Anaphoric Binding in Colloquial Sinhala1

Kumara Henadeerage
Department of Linguistics (The Faculties)
Australian National University
Canberra, Australia.
Proceedings of the LFG98 Conference
The University of Queensland, Brisbane
Miriam Butt and Tracy Holloway King
CSLI Publications

1. Introduction

In this paper, I examine the properties of reflexive binding in Sinhala (Indo-Aryan: Sri Lanka), in particular those related to the reflexive taman, the verbal reflexive, and the pronoun eyaa [FN2]. Sinhala data illustrates that reflexive pronouns that are normally required to be coreferential with an argument in the local domain may be disjoint, and personal pronouns may be bound to an argument in the local domain that they are normally required to be disjoint from. Anaphoric binding relations are conditioned by a number of different factors, one of which is the verb, which may be marked for reflexivisation and may determine the binding domain of reflexives and pronouns which occur within its nucleus. The data also provides evidence that a single anaphor can have more than one binding domain. Therefore it is not possible to define a particular anaphor as "long distance" or "clause-bound". Further the relation between argument structure and reflexive binding is evident in the interaction between the verbal reflexive, the personal and reflexive pronouns.

In this paper, I do not intend to provide a formal description based on Functional Uncertainly, however, in section 3 I will discuss some problems associated with anaphoric binding equations based on Functional Uncertainty (Dalrymple 1993), with respect to Sinhala data. First, I will survey binding relations of the reflexive pronoun in simple clauses and multi-clause constructions (i.e those with finite or non-finite subordinate clauses or infinitive clauses).

2. Anaphoric Binding Patterns

There are two ways to express the reflexive in Sinhala: one involves the use of the reflexive pronoun taman "self" (or the reciprocal tamåtaman "each other", the reduplicated form of taman) and, the other involves adding the verb gannåwa "take" to the perfect participle form of the verb. They both can occur individually or together in the same clause, in which case the gannåwa verb conditions the binding domain of taman.

The Sinhala has a pronominal system consisting of a large number of pronouns with distinct forms for humans, animals and inanimate objects. Among them, the most commonly used ones include: mamå "I", api "we", eyaa "he/she", eeka "it" (non-human/derogatory) and ekå "it" (inanimate). In this paper, I will restrict the discussion to the interaction between the reflexive verb and taman and the reflexive verb and the pronoun eyaa.

2.1 Reflexive taman

Taman "self" is the reflexive pronoun in Sinhala (Gair et al 1989, Gunasekara 1891, Inman 1994)., Taman inflects for case, just like any other noun/pronoun (see Figure 1) [FN3].
Figure 1: Case Inflections for Nouns/Pronouns
Case         Reflexive  Reciprocal     1SG "I"         "child"
Nominative   taman-ø    tamåtaman-ø    mamå-ø/mang-ø   lamåya-ø 
Accusative   taman-wå   tamåtaman-wå   maawå           lamåya-wå 
Dative       taman-tå   tamåtaman-tå   matå            lamåya-tå 
Instru./Abla.taman-gen  tamåtaman-gen  magen           lamåya-gen
Genitive     taman-ge   tamåtaman-ge   mage            lamayå-ge

Reflexive pronouns can be used as arguments (for examples see section 2.2) and possessors: see (1). Taman is inheretly 3rd person but unspecified for number and gender. Therefore it cannot be coreferential with the first and second person pronouns (see (2.a)). In (2.b), taman is not acceptable, since the antecedent is a first person pronoun.

(1) Nimal1 Siri-wå2   taman-ge1,*2   gedårå-di  dækka. 
    Nimal  Siri-ACC   self-GEN     house-at   see.PST
   "Nimal1 saw Siri2 at his1,*2 house."   (Inman 1994:51)
    [See footnote 4 for abbreviations]
(2)a. lamåya1/minissu2/ohu3/æya4/*api5    taman-ge1,2,3,4,*5 --
      child/man.PL/3SG.MASC/3SG.FEM/*1PL self-GEN        --
       -- wædak       balaa_gatta,       nam  hondai.
       -- work.INDEF  look.PP_take.PST,  if   good
      "It is better if the child1/men2/he3/she4/*we5 minded 
       self"s1,2,3,4,*5 business."
   b. api  ape/*taman-ge      wædak       balaa_gatta,      nam  hondai.
      1PL  1PL.GEN/*self-GEN  work.INDEF  look.PP_take.PST, if   good 
      "It is better if we minded our own/*self"s business."

2.2 Reflexive Verb gannåwa

When the verb gannåwa "take" is suffixed to the perfect participle form (PP) of a verb the result is a sentence with a reflexive meaning (Gair 1970:123, Gunasekara 1891:180, Reynolds 1980:185). In other words, the perfect participle of the verb suffixed with gannåwa has grammaticalised as a reflexive auxiliary (Herring 1989:179). Henceforth, these verbs will be referred to as "reflexive verbs" [FN5]. This type of reflexive is very common both in literary and colloquial Sinhala: see sentence (3-4).
(3)  amma   kaemak      hadaa_gannåwa.   
     mother meal.INDEF  make.PP_take.PRE
     "The mother is making herself a meal."

(4)  lamayi    gaha_gatta.
     children  strike.PP_take.PST
     "Children fought/ Children hit one another."
Reflexive taman may appear as an argument or in a non-argument position in a clause subcategorised by a reflexive verb. Sentence (5), which is similar to (3-4), contains a reflexive verb and the reflexive taman, which is redundant and not obligatory as can be seen in (3-4). Even without an overt reflexive pronoun, the sentence has the same meaning as it does when taman is present. In fact, sentences without an overt reflexive pronoun are preferred over those with its presence in casual conversations. In (5) the reflexive verb can not have another object which is not coreferential with the antecedent. It is because of this that the meaning "Siri hurt Gune" is not available in (5).
(5)  Sriya  (taman-wå)/*Gune-wå   tuwaalå _karå_gatta.
     Sriya  (self-ACC)/*Gune-ACC  injury_make.PP_take.PST
     "Sriya hurt herself/*Gune."
Further, note that taman is highly disfavoured in argument positions of simple sentences which do not have reflexive verbs. Compare the following sentence with (5):
(6)  Sriya  Gune-wå/*taman-wå   tuwaalå_kåla. 
     Sriya  Gune-ACC/*self-ACC  injury_make.PST 
     "Sriya hurt Gune/*herself."
Sentence (6) does not have gannåwa affixed to its verb and therefore does not have a reflexive sense. Note that the verb being non-reflexive, the sentence is not acceptable with the reflexive pronoun taman-wå .

What seems to happen is that the reflexive verb enforces the coreferential effect between its subject and an NP which occur within its nucleus. Therefore there must be an NP that can be coreferential with the subject. On the other hand, with non-reflexive verbs, taman not being coindexed with the subject, becomes highly disfavoured or unacceptable in simple sentences like (6). However, as in sentence (1) taman as a possessor is acceptable in simple sentences without a reflexive verb. I will discuss this issue in detail in section 2.5, and will show that taman which appears as an argument of a non-reflexive verb cannot bind with its subject and is required to be bound by a long distance antecedent. It is because of this reason that taman- is not acceptable in (6). In the following section, I will discuss the long distance binding and the interaction between taman and the reflexive verb in multi-clause constructions.

2.3 Long Distance Binding

The binding domain of the reflexive pronoun depends on: a) whether the verb is reflexive, b) the type of clause where the reflexive pronoun appears, ie. a simple or multi-clause construction, and c) the person of the potential antecedent.

When taman occurs in a sentence with a reflexive verb, the reflexive is always bound in the Minimal Complete Nucleus (MCN), the domain containing the reflexive, the antecedent and the reflexive verb (for instance see (7.a) and (8.a)). However, when the reflexive taman appears in a subordinate clause containing a non-reflexive verb, it is coreferential with the highest argument of the matrix clause which may or may not contain a reflexive verb: see (7.b) and (8.b).

(7)a. Piyal1  kiwwa   [Sriya2  taman-wå*1,2 --
      Piyal   say.PST [Sriya   self-ACC    --
        -- tuwaalå_kårå_gatta]       kiyåla.
        -- injury_make.PP_take.PST]  that
      "Piyal1 said that Sriya2 hurt herself2/him*1,*3."

   b. Piyal1  kiwwa    [Sriya2  taman-wå1,*2 --
      Piyal   say.PST  [Sriya   self-ACC    --
        -- tuwaalå_kåla  ]  kiyåla.
        -- injury_make.PST] that
      "Piyal1 said that Sriya hurt him1/*herself."

(8) a. Daya1 Piyal-tå2  [taman-ge1 kaarekå  --
       Daya  Piyal-DAT  [self-GEN  car.DEF  --
        -- hodå-gannå-tå]    (kiyåla)  kiwwa.
        -- wash.PP.take-INF] (that)    tell.PST
       "Daya1 told Piyal2 to wash his2/*her1 car."

    b. Daya1 Piyal-tå2  [taman-ge2  kaarekå --
       Daya  Piyal-DAT  [self-GEN   car.DEF --
        -- hodånnå-tå]  (kiyåla)  kiwwa.
        -- wash-INF]    (that)    tell.PST
       "Daya1 told Piyal2 to wash her1/*his2 car."
The embedded clause of (7.a) is parallel to sentence (5). The reflexive is bound within the embedded clause which is the domain containing the reflexive pronoun, the antecedent and the reflexive verb. Now compare the embedded clause of (7.b) and the sentence (6) which is not acceptable with taman. Taman can occur in the embedded clause in (7.b) without a reflexive verb. Further, as shown by coindexation, taman is coreferential with the highest argument in the matrix clause, but not with that of the lower clause. In (6), since the non-reflexive verb does not coindex its reflexive pronoun with the antecedent Sriya, they can not be coreferential, and therefore the sentence is ungrammatical because of the reflexive not having an antecedent to be bound by. Similarly, in the case of (7.b), the absence of a reflexive verb results in non-coreference between taman and the subject in the minimal domain, hence taman is disjoint from the lower subject Sriya and corefers with the higher subject Piyal. Thus, the evidence shows that taman requires the presence of a reflexive verb if it is to be bound in the minimal domain or else it must find a long distance antecedent. Sentence (8), which has an infinitival phrase, also illustrates this fact.

Now let's consider sentences with more than one embedded clause. Taman in such subordinate clauses corefers with the subject in the highest matrix clause, unless there is a reflexive verb in the clause in which taman appears. (9.a) has two subordinate clauses and the reflexive taman appears in the lowest clause, yet it binds with the subject in the highest clause. In (9.b), the reflexive appears in a relative clause.

(9)a. Gune1 hituwa,    [Wimåle2 lamåya-tå3 kiwwa    kiyåla --
      Gune  think.PST, [Wimale  child-DAT  say.PST  that    --
       -- [Siri4 taman-wå1 wiwecåne_kålaa   ] ] kiyåla.
       -- [Siri  self-ACC  criticise_do.PST ] ] that

      "Gune1 thought that [Wimale2 told the child3 that 
      [Siri4 criticised him1 ] ]."
   b. John1 kiwwa   [Wimåle2  dannåwa   kiyåla [ [ø taman-tå1 --
      John  say.PST [Wimale   know.PRE  that   [ [  self-DAT  --

     -- gahapu]3 miniha3 taman-ge3 wædå kårå_gattå    ]2 ]1 bawå.
     -- hit.PP]  man     self-GEN  work do.PP_take.PST]]   that.

    "John1 said that [ Wimale2 knows that [the man3 
    [ who hit self1 ]1 did his3 own work]2 ]3."

Note that clause "2" (see superscript following the bracket in 9.b) contains a reflexive verb, therefore, taman-ge "self"s" corefers with miniha "man". The reflexive taman-in clause "3" cannot bind with its null subject in the minimal domain, as the non-reflexive verb gahapu "hit.PP" does not allow a such coreference between its reflexive and the subject. Therefore, the reflexive pronoun has to find a long distance antecedent. Even though the clause "2" contains a reflexive verb, the reflexive taman-tå (in clause "3") is disjoint from miniha "man" and corefers with John, the matrix subject. The reason for disjointedness between taman-tå and the clause "2" subject, miniha is that the reflexive is not an argument of the reflexive verb in clause "2", hence the reflexive verb has no effect on the reflexive pronoun. Further this example indicates that a reflexive pronoun in a lower clause can pass through higher clauses not being blocked by reflexive verbs in such clauses, to be bound with the highest subject, and the reflexive verb only indicates coindexation between its subject argument and the reflexive pronoun which too should be a co-argument of the same verb [FN6].

The reflexive pronoun cannot bind with the 1st and 2nd person pronouns as its antecedent, as mentioned in section (2.1): (see sentence (2.a-b), for example). However, some informants find 2nd person pronouns marginally acceptable. As shown earlier, the reflexive in (10.a) binds with the matrix subject. Now consider (10.b), in which the 1st person pronoun in the matrix subject position cannot be the antecedent. Therefore, taman is disjoint from the matrix subject and corefers with the lower subject, as there is no other binder available for it. This does not in any case indicate that an object can be a potential antecedent for taman. Siri, the matrix object in this case controls the XCOMP subject position, hence it is a potential antecedent. This is further evident in sentence (c), which is ungrammatical, since it has no antecedent for taman. If an object could be a potential antecedent, we would expect taman to corefer with Siri-tå. Note that (c) does not have a discourse antecedent.

(10)a. Wimåle1   Siri-tå2  kiwwa   --
       Wimale   Siri-DAT  say.PST --

        --[ø2 taman1,*2,*3 gænå  John-tå3  kiyannå] kiyåla.  
        --[   self       about John-DAT  say.INF] that   

       "Wimale1 told Siri22 to tell John3 about self1,*2,*3]."
    b. api1  Siri-tå2  kiwwa   --
       1PL   Siri-DAT  say.PST --

        --[ø2 taman*1,2,*3 gænå  John-tå3 kiyannå] kiyåla.  
        --[   self       about John-DAT say.INF] that  

       "We1 told Siri22 to tell John3 about self*1,2,*3]."

    c.*api1  Siri-tå2  taman*1,*2  gænå   kiwwa.
       1PL   Siri-DAT  self      about  say.PST 
       ¬ 'We told Siri about himself."
       ¬ 'We told Siri about ourselves."
Sentence (10.b) does not indicate any inconsistency in binding relations of reflexive taman, as it does not bind with any subject (or any argument), when a matrix subject (or a lower subject) is not available. Although there are three arguments in (11), non of them can be an antecedent for taman-tå. The matrix subject is not a binder in this case, as it is a first person pronoun. The lower subject, Siri-tå , too, cannot be an antecedent for the reflexive (unlike in (10.b)), since the coreference between an argument and subject is acceptable when they occur within the nucleus of a reflexive verb. In other words, the reflexive verb has to licence coindexation between its subject and an argument. In the case of (10.b), there was no need for a reflexive verb, since the binding involved between the subject and a non-argument.
(11) *api1 Siri-tå2  kiwwa   -- 
      1PL  Siri-DAT  say.PST --

       --[ø2 taman-tå*1,*2,*3 John3 gænå  kiyannå] kiyåla. 
       --[   self.DAT       John  about say.INF] that 

     *"We1 told Siri22 to tell self*1,*2,*3 about John3]."
To sum up, with the presence of a reflexive verb, the reflexive should bind with the subject of the reflexive verb, otherwise, if in multi-clause construction with the absence of a reflexive verb, taman should bind with the highest matrix subject. Only if the highest matrix subject is a first or second person pronoun, the reflexive binds with the next lower subject. It is to be noted that taman cannot have a discourse antecedent (see section 2.4). Its binding relations are not based on logophoric principles and can, therefore, be accounted for by syntactic constraints alone.

2.4 Logophoricity and Taman

We discussed sentences containing matrix verbs of saying and thinking, etc. in which taman"s antecedent is the one whose views, feelings etc. are reported in subordinate clauses. As Clements (1975) discusses, in some languages logophoric pronouns are used to refer to antecedents "..whose speech, thoughts or feelings are reported or reflected..." (p. 141). Thus, when a long distance pronoun binds with a noun functioning as source of reported discourse, it is said to be in a logophoric relationship with the antecedent. The source of communication is only one of a number of semantic factors distinguishing logophoric binding from reflexive binding. Following sentences illustrate that taman is not a logophor although it shares some logophoric properties.

Long distance reflexive binding is a productive construction and not restricted to matrix clauses involving verbs such as say, think, feel, blame etc. Any type of verbs seem to appear in matrix clauses and subordinate clauses, and the subject of the matrix clause, regardless of its semantics/logophoric properties becomes the antecedent of taman. Sentence (12) illustrates this:

(12) guruwåråya1 taman-tå*1,2 danduwam_kålå     nisaa, -- 
     teacher     self-DAT    punishment_do.PP  since, --

      -- lamåya2  iskoole-tå  giye    naene.
      -- child    school-DAT  go.PST  NEG

     "The child1 did not go the school, since the teacher2 
      punished self1,*2."
Logophors can sometimes bind with a discourse antecedent, as Clements points out. However this is not possible with taman. Consider (13), in which the context is given in such a away that one would expect taman to be bound with an argument in the previous sentence. However, taman cannot corefer with lamåya, but the personal pronoun eyaa "she/he" can be coreferential with lamåya "child" (or with a discourse antecedent). Note also that the reflexive cannot be coreferential with amma "mother", because "the giver" and "the givee" of "give" cannot be the same. Hence the second sentence is ungrammatical, since the reflexive pronoun does not have a binder in the same sentence.
(13) lamåya1 badåginii kiyåla amma-tå2    kiwwa.   *Amma2  --
     child   hungry    that   mother-DAT  say.PST.  mother --

      -- eyaa-tå1,3/taman-tå*1,*2 kaemak      hadåla_dunna.
      -- 3SG-dat/self-DAT       meal.INDEF  make.PP_give.PST

    "The child1 told the mother2 that (he1) was hungry. 
     The mother2 prepared a meal and gave it to him1,3/self*1,*2."
Thus, long distance binding relations of taman in non-reflexive clauses are not based on logophoric principles.

2.5 Reflexives in Simple Clauses and Arguments vs. Non-Arguments

It was mentioned in section (2.2) that taman cannot appear in simple sentences without a reflexive verb, and the sentence (6) was given as evidence for this. (Here I use the term "simple sentence" to refer to those without embedded clauses). However, some forms of the reflexive pronoun can appear in sentences without the presence of a reflexive verb: for instance, the possessive form taman-ge "self"s" as in (1). Although informants" judgements on the acceptability of reflexive taman in sentences with non-reflexive verbs are somewhat variable, some interesting observations about the occurrences of the reflexive pronoun in simple sentences have been made. There appear to be restrictions on reflexive pronouns in simple sentences with non-reflexive verbs, and on nouns and personal pronouns which occur as arguments of a reflexive verb. In other words, acceptability or unacceptability of a given simple sentence depends on two factors: (a) presence of absence of a reflexive verb, and (b) whether the reflexive pronoun is an argument or a non-argument. Below I will present a number of sentences to illustrate that reflexives in non-argument positions (ie. possessives and complements of postpositional phrases) are acceptable when occurring with non-reflexive verbs (see (1)), while reflexives in argument positions are strongly disfavoured or unacceptable with non-reflexive verbs: see (6). Sentence (14) presents the same evidence:
(14)a. Wimåle1 Siri-tå   taman1 gænå  kiwwa/*kiyaa_gatta.
       Wimale  Siri-DAT  self   about say.PST/*say.PP_take.PST 
       "Wimale1 told Siri about himself1."

    b. Wimåle  taman-tå  Siri gænå  kiyaa_gatta/*kiwwa.
       Wimale  self-DAT  Siri about say.PP_take.PST/*say.PST 
       "Wimale muttered himself about Siri."
The reflexive verb in (14.a) is not acceptable with the reflexive pronoun, since taman appears in a non-argument position. However, the sentence is acceptable with the non-reflexive verb and taman in the PP corefers with Wimåle. On the other hand, when taman appears in the object position (14.b), the verb must be reflexive. In a non-argument position of a reflexive verb, any binding or non-binding NP, including taman as possessors and complements of PPs, can occur, as there are no restrictions from the reflexive verb. This is shown in (15). The reflexive occurs as the complements of PPs with a non-reflexive verb in (15.a), while in the same position a proper noun is allowed under a reflexive verb in (b). ((b) would have the same meaning without the reflexive verb, although with it (b) implies that the child, the chair and Siri were all close to each other).
(15)a. lamåya1 taman1 langin     putuwå    tibba.
       child   self   near.INST  chair.DEF keep.PST
       "The child1 kept the chair close to him1/*2."

    b. lamåya  Siri langin     putuwå    tiyaa_gatta.
       child   Siri near.INST  chair.DEF keep.PST_take.PST
       "The child kept the chair close to Siri."
The acceptability of (15.b) may be due to lexical properties of verbs, as not all verbs behave similarly. It is observed that those verbal stems (ie. tibba "keep.PST" in (15)) that allow taman in a sentence under a non-reflexive verb also allow non-binding nouns in the same position under a reflexive verb. On the basis of examples considered so far, we observed that the reflexive verb constrains the binding domain of the reflexive and enforces some restrictions as to what can occur within its nucleus. Further, reflexive possessors and reflexive complements of postpositions are less restricted in terms of their distribution as they can freely appear as non-arguments as well as arguments in a sentence regardless of the presence or the absence of a reflexive verb.

2.6 Pronoun eyaa

Reflexive verbs constrain not only the binding domain of taman but also what can be the antecedent of a pronoun [FN7]. A pronoun can be coreferential with the subject in a minimal binding nucleus if a reflexive verb is present (16.b). In fact, taman is rarely used in such sentences and instead pronouns are used in the place of taman to give reflexive meaning. The pronoun eya "she/he" cannot be coreferential with any argument in (16a). However, in (16.b) the presence of a reflexive verb licenses the pronoun to be coreferential with the subject. A non-pronominal like Piyal in the same position must be disjoint in reference (16.b). Thus, reflexive verbs treat pronouns as though they are anaphors.
(16)a. lamåya1 eyaa-wå*1,2/Piyal-wå  tuwaalå_kåla.
       child   3SG-ACC/Piyal-ACC    injury_make.PST
       "The child1 injured him2/her2/Piyal/*himself/*herself".

    b. lamåya1 eyaa-wå1,*2/*Piyal-wå  tuwåalå_kårå_gatta.
       child   3SG-ACC/Piyal-ACC     injury_make.PP_take.PST
       "The child1 injured himself1/herself1/*Piyal".

(17)a. lamayi1  eegollan1/*minissu2  atåre salli bedaa_gatta. 
       children 3PL/*man.PL         among money divide.PP_take.PST
       "Children1 shared the money among themselves1/*men2"

    b. lamayi1   eegollan*1,2/minissu  atåre  salli  beduwa. 
       children  2PL/man.PL           among  money  divide.PST
       "Children1 distributed the money among them*1,2/men."
The coreferentiality between eegollan (plural of eyaa) and the antecedent is clearly shown by the ungrammaticality in the sentence when eegollan is replaced by a non-coreferential NP like minissu "men" (17.a). When gatta does not appear in the verb, the pronouns must be free (17.b).

2.7 Summary

If taman appears in a clause containing a reflexive verb, it should bind with the local subject. Otherwise taman must corefer with a long distance antecedent. Personal pronouns, too, cannot be free within the local domain and should bind with the local subject when occurring with a reflexive verb. The antecedent of taman is always the subject. When the highest subject is a 1st or 2nd person pronoun, the next lower subject appears to antecede the anaphor in multi-clause constructions. In simple clauses, if the subject is a 1st or 2nd person pronoun the sentence simply becomes unacceptable. It is to be noted that the difference between personal pronouns and taman is that taman should have a binder (ie. a long distance or local antecedent) within the same sentence, while pronouns need not have a binder within the same sentence. Unlike taman, they can be free in a sentence unless they occur within the nucleus of a reflexive verb. Further we observed reflexive possessives and complements of postpositions behave differently from those reflexives in argument positions of a sentence. Taman in an argument position is restricted and generally acceptable with the presence of a reflexive verb, whereas possessors and postpositional complements are less restricted and may appear with non-reflexive verbs.


3. Towards A Solution

Pronouns and the reflexive are not in complementary distribution. Sinhala anaphora does not obey the binding principles of Government and Binding Theory (Chomsky 1981) since pronouns may bind locally as well non-locally (contrary to Principle B) and reflexives can bind non-locally as well locally (contrary to Principle A). In what follows we will discuss some problems posed by the unusual characteristics of taman for a description based on Inside-Out Functional Uncertainty (Dalrymple 1993). At the end, it should be clear that the description of Sinhala data calls for an alternative framework like Optimality Theory (Prince & Smolensky 1993).

3.1 Inside-Out Functional Uncertainty

"Functional Uncertainty" (FU) can be used to express the relationship between an anaphor and its antecedent in F-structure, one of two levels of syntactic representation in Lexical Functional Grammar. The concept, called "Outside-In FU", was first introduced by Kaplan et al (1987) for analysis of topicalisation. For example, for (18)

(18) "Mary, John claimed that [Bill telephoned yesterday]"COMP (Kaplan et al p. 5)

the Outside-In FU equation is ( | TOPIC) = ( | COMP OBJ). Here the ( | TOPIC), Mary is identified with more embedded  ( | COMP OBJ), the object of the complement, hence the name "Outside-in" FU. Dalrymple proposes "Inside-Out FU', a slight variation of Outside-In FU, for stating the relation between an anaphor and an antecedent within the F-structure. The above Outside-In FU equation can be stated with Inside-Out FU, as ((COMP OBJ | ) TOPIC). Now let"s look at a sentence with a reflexive: "Sriya hurt herself". The simplified F-structure for this sentence is given in (19):

(19)     F-structure for "Sriya hurt herself":

        __                                          __
       |                                              |
       |  PRED "hurt" <(|SUBJ) (| OBJ)>"              |
       |  SUBJ [PRED "Sriya"]                         |
       |  OBJ  [PRED "self"]                          |
       |__                                          __|
(20)     (( OBJ | ) SUBJ )s = |s

(21)     (( DomainPath | ) AntecedentPath )s = the antecedents

The binding equation relating to this sentence is given in (20), which is based on more general anaphoric binding equation (21). The anaphoric binding equation tells us what the anaphor is, what the antecedent is, and the path linking the anaphor and the antecedent. The up arrow "|" following the DomainPath is the anaphor. DomainPath is the path leading to the anaphor, which is given relative to the antecedent. For example in (20), OBJ is the path and the "|" following it is the OBJ"s value, the anaphor (referring to "self" here). The AntecedentPath is SUBJ, which is "Sriya". The constraint equation "s " indicates that the anaphor and the antecedent have the same semantic content.

3.2 Analysis of Sinhala Anaphoric Binding

Binding patterns of Sinhala personal and reflexive pronouns are given in Figure (2) below:

Figure 2: Binding Patterns of taman and eyaa

With non-reflexive verbs
With reflexive verbs
disjoint from coarguments bound with the subject in MCN Pronoun  
bound with the most distant 3rd person subject bound with the subject in MCN Reflexive 

It is not possible to propose a single domain for the pronoun and the reflexive as local or long distance, since they have more than one binding domain depending on the presence or absence of a reflexive verb. If a reflexive verb is present, which may be in finite or non-finite form, both types of pronouns bind with the subject in the minimal complete nucleus MCN: the domain containing the predicate, the subject and other arguments. If a reflexive verb is not present in the clause in which taman occurs, the reflexive should bind with the highest subject in the sentence. If the highest subject is a 1st person or 2nd person pronoun, then the reflexive binds with the next lower subject. Subordinate clauses may have non-finite (ie. infinitives) or finite verbs (i.e Ss separated by the complementiser kiyåla "that"). Pronouns are free in the local domain with the absence of a reflexive verb. As it seems rather complicated to have anaphors with more than one binding domain for each one, we can simplify binding domains as in (22).


Sinhala anaphor poses a number of problems for a solution based on FU. Dalrymple assumes that each anaphoric element has an anaphoric binding equation lexically attached to it defining its binding domain. However, when an anaphor shares more than one binding domain and these domains of the anaphor are conditioned by different factors, it is hard to propose a precise binding equation to be attached to an anaphor. Further, in multi-clause constructions, the highest subject is not necessarily the best antecedent for taman, but the one that is furthest away from the anaphor and the one that agrees with it in person. We will need to employ very complex FU equations to account for this type of long distance binding.

3.3 Reflexive Verb and Anaphoric Binding Equations

I assume that the reference to reflexive binding is encoded in the lexical entry of each reflexive verb since it indicates the coreferentiality between its subject and an anaphor which occurs within its nucleus. Therefore our binding equations must refer to the reflexive verb. We need two sets of binding equations for taman: the first one for the reflexive binding in a nucleus containing a reflexive verb, and the second one for reflexive binding in multi-clause constructions in which taman appears in a nucleus of a non-reflexive verb. The first set of binding equations involving a reflexive nucleus (see 23) refers to MCN.

(23) a.      [((DomainPath GF |) AncedentPath)s = |s

        b.     ¬ (--> SUBJ)

        c.      | REFL =c +]

"|" following the GF in the binding equation is the anaphor. With "¬(-->SUBJ)" (in 23.b) the DomainPath cannot pass through a f-structure containing a SUBJ: for instance, an anaphor, whether or not it is an argument, should bind with the first subject in its DomainPath. In other words, it should bind within the minimal complete binding nucleus: see sentence (7.a). According to the constraint equation in (23.c) this binding condition is applicable only if the verb is reflexive. I assume that the verb is lexically specified as to whether it is reflexive or not.

Now consider a sentence like (9.a), which does not have a reflexive verb and in which taman corefers with the highest subject. For these types of sentences without reflexive verbs, we need to employ the second set of binding constraints. These binding equations should contain the constraint equation (| REFL =c -), so that our binding equations will only apply to sentences with non-reflexive verbs. The binding relation in (9.a) is constrained by Dalrymple"s conditions: Coargument Disjointness Condition (p.131-134) which imposes disjointness in reference from coarguments in the same nucleus; Root S Condition (p.136) which says that any argument in the sentence can antecede the anaphor, and Subject Condition (p.137-8), according to which only subjects can be antecedents. When these three conditions are jointly applied, they allow the reflexive to be bound to any subject in a higher S. With the Root S Condition, an anaphor can appear at an indefinite distance from its antecedent, as it does not specify a particular antecedent to be bound with. The nature of this condition predicts ambiguity in sentences such as (9.a) which has a number of clauses embedded in it. However, (9.a) is not ambiguous and the antecedent is Gune who is being criticised. To account for binding relations of taman in multi-clause constructions (like 9), it is possible to use Subject Condition, Root S Condition and another new condition that we can call "Non-Highest Subject Non-Coreference Condition" (Dalrymple, p.c.), which rules out any non-highest subject being coreferential with taman. While the combination of these three conditions can account for certain long distance binding relations (like (9.a)), they fail to account for sentences like (10.b), in which the lower subject becomes the antecedent, since the highest subject is a first person pronoun. This poses a serious problem for our binding conditions.

Let"s assume that we modify our binding conditions in a such way that would rule out the highest subject being the antecedent if it is a 1st or 2nd person pronoun [FN8]. However, when we apply our modified long distance binding conditions to a sentence like (10.b), still it will not generate the correct output. The reason is that Subject Condition, Root S Condition and Non-Highest Subject Non-Coreference Condition jointly rule out all the arguments except the highest subject being the antecedent. What we added to the binding conditions to prevent any 1st or 2nd person pronoun being an antecedent, does not allow the 1st person pronoun in (10.b) to be the antecedent. Therefore, by not having an antecedent that obeys all the conditions, the sentence may be treated as ungrammatical. Therefore our binding conditions must have the flexibility to select the next lower subject as the antecedent, when the highest subject is not the optimal one.

As a theoretical alternative, Optimality Theory (OT) appears to be worth considering with regard to the Sinhala anaphoric binding. This does not mean that the anaphoric binding system based on FU is wrong or cannot describe the binding relations of taman in Sinhala. The main problem that we considered previously is that taman has a number of different binding domains depending on various conditions. Three such domains were identified: 1) the subject in the minimal complete nucleus (MCN) if the verb is reflexive; 2) the highest subject in multi-clause constructions; and 3) the next available lower subject if the highest subject is not available in a multi-clause construction. The pronoun eyaa is subject to two binding domains: the subject in MCN with a reflexive verb, and disjointness from co-arguments. We can list these domain and antecedency requirements as conditions on binding relations, as in (24) where the leftmost conditions are the overriding ones. The hierarchy of conditions given here is not intended to serve as a set of constraints of anaphoric binding based on OT, but rather as a illustration of how competing binding conditions of Sinhala anaphor can be explained in terms of the spirit of OT.

(24) Hierarchy of Domain/Antecedency Requirements of taman and eyaa

Reflexive taman (A) Disjointness from non-3rd person nominals  


(B) Coreference with SUBJ in MCN 
(in nucleus of a reflexive verb) 


(C) Coreference with SUBJ which is furthest away from taman 
Pronoun eyaa (i) Disjointness from non-3rd person nominals  


(ii) Coreference with SUBJ in MCN  
(in nucleus of a reflexive verb) 


(iii) Disjointness from coarguments 

The above-given hierarchy of conditions shows a cline less to more violable (or, alternatively, from more to less strict). In (10.b) repeated below as (25), for instance, the coreference between api "we" and taman would violate person agreement (24A) and (24B) is not applicable for this sentence. Thus, at the expense of violating the lower condition (C), taman binds with the lower subject. In a simple sentence with a reflexive verb and a first person subject, as given in (2.b) repeated as (26), taman cannot bind with the subject although it would satisfy both (B) and (C). Such coreference will result in the sentence being unacceptable, simply because the anaphor does not agree with the subject in person. Thus, it indicates that condition (A) should rank higher than (B).

(25) api1 Siri-tå2  [ø2 taman*1,2,*3 gænå  --
     1PL  Siri-DAT  [  self        about --

      -- John-tå3  kiyannå] kiyåla kiwwa.
      -- John-DAT  say.INF] that   say.PST 

      "We1 told Siri22 to tell John3 about self*1,2,*3]."

(26)  api ape/*taman-ge      wædak      --
      1PL 1PL.GEN/*self-GEN  work.INDEF --

       -- balaa_gatta,       nam  hondai.
       -- look.PP_take.PST,  if   good 

      "It is better if we minded our own/*self"s business."
Pronouns are normally free within MCN (Condition (iii) in (24)). Their binding relations are conditioned by the reflexive verb, when they occur within the nucleus of a reflexive verb. Therefore Condition (ii) should rank higher than the more general Condition (iii). Like taman, eyaa being a third person pronoun cannot corefer with a non-third person noun. Consider (27a-b):
(27)a. Daya1 eyaa-ge1,*2  kaarek   hodå-gatta.
       Daya  3SG-GEN     car.DEF  wash.PP-take.PST
       "Daya1 washed his1,*2 car."

    b. mamå1  eyaa-ge2 kaarekå  hodå-gatta.
       1SG    3sG-GEN  car.DEF  wash.PP-take.PST
       "I washed his car."
The reflexive verb in (27.a) requires eyaa to be bound with the subject. However, this is not possible in (b) as the pronoun cannot agree with the non-third person nominal. Unlike the reflexive taman, the pronoun eyaa need not have a binder in the same sentence. Therefore, (27.b) becomes acceptable indicating that the person agreement between the antecedant and the anaphor is more important than those restrictions imposed by reflexive verbs. I assume that an approach based on OT would be more suitable for explaining complex binding relations of Sinhala anaphors. Ideally, we would expect that in an OT-based account, there would be a set of universal constraints of anaphoric binding relations to provide the flexibility needed for language specific variations.

4. Conclusion

Data from Sinhala illustrates that an anaphor is subject to a combination of different binding conditions. The binding domain of an anaphor varies depending on the interaction of these binding conditions. The reflexive and personal pronouns are not in complementary distribution. The anaphoric elements discussed here cannot be characterised as either "clause-bound" or "long distance", since an anaphor can have a variable binding domain. Thus, the domain of reflexivisation is not a defining feature of a given anaphoric element.

We observed that verbs morphologically marked for reflexivisaton put constraints on coreference between anaphor and antecedent. I assume that the reflexive verb brings further insight into the interaction between the representation of argument structure, lexical semantics and morpho-syntax, and therefore requires a thorough investigation. Finally, as we very briefly showed Sinhala anaphoric binding relations may be better described in terms of Optimality Theory, which allows for the violation of certain binding conditions through constraint interaction.


(FN1) This paper has benefited from comments received from audiences at the ANU Linguistics Seminar and the LFG "98 Conference. I would like to thank Sasha Aikhenvald, Cindy Allen, Avery Andrews, Mary Dalrymple, Stuart Robinson and Jane Simpson for their helpful comments and suggestions. I alone am responsible for errors. [back]

(FN2) All the conclusions reached in this paper are based on colloquial Sinhala and the assumptions made here with regard to reflexives of colloquial Sinhala may, therefore, be quite different from that of literary Sinhala. This study represents judgements of 16 native Sinhala speakers who were asked to comment on coreference relations of arguments in sentences. While not all the sentences were tested for judgments on corefentiality with all 16 informants, most of the sentences represent judgements of about 10-12 informants. All the data in this paper is from this study, unless otherwise stated. [back]

(FN3) According to Gunasekara (1891:173), taman is plural and singular form is tama. However, he points out that regardless of its singular inflection, tama is used as plural and vice versa. Tamun, as given in Gair et al (1989:98), is assumed to be in free variation with taman, and the both forms are same in meaning and their distribution. It is worth noting that in spoken Sinhala, taman is not as frequently used as it is in literary Sinhala. This example from "Lankadiipa" newspaper illustrates the occurrence of taman in literary Sinhala:

  mee  pilibandåwå puwathpath dænwiim          palå kiriimåtå --
  this about       newspaper  advertisement.PL publish-INF --
   -- tamaa katåyutu_kå rånå   bawå   D.. Kiiya. 
   -- self  work-on.PRE       that  D.. say.PST
  "D said that self is working on advertising on 
   newspapers about this (matter)." [back]

(FN4) The following abbreviations are used:
ACC: accusative case; ANIM: animate; DAT: dative case; INAN: inanimate; INDEF: indefinite; INF: infinitive; INST: instrumental/ablative case; GEN: genitive/locative; NEG: negative; NOM: nominative case; PL: plural; PP: perfect participle; PRES: present; PST: past; SG: singular

Animate and inanimate nouns, and inanimate pronouns are unmarked for both nominative and accusative case, while animate pronouns and proper names are marked with -wå for accusative. The letter å in the text indicates the schwa. [back]

(FN5) It is also to be noted that combinations of gannåwa with verbs/verbal bases have other meanings and functions. This is an example from Paolillo (1989) quoted in Herring (1989:179) for compound continuative construction:

  Meyaage bandinå    wayåwå dæn pahu wee_genå_enåwa.
  his/her marry-AdjP age    now past become_take.PP_come.PRE
  "Her marrying age is now passing by."
As Reynolds (1980:185) points out, some sentences with gannåwa can have the meaning of "managed to ....." as in the following sentence.
  ohu dumriyå-tå  pænå_gatta.
  3SG train-DAT   jump.PP_take.PST
  "He managed to jump onto the train".  [back]
(FN6) See Mandarin (Chief 1996), where the presence of a reflexive verb in an intermediate clause blocks the reflexive being bound by a higher subject. [back]

(FN7) See Chief (1996) for a discussion of similar binding relations in Mandarin pronouns conditioned by reflexive verbs. [back]

(FN8) This might be done by adding a statement like ¬((| SUBJ = 1) ^ (| SUBJ = 2)) to binding equations. [back]


Chief, L. C. 1996. An LFG Account of Mandarin Reflexive Verbs. Paper presented at the LFG Workshop, Grenoble, France, August 1996.

Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht: Foris Publications

Clements, G. N., 1975. The Logophoric Pronoun in Eve: Its Role in Discourse. Journal of West African Languages. 10:2 141-77

Dalrymple, M., 1993. The Syntax of Anaphoric Binding. Stanford, CA: CSLI Publications

Gair, J. W. 1970. Colloquial Sinhalese Clause Structures. Mouton: The Hague.

Gair, J. W., B. Lust, L. Sumangala, and M. Rodrigo. 1989. Acquisition of null subjects and control in some Sinhala adverbial clauses. Papers and reports on child language development. Stanford, CA: Linguistics Department, Stanford University.

Gunasekara, A. M. 1981. A Comprehensive Grammar of the Sinhalese Language. Colombo: Sri Lanka Sahitya Mandalaya.

Herring, S. C. 1989. Aspectogenesis in South Dravidian: On the origin of the Compound Continuative. In H. Aertsen and R. J. Jeffers (Eds.), Historical Linguistics. 167-85. Amsterdam: Benjamins

Inman, M. V. 1994. Semantics and Pragmatics of Colloquial Sinhala Involitive Verbs. Unpublished PhD Dissertation. Stanford, CA: Department of Linguistics, Stanford University.

Kaplan, R. M., J. Maxwell and A. Zaenen. 1987. Functional Uncertainty. CSLI Monthly Newsletter. January.

Prince, A., P. Smolensky. 1993. Optimality Theory: Constraint Interaction in Generative Grammar. RuCCs Technical Report #2. Rutgers University, Center for Cognitive Science.

Reynolds, C. H. B. 1980. Sinhalese: An Introductory Course. London: School of Oriental and African Studies, University of London.

Kumara Henadeerage
Department of Linguistics (The Faculties)
Australian National University
Canberra, ACT0200, Australia.