverishment of the predictable. It is found with other types of impoverishment as well. Since impoverishment is often posited to account for neutralization, it is worth stressing that it has a differentiating effect as well.
To show that impoverishment of the predictable can also be used in an account of syncretism, let us consider Mam, a Mayan language. Its ergative/possessive markers display the following pattern:
(32) Mam ergative/possessive affixes (based on Noyer 1997:120)
The prefixes are easy to handle. The following list of Vocabulary items will do:
n/w- ⇔ [+speaker, +singular]
q- ⇔ [+speaker]
t- ⇔ [−speaker, +singular]
ky- ⇔ [−speaker]
The problem is how to treat the suffix. Suppose that Mam uses [±participant] in addition to [±speaker] and [±hearer]. The contexts in which the suffix -a shows up share the [+participant] feature. So let us posit (34).19
(34) -a ⇔ [+participant]
Inclusive, however, lacks this suffix. How can we prevent it from appearing? By deleting [+participant] from inclusive.
(35) Impoverishment of [+participant] in Mam
Delete [+participant] from the feature matrix of the ergative/possessive marker in the context of [+speaker, +hearer].
Notice that this operation is an instance of impoverishment of the predictable, since [+speaker] entails [+participant] and similarly for [+hearer].20
Here, the complexity-adding nature of impoverishment is most clearly seen. Impoverishment of the predictable [+participant] does not apply to 1st person exclusive or to 2nd person, even though either [+speaker] or [+hearer] alone predicts [+participant]. This indicates that application of impoverishment of the predictable is not automatic. As a cost added to the grammatical system, deletion of the predictable feature, like any other type of impoverishment, must explicitly specify the context of its application as in (35).21
Impoverishment of [+participant] to capture the pattern in Mam was originally proposed by Harley (1994), who formulated it in terms of a hierarchical geometry with privative features where rticipant] dominates [speaker], which in turn dominates the unanalyzed [inclusive]. Because of the geometry, impoverishment of rticipant] deletes the dominated features as well, as in (36).
Impoverishment of [+speaker] (or [speaker] in the privative system) creates a very serious problem, however, since the Vocabulary item (33b) must be used for inclusive, realizing [+speaker]. If [+speaker] is deleted, the correct form cannot be ensured for inclusive. The advantage of appealing to impoverishment of the predictable without invoking a geometrical organization of person features is rather clear here.
With the addition of [±participant], an alternative way of capturing a natural class in the Fula relative tenses becomes available, as pointed out by a reviewer. The idea is that the postverbal subject markers can be characterized as [+participant] once impoverishment of [+participant] is applied in the context of [+speaker, −hearer, −singular], namely, 1st person exclusive plural. Deletion of [+participant] in the context of [+speaker, −hearer, −singular] is again an instance of impoverishment of the predictable. Note further that this alternative also makes crucial reference to [−hearer], since impoverishment of [+participant] in the context of [+speaker, −singular] would erroneously delete [+participant] from 1st person inclusive as well. It is simply impossible to single out 1st person exclusive plural without mentioning [−hearer].
This treatment is exactly what is needed if (19c) turns out to be the correct structure for the verbal complex. Local Dislocation postposes the [+participant] markers in the relative tenses after impoverishment of [+participant] in the context of [+speaker, −hearer, −singular] (and string-vacuous Local Dislocation of V and T) under this alternative. In the subjunctive, the [+hearer] markers are subject to Local Dislocation.
It is interesting to observe that a more parsimonious account of Fula not relying on [±participant] goes hand in hand with the conclusion reached from the solution of the learnability problem discussed in section 3.2, which tentatively points to (19a) as the correct structure. Note that the value of [±participant] is always completely predictable from the values of [±speaker] and [±hearer]. The use of [±participant] in addition to [±speaker] and [±hearer], therefore, only provides superfluous information, as far as semantic interpretation is concerned. Using [±participant] on top of [±speaker] and [±hearer] and then deleting [+participant] in some context is doubly costly. Since it is possible to handle subject marker placement in Fula without counting on [±participant], as we saw in section 3, it is not obvious that the use of [±participant] is justifiable in Fula. The situation is quite different in Mam, where an exponent corresponding to [+participant] must be provided, which means that [+participant] has an unavoidable direct effect on the PF side.
It is an important topic for future research to figure out whether there is a principle dictating when [±participant] can additionally be employed in the person system with the inclusive/exclusive distinction. One may be able to motivate some economy considerations with regard to the parametric choice of features. If it turns out that additional use of [±participant] is illicit for Fula, that will reduce the learnability problem to the choice between (19a) and (19b), in which case we cannot say that knocking out the outer placement of Φ in the complex T head by means of a UG principle is the simplest solution. Ruling out (19a) will be equally simple. If, on the other hand, use of [±participant] is a legitimate option for Fula, the innermost generation of Φ will eliminate (19b) and (19c) with a single stroke and no further constraint will be needed.
To conclude the discussion of Fula, no matter which of the three structures in (19) is correct, the account of subject marker placement needs to mention [−hearer] in some way or another. Having too many options is not a blessing, however, from the learnability viewpoint. As far as Fula is concerned, an effort to narrow them down favors the analysis that fronts the [−hearer] markers from the postverbal position, namely, the one in (20), with either (19a) or (19b) as the correct structure. I now turn to a topic related to the parameterization of person features.
6 Original Motivations for the Privative Hearer Feature
Let us finally consider why the hearer feature has been considered privative in the past. The reasoning presented in the literature is that the privative treatment can avoid problems that would arise if the binary [±hearer] feature is assumed. Two such problems are discussed. The argument provided in this article for the binarity of the hearer feature, on the other hand, is an existential claim: there is a phenomenon that requires reference to [−hearer] if we want to give a principled account for it. Therefore, we need to assess how serious the two problems for the binary [±hearer] are. One of them concerns possible person systems in natural language, and it finds a relatively straightforward solution, as I will show. The other is highly debatable at the level of the empirical generalization related to the privativity of the hearer feature. I will point to lacunae in the discussion of it in the literature.
6.1 Possible Person Systems
Harbour (2006) notes that the partition created by [±hearer] is not found in the person systems of human languages. More specifically, he points out that Zwicky’s (1977) syou system, which is not attested, is incorrectly allowed to exist if the hearer feature is binary. The person categories in this unattested system are defined as follows:
(37) Zwicky’s (1977) syou system
The category of 2(+ 1 + 3) in the plural is termed syou by Zwicky. (38) shows possible and impossible feature sets for individual languages.
(38) Feature combinations for the possible and impossible person systems
[±participant, ±speaker, ±hearer]
(38a) is the feature set for Mam, as argued for in the previous section. (38b) characterizes the 3-person system exemplified by English. (38c) is the one for Fula. (38d) corresponds to the impossible syou system.
There is an obvious way of distinguishing between possible and impossible systems: [±speaker] must always be chosen. In other words, the presence of [±speaker] is not parameterized, unlike the other two person features. This constraint is quite natural, since language use inevitably presupposes the presence of the speaker, even in solitary thought. Thus, it is not surprising that it enjoys privileged status, as compared with the other notions.
The system with [±speaker] alone is also impossible, and one might wonder why.22 There are a couple of options. One is to say that only one feature can be omitted in parameterization out of the three provided by UG. Another is to posit the requirement that the person system should be able to refer uniquely to the hearer, though not necessarily by means of [±hearer]. I am inclined to the latter, but either way, one can place constraints on which of the available features can actually be chosen in parameterization, thereby circumventing the syou problem.
I would also like to point out that adoption of the privative [hearer] does not automatically solve the problem. Nevins (2007:305) chooses the inclusion semantics for the privative [hearer] as well as for the other binary features. In that case, substituting privative [hearer] in (37) does not block the syou system.23Harbour (2006), on the other hand, claims that the interpretation of [hearer] is identity with the hearer (his (5aiii)), while reserving the inclusion semantics for the binary features. This option evades the syou problem but creates a nontrivial question for the proper interpretation of [+speaker, hearer], intended to be the representation of inclusive, since the individual identified with the hearer does not contain the speaker by definition. I do not see how an alternative identity interpretation of the privative [hearer] can avoid this problem, either. All in all, the syou problem does not seem to differentiate between privativity and bivalence. Constraints on parametric choice of person features seem to be needed in any case.
6.2 Person-Case Effects
Nevins (2007) provides an account of possible parametric variations in person-case effects, which prohibit certain combinations of dative and accusative clitics or agreement markers. He identifies four distinct types in the case of the 3-person system.
Nevins argues for the privativity of the hearer feature on the basis of the putative descriptive generalization that 1st person inclusive and 1st person exclusive do not behave differently with respect to person-case effects, citing Chinook, Kiowa, Passamaquoddy, Warlpiri, and Yimas as 4-person languages with person-case effects that support the generalization.24 Here, I will not dwell on the precise mechanism that derives person-case effects and connects the putative generalization to the privativity of the hearer feature. I would just like to point out that no matter what data pattern emerges from inspection of just four or five 4-person languages with person-case effects, it is insufficient for establishing a sweeping typological generalization, especially with regard to the absence of a particular pattern. Given that four distinct types of person-case effects are recognized for the 3-person system, we would expect about as many types for the 4-person system, too. A very slight skew in the types of languages chosen for examination can create an accidental gap in the variety of attested person-case effects. Let me also add that Nevins does not discuss concrete examples of languages with the clusivity distinction, nor does he present the full data pattern. Thus, a more in-depth study is needed to give an accurate picture of this empirical domain. A fair conclusion to draw is that the case for the privative hearer feature argued on the basis of person-case effects is inconclusive.
From an analysis of Fula, I have provided straightforward evidence that the hearer feature is binary for the purposes of morphological operations. This conclusion still leaves open how person features behave in narrow syntax. In this connection, it is interesting to point out that Harbour (2007, 2011b) proposes that uninterpretable features come in the form of [+F, −F]. This hypothesis presupposes that both the plus and minus values of a feature are available. Harbour’s concern is with number, but if his proposal is taken to be a general claim about uninterpretable features, as it should be, the person features are expected to behave in the same way. This means bivalence of person features in syntactic agreement. Furthermore, as Adger and Svenonius (2011) point out, if agreement involves valuation, as suggested by Chomsky (2001), it is incompatible with the privative system, because valuation implies two different states of a single feature. Of course, the working of valuation differs depending on whether uninterpretable features are simply valueless as in Chomsky’s original formulation or take the form of [+F, −F] as in Harbour’s new proposal. See Harbour 2011b for one precise implementation under the new conception. See Watanabe 2012 for exploration of an alternative possibility in making use of [+F, −F]. Note that Nevins’s (2011b) claim that bivalence of [±singular] should be limited to morphology, mentioned in section 1, is at odds with these general considerations about agreement. How this conflict can be resolved is a topic for future research.
The proposed analysis of Fula counts on Local Dislocation to handle mixed placement of subject agreement markers. Without Local Dislocation, some markers would be haphazardly treated as prefixes, others as suffixes, as in Stump’s (2001) account, which misses an important regularity hidden in the grammatical system of Fula. Thus, the proposed account also lends strong support to the theory of morphology that posits that Local Dislocation is a postsyntactic operation. A novel aspect of the analysis is that a natural class made possible by impoverishment defines the environment in which Local Dislocation applies. Impoverishment has mainly been used to capture syncretism in the past, but the present study has shown that it has further potential in the area of mixed placement of agreement markers. Though this type of phenomenon is not common according to Cysouw (2003:11), its evidential value concerning the feature system is rather direct, compared with that of syncretism, which may involve an elsewhere form used for the residue that does not necessarily form a natural class. Given the additional possibility of accidental homophony (e.g., Bobaljik, Nevins, and Sauerland 2011, Harbour 2008), data concerning syncretism require extreme caution in handling, especially when multiple features interact. Mixed placement of agreement markers is free from such complications.
The proposed new type of impoverishment operation is motivated by cooccurrence restrictions in the interaction of person features with number features, thereby falling squarely within the research agenda that tries to ground impoverishment in the logical properties of the feature system. Markedness considerations have already been discussed in the literature. The current study points to predictability due to cooccurrence restrictions as another significant factor. Given the role that markedness and redundancy have played in the phonological feature system (Dresher 2009), it is not surprising to find that a similar idea is useful in the domain of morphology. Adding impoverishment of the predictable completes the picture in this respect.
The core part of the analysis of Fula was presented as a poster at the eighth Mediterranean Morphology Meeting (Cagliari, 2011). I would like to thank the participants for their interest in my work. Comments and suggestions from two anonymous reviewers as well as from Chizuru Nakao and Uli Sauerland have also helped me hammer out the final version. The research reported in this article was supported by Grant-in-Aid for Scientific Research (C) 22520492 from the Japan Society for the Promotion of Science.
1 Harley and Ritter use [addressee] instead of [hearer]. Since the name itself does not matter, I will use [hearer] throughout for the sake of consistency.
2Harbour (2011b) posits a version of privativity, called plus~minus privativity, that employs both the plus and minus values, as distinct from presence~absence privativity, which, as the name indicates, cares only about the presence or absence of a feature. Plus~minus privativity, which Harbour eventually argues against, counts as one type of binarity in the terminology adopted here.
3 For languages without the clusivity distinction, Harley and Ritter (2002) claim that [speaker], as the default person, need not be represented explicitly. They take this status to be responsible for early acquisition of 1st person pronouns, but since in their account [speaker] is not treated as the default person in languages that distinguish inclusive from exclusive, the link to the acquisition process is not so obvious: by the time children have figured out whether or not the target language makes the clusivity distinction, they must already have acquired the entire 1st person pronoun system. I will comment on a different respect in which the speaker feature is special in a way relevant to language acquisition in section 6.
4 The same problem arises in the privative system.
5 Arnott treats the 1+2(+3) category as 2nd person inclusive on the grounds that it has more in common with 2nd person than with 1st person. Such a terminological decision has no status in the discussion of more fundamental person features. I will stay with the traditional labels.
6 See Prunet 1992 for one approach to decomposing these endings in a slightly different dialect into separate voice and tense/aspect morphemes. I will abstract away from the morphophonological processes that create the allomorphy in each of the categories in (14). I will also refrain from exploring the nature of the alternation of the [+hearer] subject markers such as -aa ~ -ɗaa, except to note that the forms with /ɗ/ are used in all the voice types in the relative tenses and the subjunctive other than relative future active and subjunctive active.
7 In the subsequent discussion, I will continue to talk about the tense/aspect information encoded in the endings in (14) and use the categories labeled by Arnott, even if McIntosh’s analysis turns out to be on the right track. Marking of subjunctive and relative should be on T regardless of the treatment of tense categories themselves.
8 In the subjunctive, an independent particle to is placed before the subject, as in (i), where I omit the accent mark from the negative particle.
To Bello nasta.
neg Bello come.in
‘Bello is not to come in.’
This negative particle is placed in the C-domain. Note also that no subject marker appears when the subject is not a pronoun. Only pronominal subjects trigger overt agreement in Fula (Arnott 1970:144).
9 If it turns out that the verb is not raised to T in Fula despite the evidence given in (17) and (18), the argument for postverbal placement of subject agreement markers is pretty straightforward, since the T node must be lowered to the verb, creating the structure [V0 V0 + T0] (see Embick and Noyer 2001:561). The Φ node must be generated under T0, given that it carries the agreement features. Apart from the fact that Embick and Noyer regard Lowering as a morphological operation, it has essentially the same structural properties as in Chomsky’s (1995:chap. 2) account, applying to syntactically separate heads.
10 Potentially relevant is the fact that the vowel of the preterite element is shortened when the postverbal subject marker has a long vowel as in (ia), where the relative future active is used again for illustration.
‘you(sg/pl) were going to help’
In other words, there is a trade-off between the preterite element and the subject marker with regard to vowel length. I am not sure about the weight carried by this phenomenon, though.
11 Strictly speaking, ⊕ indicates adjunction of a sublexical terminal embedded in a larger head to another sublexical terminal. Since the analysis of Fula involves such sublexical terminal elements, I limit the discussion to these cases of Local Dislocation.
Incidentally, the other variety of Local Dislocation allowed in Embick and Noyer’s (2001) theory involves maximal head-level constituents. It is not allowed to refer to intermediate constituents within a complex head in either variety of Local Dislocation. If it turns out in future research to be necessary to modify the theory of Local Dislocation in such a way as to allow reference to an intermediate constituent, however, (20) remains intact.
12 If V contains further structural details such as a light verb and an Aspect head, these nodes must also undergo Local Dislocation cyclically.
13 The sensitivity to the tense distinction may provide one consideration for ruling out the possibility that the original position of Φ is preverbal in Fula. The idea is that derivation of postverbal agreement markers from [Φ * [V ⊕ T]] by Local Dislocation must refer illicitly to the nonadjacent tense information. I am not aware of other cases where this condition is crucial in the application of sublexical Local Dislocation. Future research should address this question seriously. I would like to thank a reviewer for bringing my attention to the issue.
14 Independent pronouns in Shuswap make a singular/plural distinction in 3rd person, unlike agreement suffixes. See Déchaine and Wiltschko 2002.
15 Even the pattern in (28) becomes inconclusive once [±participant] is added on top of [±speaker] and [±hearer]. C can be a realization of [+participant] once 1st person exclusive plural is stripped of [+participant] by omission of predictable features. Then, B will be able to emerge as the default form. See section 5 for an analysis of Mam that employs [±participant] in addition to [±speaker] and [±hearer].
16 This conjecture is probably supported by the figures provided in Pertsova’s (2011) study of the logical types of syncretism.
18 See Harbour 2011a for the care that one must take in dealing with the notion of markedness.
19 Why [±speaker] (together with a number feature) is realized as a prefix whereas [+participant] appears as a suffix is an interesting question that needs to be addressed in future research, though orthogonal to the present purposes.
20Noyer (1997) thought about using [+participant], but fell just short of invoking impoverishment in the fashion discussed here, discarding the possibility. Instead, he appealed to the α notation, claiming that the suffix -a appears in the contexts of [αspeaker, −αhearer], where α ranges over + and −. No such device is needed in the account proposed in the text.
21 A reviewer wonders whether an elsewhere impoverishment rule is possible. An elsewhere item for Vocabulary Insertion can be defined because relevant terminal nodes are already given. If the same is true for impoverishment, an elsewhere specification of the context for its application is conceivable when multiple deletion rules target a single node, though a concrete example is yet to be found. Local Dislocation is different, since its operation simply picks up two adjacent items.
22 It is well-known that sign languages in general only distinguish between 1st person and non-1st person (see Lillo-Martin and Meier 2011 for a recent overview). But this is only a matter of morphology. Semantic interpretation needs to be able to single out the hearer from the rest, as Aronoff and Padden (2011) point out. It is an open question whether narrow syntax is insensitive to distinctions other than that between 1st person and non-1st person.
23 A strictly literal interpretation in fact creates a slightly different problem, since there is no way of unambiguously designating the speaker in the system with [±participant] and [hearer]. Invoking pragmatic considerations to avoid this problem allows the syou system to emerge.
24 Yimas should not be included here, since it does not distinguish between inclusive and exclusive (Foley 1991).