The whole picture: Disentangling locality, logophoricity and subjecthood in English picture noun anaphora

This article provides a solution to the long-standing puzzle of English anaphors within so-called picture noun phrases, which superficially exhibit an exceptional binding behavior. In particular, picture noun anaphors seem, under certain conditions, to escape the locality conditions imposed by Condition A of Binding Theory. Previous proposals attribute such apparently exceptional behavior to various sources: the classical Binding Theory appeals to the possible presence of covert agents within NPs; predicate-based theories introduce the possibility of exemption from Condition A; others capitalize on possible homophony with (logophoric) pronouns. While all of these proposals provide valuable insight into some aspect of the puzzle, we show that they all fail to capture the full empirical picture. Based on a detailed examination of their behavior in various syntactic and interpretive conditions, we instead propose that English picture noun anaphors, like any other anaphor, systematically obey Condition A. Their apparent exemption from it in some cases derives from the possible implicitness of some binders, in particular, logophoric pronouns or nominal subjects. Furthermore, the availability of such covert binders is crucially affected by a binding-independent competition principle between weaker and stronger forms. Thus, the apparently irregular behavior of English picture noun anaphors results from the interaction between several factors (syntactic representation of logophoricity, syntactic projection of subjects in nouns, pronominal competition), which is responsible for the illusion that Condition A does not apply systematically. By disentangling these factors, we propose a solution that integrates previous insights without compromising on empirical adequacy or analytical parsimony.


Introduction
Since Warshawsky (1965) coined the term picture nouns to refer to phrases headed by representational nouns like picture or story, the behavior of English anaphors within such phrases has remained an outstanding issue for theories of binding. In particular, reflexives and reciprocals in picture noun phrases (henceforth, Picture Noun Anaphors or PNAs) seem to routinely disobey the locality conditions imposed by Condition A of Binding Theory, as illustrated in (1)-(2).
(1) Tom i believes that there is a picture of himself i hanging in the post office. (Jackendoff 1972: 133) (2) [The men] i knew that there were pictures of [each other] i on sale. (Pollard and Sag 1992: 267) This type of observation led many (starting with Postal 1971 1 -see also Bouchard 1984; Rooryck and Vanden Wyngaerd 2011; among many others) to assume that PNAs form an exceptional class of anaphors. Drummond, Kush and Hornstein's assumption is representative in this respect: "a reflexive within a picture noun phrase that is bound from outside its containing noun phrase is not a true reflexive subject to principle A. Rather, it is a pronominal with special logophoric requirements. This follows a long tradition of analysis [...]" (Drummond et al. 2011: 401).
A closer look at the literature reveals that the two main theories of anaphor licensing do not, however, assign a specific status to PNAs. First, Chomsky (1981Chomsky ( , 1986 supposes that just like any anaphor, PNAs obey Condition A of Binding Theory, and must thus be bound within the smallest phrase containing them and a subject distinct from them. 2 Chomsky further posits the possible presence of a PRO-like implicit subject within DPs to account for the fact that anaphors like each other in (3) (vs. (4)) are not in complementary distribution with pronouns as expected under the classical Binding Theory (see further discussion in Sect. 2.1.1.3); the contrast between (3) and (4) derives from the meaning difference between tell and hear.
(3) a. They i heard stories about [each other] i . 1 Postal (1971: 12, fn. 6) explicitly argues that PNAs are not ordinary reflexives, so that their behavior should not motivate reformulation of the reflexivization rule (Lees and Klima's 1963 rule or any other version). This assumption contrasts with Jackendoff (1972) or Ross (1967Ross ( , 1970, who (implicitly) suggest that PNAs should fall under the scope of the reflexivization rule, thus setting up the debate about PNAs until today. 2 The formulation of Condition A provided in the text is a paraphrase of Chomsky (1981Chomsky ( , 1986. The requirement for the binding domain of the anaphor to contain a subject distinct from the anaphor is introduced by the notion of accessible subject (where a subject is not accessible if the i-within-i filter is violated) in Chomsky (1981Chomsky ( : 213-214, 1986. This point is meant to explain the grammaticality of each other in sentences like (i), which involve an anaphor bound from outside the subject of an embedded clause. See fn. 16 for further discussion about this point.
(i) [The children] i thought that pictures of [each other] i were on sale. (Chomsky 1986: 173) b. They i heard [PRO k stories about them i ]. (Chomsky 1986: 166-167) (4) a.  (Chomsky 1986: 166-167) Second, predicate-based theories (henceforth PBTs-see Pollard and Sag 1992;Reinhart and Reuland 1993; and subsequent versions thereof) do not treat PNAs as a special class either. Unlike Chomsky, they argue that PNAs in possessorless DPs are exempt from Condition A, which they redefine as obligatory coargument binding (see further discussion in Sect. 2.1.1.2). But in this respect, PNAs are no different from all other instances of anaphors lacking a coargument, such as (5a) (vs. (5b)). Exempt PNAs also pattern with other exempt anaphors in being subject to perspective-related discourse conditions: himself in (5a), for example, is licensed by the fact that the clause containing it represents the point of view of the referent of its antecedent, Max.
(5) a. Max i boasted that the queen invited Lucie and himself i for a drink. b. *Max i boasted that the queen invited himself i for a drink. (Reinhart and Reuland 1993: 670) On the other hand, PNAs in possessed DPs are assumed to be subject to Condition A (at least under early PBT versions treating possessors as subjects; see further discussion in Sect. 2.1.2.1); in cases like (6b), the PNA must therefore be bound by the possessor. According to PBTs, this explains the reported contrast between (6a) and (6b).
(6) a. Lucie i liked a picture of herself i . b.*/?Lucie i liked your picture of herself i . (Reinhart and Reuland 1993: 682) Thus, PNAs are consistently treated as plain anaphors (in Charnavel and Sportiche's 2016 terminology) under the Chomskian theory, while under PBTs, PNAs divide into plain anaphors in possessed DPs and exempt anaphors (in Pollard and Sag's 1992 terminology) in possessorless DPs. In both cases though, their occurrence within phrases headed by the descriptive class of picture nouns does not translate into a specific behavior. 3 The nevertheless persistent idea of PNA exceptionalism may come from the failure of both theories to capture the full behavior of PNAs. As we will see, Chomsky's theory is indeed unable to predict the contrast between PNAs under different perspectival conditions. For example, his PRO-based hypothesis should presumably imply that himself will have the same status in (7) and in (1) as long as the author of the picture is the same in both sentences (see further discussion in Sect. 3.2.3): in both, himself should only be acceptable if John took the picture, whether or not his point of view is represented in the clause, contrary to fact.
(7) *Mary said about John i that there was a picture of himself i in the post office. (Kuno 1987: 126) Such sensitivity of PNAs to point of view (argued by Kuno 1972Kuno , 1987Cantrall 1974;Keenan 1988;Zribi-Hertz 1989;i.a.) is what motivated PBTs to develop a theory of exemption, under which himself in (1) and (7) is not subject to Condition A, but to perspective-based discourse conditions. 4 But conversely, PBTs thereby fail to predict the different grammaticality status of PNAs under different syntactic conditions, as shown for French by Charnavel and Sportiche (2016): in particular, the inanimate elle-même in (8), which lacks a coargument, should under PBTs be excluded in both (8a) and (8b) regardless of the position of its antecedent, given that inanimates cannot take perspective and, hence, cannot satisfy discourse conditions on exemption. '[This law] i is so important that the journalists predict the publication of a book about itself i and about its author.' (Charnavel and Sportiche 2016: 49) The goal of this article, which will concentrate on English reflexives (leaving the investigation of reciprocals and crosslinguistic anaphors for future research), is to solve the long-standing puzzle posed by English PNAs by integrating these various perspectives. 5 As we will see, each of these theories provides valuable insight into the puzzle, but misses at least one crucial aspect of it. Instead, we propose a new combination of mostly existing ingredients that leads to a full solution to the PNA puzzle without assigning an exceptional status to PNAs. In line with Chomsky (vs. PBTs), we argue that English PNAs, just like any anaphor, are uniformly subject to Binding Theory: Condition A suffers no exception (and is antecedent-based). We also agree with PBTs (vs. Chomsky) that, descriptively speaking, anaphors exhibit a heterogeneous behavior, being either plain or exempt. To resolve this apparent paradox, we adopt Charnavel's (2020aCharnavel's ( , 2020b hypothesis that the descriptive heterogeneity of anaphors can be reduced to the heterogeneity of their local binders. In particular, besides standardly postulated binders, anaphorsincluding PNAs-can take covert logophoric binders, which syntactically represent the locally relevant perspective center. Thus, anaphors that are descriptively exempt in fact covertly comply with Condition A. Charnavel's logophoric A-binder solution is not sufficient to fully solve the English PNA puzzle, however. As we will see, there is another factor at stake-hinted at by both Chomsky and PBTs, albeit in very different ways-which further complexifies the superficial behavior of PNAs: as shown by (5b) (vs. (5a)), English anaphors cannot always be logophorically bound even under appropriate discourse conditions. We will attribute this fact to an independent constraint, which, descriptively, blocks logophoric binding in the presence of a coargument subject. The theoretical relevance of subject coargumenthood for English reflexives-outside PBTs-is demonstrated in Ahn (2015), which distinguishes anaphors that are bound by a coargument subject from other anaphors on the basis of prosody-based diagnostics. To explain the blocking of logophoric binding in sentences like (5b), we propose to complement this insight with a strong/weak competition hypothesis à la Cardinaletti and Starke (1999): the apparent competition between some binders (i.e. logophoric binder and coargumental subject) in fact results from a competition between some possible bindees (strong herself vs. weak herself and weak her), which falls under a general competition principle between weaker and stronger forms. This hypothesis thus partly reintegrates in the account of PNAs both the relevance of coargumenthood, which is at the root of PBTs, and the possible presence of implicit PRO-like subjects within nominals, which is crucial to the Chomskian theory.
In sum, the seemingly heterogeneous distribution of the descriptive class of English PNAs illustrates how the interaction between simple and general principles-Condition A, logophoricity, weak/strong competition-can yield superficially complex behaviors. Such apparent complexity, in our opinion, does not warrant a relaxation of parsimony or rule generality, but a disentanglement of the various interacting factors.
The outline of the paper is as follows. The first part (i.e. Section 2) will re-examine the various properties purported to distinguish between plain and exempt anaphors in order to determine which of these properties actually characterize English PNAs in various syntactic contexts. To this end, we will mainly follow the proposals of Charnavel and Sportiche (2016) and Charnavel (2020aCharnavel ( , 2020b in order to independently determine the local domain relevant to PNA binding as well as the notion of logophoricity relevant to apparent exemption. This empirical exploration will motivate unification of the descriptively double (plain/exempt) behavior of English PNAs by adapting Charnavel's (2020aCharnavel's ( , 2020b logophoric A-binder hypothesis, which reduces exempt to plain behavior (as roughly represented in (9b) as compared to (9a)). In the second part of the paper (i.e. Section 3), we will investigate the blocking of logophoric binding by the presence of (overt or covert) coargument subjects (i.e. explore the conditions of application of cases (10a) vs. (10b)).
(10) a. ... (DP i  To establish the generalization, we will first concentrate on the verbal domain, where the obligatory overtness of subjects removes a complicating factor. The generalization will be explained using Ahn's (2015) discovery about the prosodic behavior of anaphors bound by coargument subjects along with Cardinaletti and Starke's (1999) general principle of competition between weaker and stronger forms. We will then come back to the nominal domain, where we will explore the consequences of this competition when the possible implicitness of nominal subjects is added to the general picture.

Unifying plain and exempt PNAs: The logophoric A-binder hypothesis
English PNAs have received specific attention in the literature because, unlike other anaphors, they seem to be generally exempt from the structural conditions of locality defined by Condition A of Binding Theory. The goal of this section is to challenge this claim on both descriptive and analytical levels. Descriptively (Sect. 2.1), we show that PNAs can in fact be plain or exempt; we reach this conclusion by re-examining their distribution using a criterion independent of Condition A (i.e. Charnavel and Sportiche's 2016 inanimacy-based tool) to distinguish between plain and exempt instances of anaphors. Analytically (Sect. 2.2), we propose that PNAs are in fact never exempt, but consistently obey Condition A; we obtain this result by adopting Charnavel's (2020a, 2020b) logophoric A-binder hypothesis, which reduces exempt behavior to local binding by an implicit logophoric binder. This hypothesis correctly predicts the descriptively dual behavior of English PNAs while avoiding postulation of any kind of homophony or restriction of the scope of Condition A.

PNAs in possessorless DPs
In this section, we concentrate on the prototypical case of PNAs, namely reflexives in picture noun phrases that lack a possessor (henceforth possessorless PNAs). We show that, descriptively speaking, they are neither uniformly plain anaphors as implied by the Chomskian theory, nor uniformly exempt anaphors as implied by predicatebased theories, but in fact exhibit a dual behavior. Recall that by plain anaphors, we mean anaphors that standardly obey Condition A and by exempt anaphors, anaphors that seem to disobey Condition A; as we will see, the content of these terms thus depends on the definition of Condition A. The bulk of our argumentation challenging the earlier literature will therefore consist in independently determining the scope of Condition A and conditions on exemption.   Table 1 and detailed below, possessorless PNAs are usually claimed to exhibit four distributional properties that distinguish them from plain anaphors (see Bouchard 1984;Lebeaux 1984;i.a.). It is generally assumed that possessorless PNAs are not unique in this respect, but share these characteristics with other instances of anaphors, in particular those within a conjoined DP (e.g. Mary and herself ), within like-phrases (e.g. physicists like herself ), within as for-phrases (e.g. as for herself ) or within exceptive constructions (e.g. no one but herself ) (see Ross 1970;Kuno 1972Kuno , 1987Keenan 1988;i.a.). First, as mentioned at the start, it has been observed that possessorless PNAs need not be locally bound (see Helke 1970;Ross 1970;Jackendoff 1972;Cantrall 1974;Lebeaux 1984;Bouchard 1984;Kuno 1987;Zribi-Hertz 1989;Pollard and Sag 1992;Reinhart and Reuland 1993;i.a.). Example (11) below shows that himself need not be c-commanded by its antecedent; example (1) above illustrates that the antecedent of himself does not have to be in the smallest clause containing it, and example (12) that the antecedent does not even have to be within the same sentence. These reflexives thus appear to escape the locality conditions on anaphors under any definition of locality. (11) The picture of himself i in Newsweek dominated John i 's thoughts. (Pollard and Sag 1992: 278) (12) John i was going to get even with Mary. That picture of himself i in the paper would really annoy her, as would the other stunts he had planned. (Pollard and Sag 1992: 274) Importantly, these examples remain apparent exceptions to the Chomskian Condition A even if we extend Chomsky's PRO-based solution to them. 6 Recall that in (3)-(4) above, Chomsky posits that the DP containing the picture noun can include a PROlike subject, thereby allowing the object pronoun to corefer with an apparently local antecedent. If we adopt this hypothesis, we might further assume that the PRO-like subject can serve as the local binder of an anaphor. Extending this to examples like (1), (11) or (12), we may conclude that the appearance of exemption follows from local binding by the null nominal subject; this in turn would imply that in such ex-amples, it is the referent of the antecedent of the reflexive that took the picture. 7 But crucially, these examples do not require this type of interpretation: for instance, (11) is perfectly acceptable in a context where John did not take the picture-in fact, this is the most natural interpretation. That is not to say that Chomsky's hypothesis must be abandoned without further discussion. We will see in Sect. 3.2.3 that it is in fact part of the solution. But the acceptability of these examples under a non-agentive interpretation of the antecedent shows that an extension of Chomsky's hypothesis is not sufficient to account for all instances of non-locally bound PNAs. The aim of this first part of the paper (i.e. Section 2) is to investigate the cases that cannot fall under the Chomskian explanation; unless otherwise stated, all examples should therefore be read under the aforementioned non-agentive interpretation. Second, possessorless PNAs are claimed to contrast with plain anaphors in allowing split antecedents (see Helke 1970;Bouchard 1984;Lebeaux 1984;Pollard and Sag 1992;i.a.). (13) a. John i told Mary k that there were some pictures of themselves i+k inside. b. *John i told Mary k about themselves i+k . (Lebeaux 1984: 346) Third, it is often assumed that possessorless PNAs can trigger sloppy or strict readings in ellipsis contexts, while plain anaphors only exhibit sloppy readings (see Bouchard 1984;Lebeaux 1984;Reinhart and Reuland 1993;Kiparsky 2012;Runner et al. 2002; i.a.).
(14) John i thought that there were some pictures of himself i inside, and Bill did too. a. Bill thought that there were some pictures of himself inside too. b. Bill thought that there were some pictures of John inside too. (Lebeaux 1984: 346) (15) John i hit himself i , and Bill did too. a. Bill hit himself too. b. #Bill hit John too. (Lebeaux 1984: 346) Finally, it is commonly supposed that possessorless PNAs, unlike plain anaphors, are in free variation with pronouns (see Jackendoff 1972;Lebeaux 1984;Chomsky 1986; i.a.).
(16) a. John i knew that there were some pictures of {himself i /him i } inside. b. John i likes {himself i /*him i }. (Lebeaux 1984: 346) 2.1.1.2 Possessorless PNAs as exempt under PBTs' approach to exemption These four purported distributional properties of PNAs and other instances of anaphors, which are properties of pronouns, have caused a widespread and persistent assumption that PNAs are in fact not real anaphors, but pronouns (see Bouchard 1984;Safir 2004;Drummond et al. 2011;Rooryck and Vanden Wyngaerd 2011;i.a.). However, this hypothesis implies some kind of lexical ambiguity or homophony: English herself would have two-related-lexical entries, one for plain behavior, one for exempt behavior. This assumption clearly goes against parsimony, especially since the generalization holds cross-linguistically: it is not just in English, but also in many other unrelated languages like Chinese, Korean or Turkish, that reflexives exhibit a dual behavior (see review in Charnavel 2020a). Instead of postulating homophony between an anaphor herself and a pronoun herself, PBTs (Pollard and Sag 1992;Reinhart and Reuland 1993; i.a.) develop a theory of exemption that restricts the scope of Condition A. Under these proposals, Condition A, which is redefined as a condition on predicates rather than a condition on antecedence, requires that anaphors be bound by a syntactic coargument whenever they have one. 8 Crucially, this implies that anaphors lacking a coargument, such as possessorless PNAs that are the only argument of the picture noun, are exempt from Condition A. Such exempt anaphors, PBTs argue, are instead subject to discourse conditions related to perspective. 9,10 Indeed, it has long been observed that possessorless PNAs and other exempt anaphors are licensed in clauses expressing the point of view of their antecedent (Kuroda 1965(Kuroda , 1973Cantrall 1969Cantrall , 1974Kuno 1972Kuno , 1987Clements 1975;Sells 1987;Zribi-Hertz 1989, i.a.). This generalization, exemplified by the contrast between (1) and (7) above, is also illustrated by the contrast between (12) and (17). (17) Mary was quite taken aback by the publicity John i was receiving. That picture of him(*self i ) in the paper had really annoyed her, and there was not much she could do about it. (Pollard and Sag 1992: 274) PBTs thus directly account for the first distributional property of possessorless PNAs mentioned above: possessorless PNAs need not be locally bound since, in the absence of a coargument, they are exempt from Condition A. The second property of PNAs, namely the acceptability of split antecedence, also follows from PBTs, ac-8 This paraphrase of PBT Condition A glosses over some differences across accounts. In particular, Pollard and Sag (1992: 287) hypothesize that only less oblique coarguments are relevant; this implies that subject anaphors (e.g. each other's pictures) are exempt, contrary to what is predicted by Reinhart and Reuland (1993). Conversely, Reinhart and Reuland (1993: 678) assume that coargumenthood is only relevant in syntactic predicates, which must contain a subject; this implies that anaphors with an object coargument but no subject coargument (e.g. a letter to John about himself ) are exempt, contrary to what is predicted by Pollard and Sag (1992). These variations in predictions will be taken into consideration when relevant to the argumentation. 9 As noted in fn. 4, Reinhart and Reuland (1993: Sect. 6; see also Reuland 2011) additionally propose that anaphors exempt from Condition A may in some cases be locally bound via A-chain formation (see also Charnavel and Sportiche 2016: 49-50 for discussion). But crucially, Reinhart and Reuland (1993: 702, 705) claim that possessorless PNAs do not form a chain with their antecedent. Hence, the theories of chains pursued in those works do not affect predictions for possessorless English PNAs. 10 More specifically, Pollard and Sag (1992: 271-279) assumes both processing (i.e. intervention) and discourse (i.e. point of view) constraints on exempt anaphors, while Reinhart and Reuland (1993: 672-673) supposes that the main (but non-exhaustive) uses of exempt anaphors (logophors, in their terms) are perspective and focus uses. Based on the rest of the literature on the discourse constraints on exempt anaphors (see review in Charnavel 2020a and references therein, some of which are mentioned in the text above), we subsume intervention under perspectival constraints and exclude focus uses: as argued in Charnavel (2020a: 36-39), focus is neither sufficient nor necessary for exemption.
cording to Pollard and Sag (1992: 270): exempt anaphors, which are not subject to Condition A, are free to refer to a group entity formed in the discourse, regardless of whether this entity is expressed as a single DP in the syntax. The availability of strict readings, Reinhart and Reuland (1993: 674) claim, is also accounted for under their theory (complemented with Rule I; see e.g. Grodzinsky and Reinhart 1993): possessorless PNAs, which are exempt from Condition A, can be related to their antecedent either by variable binding or by coreference, whereas anaphors with a coargument, which obey Condition A and must thus be coindexed with a coargument, can only be interpreted by variable binding. As for non-complementarity between PNAs and pronouns, it is also predicted by PBTs: given that PBT Condition B forbids pronouns from being bound by coarguments, it follows that exempt anaphors, which by definition lack a coargument, can alternate with pronouns (under appropriate discourse conditions). Furthermore, the distinction between syntactic and semantic coarguments made by Reinhart and Reuland (1993) explains the contrast between (3), where the PNA alternates with the pronoun, and (4), where it does not: in sentences like (4), the pronoun, unlike the reflexive, is ruled out by the semantic representation of the agent role associated with the picture noun, because only Condition B (vs. Condition A) is sensitive to semantic co-argumenthood. 11

Possessorless PNAs as plain or exempt under Charnavel and Sportiche's (2016)
approach to exemption Because PBTs redefine Condition A as obligatory coargument binding, the core property held to be responsible for the exempt behavior of possessorless PNAs is their lack of a coargument. As argued in detail in Charnavel and Sportiche (2016) based on French anaphors, this approach to exemption is empirically incorrect; we confirm this conclusion for English PNAs below. Instead, Charnavel and Sportiche (2016) propose a criterion independent of Condition A to descriptively tease apart plain and exempt anaphors, namely inanimacy. Their reasoning is based on the widespread observation (also adopted by PBTs) that, crosslinguistically, exempt reflexives are subject to logophoric conditions, i.e. their clause must express the point of view of their antecedent (see aforementioned references). Given the controversial and often imprecise definition of logophoricity (see review in Charnavel 2020a), this notion cannot be directly used to detect exempt anaphors. But non-logophoricity, Charnavel and Sportiche (2016) argue, can conversely be used to identify plain anaphors. Specifically, they hypothesize that inanimate anaphors cannot be logophoric, as by nature they lack a mental state, which is required to take perspective under (virtually 12 ) all definitions of logophoricity, and therefore are not eligible for exemption. 11 Reinhart and Reuland's (1993: 685-686) solution is thus in some sense similar to Chomsky's (1986) PRO-based solution mentioned above: instead of positing the syntactic presence of a PRO subject in the subject of picture noun complements of verbs like tell (a story) or take (a picture), they assume (inspired by Williams 1985) that the agent role of these nouns in these configurations is present semantically although it is not syntactically realized. 12 One type of perspective does not require a mental state, namely spatial perspective, which only relies on physical location and orientation. Spatial perspective is sometimes subsumed under logophoricity as in Sells (1987: 456), which defines one logophoric role (i.e. Pivot) as the center of deixis "in a very physical sense," even if inanimates are not discussed. But Charnavel (2020aCharnavel ( , 2020b demonstrates that pure deictic perspective is in fact not sufficient for exemption, which requires mental perspective (i.e. attitude or empathy, as will be discussed in Sect. 2.2.1). Here, we will simply avoid contexts involving spatial perspective.
This inanimacy-based tool can be used to re-examine the distributional properties of English possessorless PNAs (cf. Bassel 2018 on Hebrew). It reveals that, contrary to the predictions of PBTs, possessorless PNAs do not consistently exhibit an exempt behavior. First, inanimate possessorless PNAs do obey locality conditions: 13 unlike himself in (1), (11) or (12) above, itself must be bound (see (18a) vs. (18c)) and cannot be bound across a subject (see (18a) (21)- (22) and (24) below) should similarly be ruled out under PBTs since they also contain non- 13 The few examples of inanimate PNAs mentioned in the literature (by e.g. Minkoff 1994Minkoff , 2000Minkoff , 2004Postal 2006) support this generalization.
(i) *That ugly picture of itself i hurt [the car] i 's steering wheel (by falling on it). (Minkoff 1994: 127) (ii) a.
[Winston Q. Felix] i insisted that any criticisms of himself i would be based on prejudice. b.
[The Nature of It All] i insisted that any criticisms of it i (*self) would be based on prejudice. c.
[Winston Q. Felix] i rejected in advance future criticisms of himself i . d.
[The Nature of It All] i rejected in advance future criticisms of it i (self). (Postal 2006: 11) Note though that the antecedent of itself in Postal's (2006) examples is not strictly inanimate: the choice of verbs (insist, reject) suggests that The Nature of It All, which is meant to be the title of a book, could arguably stand for its author. Such cases raise interesting questions about the notion of inanimacy, which we leave for future research. Here, we simply avoid these cases by restricting ourselves to inanimate anaphors with clearly non-mental antecedents. 14 Unless otherwise noted, examples that follow are our own. As is well-known from the literature, judgments about anaphors vary across speakers (see some discussion in fn. 18; see also Chomsky 1981: 214-216, i.a.); as is standard, we thus use "ok/*" to indicate contrasts in acceptability within sets of sentences rather than absolute grammaticality judgments. Acceptability contrasts indicated on novel English examples reflect judgments of several native speakers of American English, including the second author; judgments were collected using acceptability questionnaires. French judgments likewise reflect those of several native speakers of French, including the first author.
perspectival anaphors without coarguments. Application of the inanimacy-based tool to possessorless PNAs thus reveals that the dividing line between plain and exempt anaphors should not be based on coargumenthood. In fact, all the observations above also hold for all other types of anaphors without coarguments, as shown in (20), for example.
(20) a. [The witty play] i refers to itself i and its author. b. *[The witty play] i led {Bob / newspapers} to provide information about itself i and its author.
The correct descriptive generalization, then, should be formulated as follows: all inanimate anaphors must be locally and exhaustively bound, but some animate anaphors (i.e. perspectival ones, see Sect. 2.2.1) need not be locally or exhaustively bound. Possessorless PNAs are not special in any way, but exhibit this dual plain/exempt behavior accordingly. With respect to the remaining two purported distributional properties of exempt anaphors, inanimate possessorless PNAs pattern with animate ones: as illustrated in (21)-(22), they can trigger strict readings and alternate with pronouns. As we show below, this fact does not indicate that possessorless PNAs are exempt, but rather, that these two properties do not accurately distinguish exempt from plain anaphors.   Reinhart and Reuland (1993: 674-675) attempts to account for such examples by appealing to focus as a licensing factor for exemption: according to them, it is because they are focused that the anaphors in such examples can trigger strict readings despite the presence of a coargument. However, as argued in Charnavel (2020a: 36-39), focus is neither sufficient or necessary for exemption. That said, Reinhart and Reuland's empirical observation may be relevant to accounting for the availability of strict readings in some cases. As we will show in Sect. 3, descriptively exempt anaphors must be strong due to weak/strong competition independent of Condition A. Given that such competition only arises if the interpretation remains the same, the possibility of strict readings in ellipsis may license a strong form where it is usually not licensed and thus give rise to the illusion that focus licenses exemption. Thus, our findings ultimately suggest a way to account for the ellipsis facts, but a full exploration is beyond the scope of this article (see Charnavel and Sportiche to appear for further discussion).
[That magnet] i attracts paper clips to it i (self). (Minkoff 2000: 584-585) Just as in the case of strict readings, we will not provide an explanation for this fact here, which bears on Condition B of Binding Theory; the observation that noncomplementary distribution with pronouns, just like strict readings, is not a specific property of exempt anaphors, is sufficient for our purposes. But note that the fact illustrated in (25) (and more generally, our whole paper) is consistent with the standard hypothesis that pronouns must be disjoint from local binders (Condition B) and further supports the idea that the local domain relevant to Condition B is smaller than the local domain relevant to Condition A (Huang 1983;Chomsky 1986; i.a.; see fn. 24, 31, 51 and Sect. 3.2.3 for some further discussion about Condition B).
In sum, only two distributional properties reliably distinguish exempt from plain anaphors, as summarized in Table 2.
Crucially, in contrast with the predictions of both Chomskian Binding Theory and PBTs, possessorless PNAs can behave as plain or exempt with respect to these two properties. The descriptive generalization we have reached in this section is thus the following: Descriptive generalization about possessorless PNAs: Inanimate PNAs must be bound exhaustively and locally (i.e. within the smallest tensed TP containing them, without any subject intervening between them and their antecedent), whereas some animate PNAs need not be.
This generalization matches the generalization that Charnavel and Sportiche (2016) formulate on the basis of French inanimate anaphors in general (not just PNAs), which leads them to redefine Condition A as in (27). 16 (27) Condition A (Charnavel and Sportiche 2016: 71): A plain anaphor must be bound within the minimal spellout domain containing it (i.e. tensed TP, or any other XP with a subject distinct from the anaphor).
We will therefore adopt the definition of Condition A given in (27) for the remainder of our investigation of English PNAs. Note that we will not attempt to derive Condition A, as only the generalization about the binding domain of anaphors is relevant to our purposes. But as discussed in Charnavel and Sportiche (2016, Sect. 5.4.3), this formulation is compatible with movement-based approaches to binding (see Drummond et al. 2011;Kayne 2002; Charnavel and Sportiche to appear; i.a.).  (31) a. John i 's description of himself i was flawless. b. *The fact that Mary k 's description of himself i was flawless was believed to be disturbing John i . (Pollard and Sag 1992: 265) But these contrasts are not robust, as already suggested in Reuland (1993: 683, citing Ben-Shalom andWeijler 1990), prompting speculation therein that NPs may in fact never contain a subject, as proposed in Williams (1985). 18 Many experimental studies Keller and Asudeh 2001;Runner et al. 2002Runner et al. , 2003Runner et al. , 2006Jaeger 2004;Runner and Kaiser 2005; i.a.) have since confirmed that possessed PNAs can be bound from outside their picture NP. For example, the magnitude estimation task used by Keller and Asudeh (2001: 7) revealed no significant acceptability difference between (32a) and (32b), which are both highly acceptable.

Locality constraints of possessed PNAs in previous studies
(32) a. Hannah i found Peter k 's picture of herself i . b. Hannah i found Peter k 's picture of her i . (Keller and Asudeh 2001: 5) According to these experimental studies, the empirical observations in (32) do not challenge PBTs but, rather, the status PBTs assign to the possessor. Building on Williams (1985) and Barker (1995), Asudeh and Keller (2001) argue that the possessor is not an argument of the head noun (cf. Keller and Asudeh 2001;Runner and Kaiser 2005;i.a.). Under that revised assumption, possessed PNAs do not have any coarguments. They are thus predicted to be exempt and, therefore, able to take an antecedent outside their DP (under appropriate discourse conditions). Inspired by Reinhart and Reuland (1993: fn. 49) and Runner (2007), Reuland (2011: 254) presents another solution to the issue, which relies on a modification of PBT Condition A: obligatory coargument binding only applies to eventive predicates. 19 What this hypothesis predicts for PNAs depends on the extent to which (some) picture nouns can be treated as eventive. Reuland (2011: 254-255, 381-382, 17 Depending on the specific account (see fn. 8), this prediction results from Pollard and Sag's (1992: 266) hypothesis that the possessor is the least oblique argument within an NP, or from Reinhart and Reuland's (1993: 682-683) hypothesis that the presence of a subject in the NP causes the noun to form a syntactic predicate. These two versions of PBT also make different predictions about anaphors that are themselves the possessor: they are exempt anaphors in Pollard and Sag (1992: 264-266), but plain anaphors in Reinhart and Reuland (1993: fn. 4, 39; but see fn. 18). We predict that English possessive anaphors like her/his/its own and each other's exhibit a dual (plain/exempt) behavior (cf. French son propre in Charnavel and Sportiche 2016), but a detailed investigation of this prediction is beyond the scope of this article, which focuses on PNAs in the narrow sense. 18 Under that revised assumption, possessed PNAs are thus exempt, and the variation in judgments, Reinhart and Reuland (1993: 683) further suggest, may be due to the discourse accommodation required by logophors (cf. Kuno 1987: 75, 169). Pollard and Sag (1992: 278) also mention some examples of possessed PNAs bound from outside their DP, which are discussed in Zribi-Hertz (1989) and attested in the works of various British writers; according to them, the acceptability of these anaphors is to be related either to differences among varieties of English with respect to Condition A or to the possibility of relaxing Condition A in highly stylized narrative. 19 Reuland (2011: 255) speculates that having an eventive role and projecting a subject may be two sides of the same coin. In that case, the revised formulation of Condition A based on predicates with an eventive fn. 7-8) does not investigate in detail the question, but suggests three possibilities. Under the first option, nouns never have an eventive role, which makes the same prediction as the previous hypothesis: all possessed PNAs are exempt. Under the second option, only Grimshaw's (1990) complex event nominals (i.e. nouns that have an internal aspectual analysis) are eventive; this predicts that possessed PNAs are generally exempt, except for anaphors in complex event nominals. This prediction is supported by the contrast between sentences like (32a) and sentences like (33a-33b). 20 (33) a. *Jill i found Matt k 's fear of herself i surprising. b. *Joanna i was irritated by Mark k 's pride in herself i . (Runner 2007: 83;Sturgeon 2003: 508, fn. 10) Finally, under the third option, only concrete nominals, which denote a physical object, lack an event role. Concrete nominals contrast not only with complex event nominals, but also with result nominals, which denote the outcome of an event. Applied to picture nouns, the concrete/result distinction tracks with a distinction in sense (see Davies and Dubinsky 2003): picture nouns pattern with concrete nominals when referring to a physical object, but pattern with result nominals when referring to informational content. Hence, this third option predicts that possessed PNAs are exempt only when the picture noun is object-referring. Reuland (2011) mentions as support for this hypothesis the contrast between (34a) and (34b) role is equivalent to the original formulation based on predicates with a subject, and the discussion in the text amounts to the following question: to which extent do (some) picture nouns have a subject? We will reexamine this question in Sect. 3.2 from a different angle. 20 Note that while fear and pride denote psychological states rather than dynamic events, they nevertheless qualify as complex event nominals: they take the same arguments as their corresponding verbs (e.g. fear) and pass other diagnostics for complex event nominals proposed by Grimshaw, for instance permitting aspectual modifiers (e.g. Matt's momentary fear of her). 21 According to Runner (2007) (reinterpreting Davies and Dubinsky's 2003 proposal that takes result nominals to have non-argument participants and concrete nominals to have no participants), these two different interpretations of picture nouns correlate with two different argument structures: result nominals, but not concrete nominals, take arguments. Under his version of PBT Condition A (requiring binding by a higher coargument as in Pollard and Sag 1992), this predicts (just like Reuland's 2011 third hypothesis discussed in the text) that only possessed PNAs within concrete nominals are exempt. As we will discuss in Sect. 3.2.4, example (34b), which is supposed to confirm the prediction, presents a crucial confound: it involves the creation verb write.

Re-examining the locality constraints of possessed PNAs
The difficulty in pinning down the status of possessed PNAs is due to the controversy surrounding several variables simultaneously: the definition of Condition A, the potential conditions for exemption from it, the status of the so-called possessor. As in the case of possessorless PNAs, Charnavel and Sportiche's (2016) inanimacy-based tool can be used to at least partially settle the issue by providing a criterion independent of Condition A to tease apart plain and exempt anaphors. 22 Specifically, the inanimacy-based tool can be used to test the hypothesis of late PBTs that possessed PNAs are exempt (at least with non-eventive nouns; see Sect. 2.1.2.1). Recall that PBTs take anaphors to be exempt in the absence of syntactic coarguments, and that exempt anaphors are assumed to be subject to discourse conditions relating to perspective; hence, if the possessor does not comprise a syntactic coargument of a possessed PNA, then possessed PNAs are predicted to be licensed only if logophoric. Given that, as we saw, inanimates are non-perspectival and thus ineligible for exemption, this hypothesis would entail that inanimate possessed PNAs should never be acceptable (at least with non-eventive nouns). But the contrast between (35) and (36) shows that this prediction is not borne out: only inanimate possessed PNAs that are bound across the possessor are unacceptable. 23 This is true even in concrete nominals ((35b) vs. (36b)), contrary to the most conservative hypothesis of Reuland (2011), under which only concrete nouns are treated as non-eventive. Hence, it is neither the case that possessed PNAs uniformly behave as exempt, nor that the locality constraints on possessed PNAs depends on the nature of the noun predicate, as implied by the latest PBTs. Rather, the behavior observed of possessed PNAs depends on the logophoric potential of the reflexive: just like possessorless PNAs, possessed PNAs exhibit a dual behavior (plain vs. exempt), irrespective of the interpretation of the picture noun. Specifically, non-perspectival PNAs like itself in (35)-(36) must be bound by the possessor, while perspectival PNAs like himself /herself in (32a) and (34a) need not be bound by the possessor. This observa-22 There has been a previous attempt to use the notion of perspective in order to clarify the status of possessed PNAs: to test the hypothesis that possessed PNAs are uniformly exempt, Kaiser et al. (2009) aim to check whether possessed PNAs preferably refer to sources rather than perceivers of information.
But the results of this experiment are inconclusive for two reasons: first, it uses a choice task rather than a grammaticality judgment task; second, Charnavel (2020a: 153-154) shows that the notion of source of information is in fact irrelevant to the notion of perspective that licenses exemption. 23 As suggested by an anonymous reviewer, the acceptability of the sentences in (36) could perhaps be predicted under Reinhart and Reuland's (1993) theory of chains. However, this theory cannot account for cases in which an inanimate PNA occurs within a coordinated structure, e.g. Mary polishes the castle's replicas of itself and the neighboring church (cf. (20a); see Reinhart and Reuland 1993: 705). Hence, possessed inanimate PNAs cannot be straightforwardly explained by existing predicate-based analyses even if binding via chain formation is taken into consideration.
tion supports the Chomskian notion of locality revisited by Charnavel and Sportiche (2016): the presence of a so-called possessor does turn the DP into a binding domain. This is not due to the argumental status of the possessor as implied by early PBTs (and correctly questioned by late PBTs; see discussion in Sect. 3.2.2). Rather, formation of a binding domain follows from the position of the possessor in the specifier of DP, which entails the formation of a spellout domain (see Charnavel and Sportiche 2016). This conclusion about the dual behavior of possessed PNAs is further supported by the observation that non-exhaustive binding of possessed PNAs is only possible with animates. 24 (37) a. Mary i looks like Sue k in the library's picture of themselves i+k . b. *[The museum] i looks like [the castle] k in the library's picture of themselves i+k .
(38) a. Mary i will soon buy Sue k 's sculpture of themselves i+k . b. *[The castle] i will soon contain [the museum] k 's replica of themselves i+k .
As captured by the contrasts in (37) and (38), possessed PNAs can behave as exempt by licensing split antecedents if animate (cf. Helke 1970: 116), but exhibit plain behavior by requiring exhaustive binding if inanimate: whether the picture noun is interpreted as a result nominal (as in (37b)) or a concrete nominal (as in (38b)), split antecedents for inanimate PNAs are ruled out. Thus, possessed PNAs are no different from possessorless PNAs in displaying both plain and exempt behavior (see generalization (26)) as summarized in Table 3.

The analytically uniform behavior of English PNAs: Extending the logophoric A-binder hypothesis
In Sect. 2.1, we showed that just like any anaphor, both possessorless and possessed PNAs can descriptively behave as plain or exempt anaphors. We obtained this result by applying Charnavel and Sportiche's (2016) inanimacy-based tool, which allowed us to determine the distributional properties distinguishing exempt from plain anaphors in a reliable way. This result challenges all previous theories: first, contrary to Chomskian predictions, PNAs can superficially be exempt from Condition A even under non-agentive interpretations, that is, when the referent of the anaphor is not the author of the picture; second, contrary to PBTs, the dividing line between plain and exempt anaphors does not lie in the presence of a coargument (whether the possessor counts as such or not), but in the perspectival interpretation of the anaphor; finally, contrary to the pervasive assumption represented by the claim in Drummond et al. (2011) above, that PNAs are logophoric pronouns (homophonous with anaphors), PNAs do not uniformly display pronominal properties, but can behave like plain anaphors.
The goal of the present section is to account for this apparently dual behavior of PNAs without appealing to homophony or restricting the scope of Condition A. We will reach this goal by adopting Charnavel's (2020a, 2020b) logophoric A-binder hypothesis-thereby further supporting it by demonstrating that it also makes correct predictions for English PNAs. In the spirit of Chomsky (1986), this hypothesis retains the general applicability of Condition A by assuming the possible presence of implicit binders for anaphors. However, it introduces a new type of covert binder, which, unlike Chomsky's PRO-like subject, does not entail an agentive interpretation, but derives the perspective-based contrasts observed above.
We begin by introducing Charnavel's (2020a, 2020b) logophoric A-binder hypothesis in further detail in Sect. 2.2.1. Then, we extend this hypothesis to English PNAs: just as in Sect. 2.1, we will first concentrate on possessorless PNAs in Sect. 2.2.2, before applying the analysis to possessed PNAs, which raise further challenges, in Sect. 2.2.3.

The logophoric A-binder hypothesis
According to Charnavel's (2020aCharnavel's ( , 2020b hypothesis, descriptively exempt anaphors are in fact not exempt from Condition A, but are locally bound by a covert logophoric pronoun. Thus, the properties that characterize descriptively exempt anaphors do not come from the anaphors themselves, but derive from the nature of their binder. Recall from Sect. 2.1.1.3 (see (27)) that we adopt a version of Condition A according to which anaphors must be bound within the minimal spellout domain containing them. Based on the same assumption, Charnavel (2020aCharnavel ( , 2020b hypothesizes that each spellout domain can contain a verb-like logophoric operator OP LOG introducing a logophoric pronoun pro log as its subject. This is represented in (39a), in which pro log locally binds a PNA. Note that the category of the spellout domain shown in (39a) will depend on the syntactic configuration in which the PNA occurs: if the DP containing the PNA also contains a subject, then the DP will be the spellout domain relevant for binding (and can contain pro log as a potential binder 25 ); if the DP lacks a subject, then the spellout domain will instead be the smallest phrase containing both the DP and a subject (or the smallest tensed TP containing the DP if the DP is the subject of a tensed TP). While pro log refers to the locally relevant logophoric center, OP LOG imposes the first-person perspective of that center on its complement α, as formulated in (39b). This hypothesis codes the intuition that the locally relevant perspective center, which is independently determined by a combination of syntactico-semantic and discourse factors, can be syntactically represented in each phase. Charnavel's hypothesis is inspired by the literature on logophoric operators and perspectival projections (see Koopman and Sportiche 1989;Kinyalolo 1993;Jayaseelan 1998;Speas and Tenny 2003;Adesola 2006;Anand 2006; i.a.), but differs from it in two main respects. First, it builds on Sells (1987) and Oshima (2006) in proposing a specific definition of logophoricity as first-person mental perspective (encompassing de se attitude and empathy) on the basis of anaphora-independent tests such as the epithet test for de se attitude shown in (40) and the French possessive cher test for empathy adapted to English in (41); this methodology circumvents the aforementioned issue regarding how to characterize the notion of logophoricity.

(40)
Epithet Test for detecting attitude holders in their attitude contexts: To simultaneously check whether a given DP 1 is in an attitude context and who the relevant attitude holder is, replace DP 1 with an epithet and determine its referential possibilities in unmarked situations (i.e. without using non-de se scenarios). If there is a DP 2 that does not locally-c-command the epithet but which the epithet cannot take as antecedent, then the epithet (and DP 1 it replaced) are in an attitude context and the referent of DP 2 is the attitude holder of that context. (cf. Charnavel 2020a: 146) (41) Possessive dear Test for detecting empathy loci in their empathy contexts: To identify the possible empathy loci in a context containing a given DP, replace this DP with a possessive DP containing dear and determine its referential possibilities. If her dear is acceptable, its referent can be construed as the empathy locus of the context of the DP. Otherwise, only the speaker can be interpreted as the empathy locus.
(cf. Charnavel 2020a: 169) Second, it makes two modifications to the syntactic representation of logophoricity by restricting logophoric domains to spellout domains (instead of full CPs as previously assumed; see Charnavel 2020b: 709-711 for discussion) and by treating OP LOG as a verb-like operator introducing a subject (pro log ); this twofold innovation crucially entails that pro log can serve as an A-binder for anaphors. Under Charnavel's hypothesis, apparently exempt anaphors like herself in (39a) are thus in fact bound locally (i.e. within their spellout domain) by the implicit logophoric pronoun pro log . This predicts that descriptively exempt anaphors must be logophorically interpreted, and thus derives (instead of postulating) the correlation between logophoricity and exemption observed in many unrelated languages (see e.g. French elle-même and son propre, Korean caki-casin, Mandarin ziji or Icelandic sig, 26 as discussed in aforementioned references and reviewed in Charnavel 2020a). 27

Testing for the logophoric interpretation of exempt PNAs
Applying Charnavel's (2020a, 2020b) hypothesis to our specific case study, we propose that descriptively exempt English PNAs are analytically plain, covertly complying width Condition A by virtue of binding by pro log . This treatment entails that descriptively exempt PNAs must be logophorically interpreted (under a non-agentive interpretation 28 ) in the sense explained above. This correctly predicts our finding in Sect. 2.1.1.3 that inanimate PNAs are unacceptable unless locally bound by an overt binder: due to the incompatibility between inanimacy and mental perspective, inanimate PNAs cannot take pro log as an antecedent. As we will now illustrate with possessorless PNAs, this also correctly predicts that animate PNAs that lack an overt local binder can only occur in phrases expressing the first-person mental perspective of their antecedents.
This first means that seemingly exempt PNAs must pass the tests described in (40)-(41). For example, neither himself in the complement of the attitude verb believe in (1) (repeated as (42a)), nor himself in the intended Free Indirect Discourse in (12) (repeated as (43a)) can be replaced with a coreferential epithet, as shown in (42b)-(43b); this demonstrates that the antecedents of these anaphors are relevant attitude holders in the clause containing them. 29 (42) a. Tom i believes that there is a picture of himself i hanging in the post office. b. *Tom i believes that there is a picture of [the idiot] i hanging in the post office.
(43) a. John i was going to get even with Mary. That picture of himself i in the paper would really annoy her, as would the other stunts he had planned. b. #John i was going to get even with Mary. That picture of [the idiot] i in the paper would really annoy her, as would the other stunts he had planned.
Himself in (11) (repeated as (44a)) can be replaced by an epithet, as shown in (44b), but also by a possessive DP containing dear as in (44c); this reveals that its antecedent, John, cannot be construed as an attitude holder, but can be construed as the empathy locus in the phrase containing himself. However, himself in (7) (repeated as (45a)), which is unacceptable, can alternate with a coreferring epithet, but not (under a non-ironic reading) with a possessive DP including dear (see (45b-45c); this shows that the antecedent of himself neither refers to an attitude holder, nor to an empathy locus, which makes himself non-logophoric and unable to be bound by pro log . Second, the phrase containing exempt PNAs must express the first-person perspective of their antecedent. For instance, this predicts that himself in (42a), which refers to an attitude holder, must be read de se, as confirmed by (46) Similarly, this implies that John's act described in (43) must be considered as a stunt by John himself; for example, the sentence is infelicitous if only the speaker considers this act as a stunt, but John considers it as an act of kindness.

Deriving the distributional properties of exempt PNAs
We have shown that the logophoric A-binder hypothesis correctly predicts the logophoric interpretations required for descriptively exempt PNAs. Adopting this hypothesis also allows us to derive the distributional properties that distinguish descriptively exempt from plain PNAs. As captured in (39) above, the apparent absence of locality constraints for exempt PNAs follows from the implicitness of their binder, pro log , which like any pronoun need not be locally bound nor even take an overt antecedent in the sentence. Similarly, the availability of non-exhaustive binding for exempt PNAs is an illusion due to the pronominal nature of their binder. As shown in (47) below (repeating example (13)), themselves, which descriptively takes a split antecedent (John and Mary), is in fact exhaustively bound by pro log . The pronominal nature of pro log , however, allows it to take a non-exhaustive antecedent; the Ewe example in (48) independently confirms that logophoric pronouns do not differ from other pronouns in this respect. Apparent non-exhaustive binding of the anaphor is thus analyzed as non-exhaustive antecedence of the exhaustive binder of the anaphor.
(47) John i told Mary k that pro log-i+k there were some pictures of themselves i+k inside.
go. out 'Kofi i saw that they i+k had come out.' (Sells 1987: 449) In sum, the logophoric A-binder hypothesis allows us to reduce descriptively exempt PNAs to plain anaphors by deriving all their specific properties from the nature of their binder pro log . Charnavel (2020aCharnavel ( , 2020b provides additional arguments for the logophoric A-binder hypothesis independent of the properties of descriptively exempt anaphors. We show below that these arguments, which she uses to account for the distribution of French exempt anaphors, also apply to the case of English PNAs, thus further supporting extension of the logophoric A-binder analysis. First, just like exempt anaphors in French (see Charnavel 2020aCharnavel , 2020b or Mandarin (see Pan 1997; Huang and Liu 2001;Anand 2006;i.a.), we observe that locally co-occurring exempt PNAs in English must exhaustively corefer. For example, (49) shows that even if himself (in (49a)) and themselves (in (49b)) can be descriptively exempt, they cannot co-occur in the same clause in (49c).

Independent arguments for the logophoric A-binder hypothesis
(49) a. John i told Mary that there was a story about himself i in the newspaper. b. John i told Mary k that there were some pictures of themselves i+k in the newspaper. c. *John i told Mary k that there were some pictures of themselves i+k and a story about himself i in the newspaper.
This ban on non-exhaustive coreference between locally co-occurring exempt PNAs directly follows from the logophoric A-binder hypothesis. Recall that under this hypothesis, descriptively exempt PNAs are in fact plain anaphors and must thus be exhaustively bound within their spellout domain. Given that there is only one possible binder in the spellout domain (TP) containing themselves and himself, namely pro log , both themselves and himself must be exhaustively bound by pro log , which entails that they must exhaustively corefer. 30 Whether pro log syntactically references John as the logophoric center in the embedded clause as in (51a) (cf. (50a)) or the sum of John and Mary as in (51b) (cf. (50b)), one of the anaphors will not be able to be exhaustively bound (namely, themselves in (51a) and himself in (51b)), thus ruling out the sentence.
(50) a. John i told Mary that pro log-i there was a story about himself i in the newspaper. b. John i told Mary k that pro log-i+k there were some pictures of themselves i+k in the newspaper.
(51) a. *John i told Mary k that pro log-i there were some pictures of themselves i+k and a story about himself i in the newspaper. b. *John i told Mary k that pro log-i+k there were some pictures of themselves i+k and a story about himself i in the newspaper.
Note that alternative hypotheses discussed above cannot explain this exhaustive coreference constraint. For example, under the hypothesis that PNAs are in fact logophoric pronouns subject to discourse requirements, ban on disjoint exempt anaphors in the same domain could presumably derive from a pragmatic principle ruling out perspective conflicts (cf. Pan 1997; Huang and Liu 2001, i.a.). But this explanation cannot hold for (49c), which does not involve disjoint, but partially coreferential anaphors, and thus does not entail perspective conflict, as attested by the perfectly viable direct discourse counterpart in (52).
(52) John told Mary: "There were some pictures of us and a story about me in the newspaper." This coreference constraint cannot straightforwardly derive either from another version of this alternative hypothesis, according to which PNAs would be logophoric pronouns that must be bound by a logophoric operator (cf. Anand 2006) and at most one logophoric operator per clause can be present (cf. Koopman and Sportiche 1989).
As mentioned above, nothing should prevent a logophoric pronoun from being partially bound; in fact, Adesola (2006) explicitly allows non-exhaustive binding by a 30 The same result would obtain under the assumption that themselves and himself can occupy two different spellout domains (e.g. the DP containing them, if one supposes, contra Charnavel and Sportiche (2016), that spellout domains are not restricted to phrases with subject; recall from fn. 28 that we do not consider DPs with PRO-like subjects in this section). As shown in (50), either John or the sum of John and Mary can be syntactically represented as the relevant logophoric center in the domain (Mary would be a third option). But as argued in Charnavel (2020aCharnavel ( , 2020b, this does not imply that they could be represented simultaneously. Several logophoric pronouns co-occurring in the same clause can syntactically reference different logophoric centers only if the relevant discourse conditions are met, e.g. if the syntactico-semantic conditions introduce a new logophoric center between the two logophoric pronouns; this is exemplified in (i) where the report about the content of Paul's granddaughter's diary introduces her as a new logophoric center within the attitude clause (which has Paul's daughter as attitude holder).  (Charnavel 2020a: 225) logophoric operator. The restriction of one logophoric operator per clause does not therefore entail any ban on partially coreferring logophoric pronouns within the same clause. Second, the logophoric A-binder hypothesis predicts a Condition C effect if a descriptively exempt PNA locally co-occurs with an overt DP that refers to the logophoric center anteceding the anaphor (see Charnavel 2020a: 228). The contrast between (53a) and (53b) indicates that this prediction is borne out. As shown in (54), herself is licensed by the presence of pro log in its spellout domain (the bracketed DP), which syntactically represents the logophoric center, Lucy; the presence of the DP Lucy in the same domain therefore entails a Condition C violation. 31 (53 To wrap up, both anaphora-based and anaphora-independent arguments thus motivate the hypothesis that the possible syntactic representation of implicit logophoric pronouns, which can serve as A-binders, is responsible for the illusion of PNA exemption. This hypothesis parsimoniously reduces apparently exempt PNAs to plain anaphors subject to a fully general Condition A.

The logophoric A-binder hypothesis applied to possessed PNAs
So far, we have focused our analysis on possessorless PNAs. Given that one of the motivations (and consequences) of the logophoric A-binder hypothesis is to unify all instances of anaphors, we also apply it to possessed PNAs, as illustrated in (55) and (56)  Just as in the case of possessorless PNAs, the acceptability of the descriptively exempt anaphor in these cases derives from the presence of pro log with an appropriate reference in its spellout domain, namely in the possessed DP; the relevant interpretation relies on whether the discourse conditions allow the intended antecedent of the anaphor to be construed as the logophoric center in the DP (e.g. as empathy loci in (55)-(56)).  Charnavel (2020a: 228-230) argues that covaluation with pro log can never violate Condition B (coargumenthood seems to be the relevant notion of locality for Condition B, and no pronoun can be a coargument of pro log , since the complement of OP LOG is always a bigger phrase (see α in (39)).
Note that the logophoric center that is syntactically represented in a domain need not be the closest available one in the sentence, as shown by examples like (57) (containing an English exempt anaphor) or (58) (including an Ewe logophoric pronoun). Therefore, the presence of a disjoint animate possessor in sentences like (55) does not necessarily create an intervention effect (but may be responsible for some variability in judgments about possessed PNAs 32 ).
(57) John i asked Bill k who he k thought had stolen the picture of himself i . (Cantrall 1974 'Kofi i believed that Ama k said that he i beat her k .' (Clements 1975: 173)

Descriptive generalization (for all English anaphors including PNAs):
Non-perspectival anaphors must be locally and exhaustively bound, whereas perspectival anaphors may appear not to be.
(60) Condition A: All anaphors (including PNAs) obey Condition A, i.e. must be bound within the minimal spellout domain containing them. Possible Abinders include overt DPs and covert ones such as PRO or pro log .
But the picture is not complete yet: a closer look at the data suggests that this solution is not sufficient. Specifically, we observe that in some configurations, possessed PNAs cannot be logophorically bound even under the appropriate discourse conditions; we will henceforth refer to this observation as the Logophoric Blocking Effect or LBE. This is first the case of PNAs in complements of creation verbs 33 such as himself in (61b) (repeating (34b), cf. (29a)).
(61) a. /?Joe i destroyed Harry k 's book about himself i . b. ?/*Joe i wrote Harry k 's book about himself i . 32 Even if pro log need not refer to the closest logophoric center, the presence of a closer logophoric center may make some readings harder to access and therefore apparently ungrammatical for some speakers. Under this view, the logophoric potential of the possessor is an important factor, which depends on two variables: the nature of the DP occupying the possessor position (whether or not it can-in principle and in the context-be construed as a logophoric center; inanimates never can, for example) and the nature of the head noun (whether or not it can make its possessor a logophoric center in its complement; nouns like opinion or belief can, for instance). 33 Like Davies and Dubinsky (2003) or Jaeger (2004), we hypothesize that the relevant category here is the class of creation verbs. This is not the only hypothesis entertained in the literature: in particular, Jackendoff (1972) characterizes this problematic group of verbs as verbs marking their subject with the thematic relation Agent, while Keller and Asudeh (2001) treat them as [+existence] accomplishment verbs. Jackendoff's (1972) hypothesis seems to be too broad in incorrectly including verbs like destroy in (61a). Keller and Asudeh's (2001) hypothesis seems to be empirically equivalent, but attributes the effect to the aspect of the verb, which does not seem to be the relevant factor (see further discussion in Sect. 3.2.4).
(62) a. *John i took Mary's pictures of him i . b. John i found Mary's pictures of him i . (Williams 1987: 156) The particular behavior of reflexives and pronouns (as in (62)) in this type of syntactic contexts has been long noticed in the literature (Jackendoff 1972: 166-168;Chomsky 1986: 166-167;Williams 1987: 155-156;Reinhart and Reuland 1993: 685;Runner 2002;Davies and Dubinsky 2003: 25-27;i.a.) and has more recently been experimentally investigated Jaeger 2004;Bryant and Charnavel 2020;i.a.). The logophoric A-binder hypothesis is too weak to account for the contrast in (61): just like in (55) above, it predicts that himself in (61b) can be bound by pro log , which should be able to refer to Joe, as implied by (61a). Another constraint must thus be responsible for ruling out (63).
(63) *Joe i wrote [pro log-i Harry k 's book about himself i ].
(64) *The fact that Mary k 's description of himself i was flawless was believed to be disturbing John i .
(65) *Jill i found Matt k 's fear of herself i surprising.
Finally, novel experimental findings presented in Bryant and Charnavel (2020) reveal that possessed PNAs cannot be logophorically bound either when they stand as the goal argument of the noun, as illustrated in (66).
(66) *Chelsey i found Brandon k 's letter to herself i . (Bryant and Charnavel 2020: 11-12) In our view, these three types of data fall under the same category and demonstrate that an explanatory factor is yet to be added to the logophoric A-binder hypothesis to reach a full resolution of the PNA puzzle. The goal of the second and last part of the paper is to specify the nature of this additional factor that interacts with the logophoric A-binder hypothesis.
To give a preview, the additional factor that gives rise to LBE does not specifically target possessed PNAs: we hypothesize that the unacceptability of (63)-(66) results from the same constraint as the unacceptability of (5b) (vs. (5a)) repeated in (67b) (vs. (67a)), which involves a reflexive as direct object of the verb.
a. Max i boasted that the queen invited Lucie and himself i for a drink. b. *Max i boasted that the queen invited himself i for a drink. (Reinhart and Reuland 1993: 670) The ungrammaticality of examples like (67b), which seems to be responsible for the widespread assumption of PNA exceptionalism, remains another crucial outstanding issue that appears to undermine the logophoric A-binder hypothesis beyond PNAs. Recall indeed from Sect. 2.1.1.1 that, typically, English anaphors can be descriptively exempt only under some configurations: in particular, when they are within picture noun phrases (e.g. picture of himself ), as well as within a conjoined DP (e.g. Lucie and himself as in (67b)), within like-phrases (e.g. physicists like herself ), within as for-phrases (e.g. as for herself ) or within exceptive constructions (e.g. no one but herself ). This observation is what motivated the development of PBTs, which tie exemption to the absence of coarguments. Now, we have explained at length (especially on the basis of inanimate anaphors) why it is empirically incorrect to base the dividing line between plain and exempt anaphors on coargumenthood. We therefore do not intend to reincorporate the notion of coargumenthood into our account of apparent exemption from Condition A. But we will show that, at least descriptively, this notion does indirectly play some role in the licensing conditions on reflexives.
Specifically, we will conclude that the factor descriptively responsible for LBE in (67b) is the presence of a coargumental subject: the queen blocks logophoric binding in (67b), but not in (67a), because it is a subject coargument of himself only in (67b). And the reason why logophoric binding is excluded in the presence of coargumental subjects is because this configuration licenses alternative pronominal elements that are weaker and yield the same interpretation. We will thus hypothesize that LBE falls under a general principal of competition between weaker and stronger forms à la Cardinaletti and Starke (1999), which is fully independent of Condition A or exemption. (68) Logophoric Blocking Effect (LBE): Herself cannot be logophorically bound in the presence of a coargumental subject.
(69) Weak/strong competition (Cardinaletti and Starke 1999): Weaker forms exclude stronger forms if they can yield the same interpretation.
Crucially, we will see that the same principle of competition can derive LBEs for PNAs in examples such as (63)-(66) once we clarify two issues specific to the nominal domain, namely the distinction between subjects of NP and other sources for possessors, and the conditions on implicit projection of nominal subjects. We will conclude that the ungrammaticality of (63)-(66) ultimately results from the obligatory presence of an implicit subject in NP in those cases. Besides solving the PNA puzzle, this investigation of LBE will thus provide a new probe into the controversial argument structure of NPs.

Deriving the logophoric blocking effect: The weak/strong competition hypothesis
This second part of the paper aims at explaining why the empirical scope of the logophoric A-binder hypothesis seems to be restricted, namely why logophoric binding is impossible for PNAs in some configurations, e.g. (63)-(66), in spite of favorable discourse conditions. First, we will explore such Logophoric Blocking Effects (LBEs) in the verbal domain, e.g. (67b): in Sect. 3.1, we will build on Ahn (2015) to establish the empirical generalization capturing the conditions under which LBEs arise for anaphors in verbal complements, and we will derive this generalization from a weak/strong principle of competition inspired by Cardinaletti and Starke's (1999) work. Only then will we be in a position to examine the consequences of this hypothesis in the nominal domain, where it interacts with additional factors: in Sect. 3.2, we will examine its predictions for both possessorless and possessed PNAs and thus solve the remaining cases of LBEs, e.g. (63)-(66). We will thereby open new avenues for the investigation of nominal structures.

The logophoric blocking effect in the verbal domain
The goal of this section is to derive LBEs in the verbal domain (i.e. outside PNAs), where the generalization that will ultimately be relevant to PNAs is easier to establish in the absence of complicating factors specific to the nominal domain. In a nutshell, we will show, using prosody as a diagnostic, that logophorically bound herself is necessarily a strong form. Due to a principle of competition between weaker and stronger forms, LBEs therefore arise when anaphors occur in positions that can host weak elements, typically in positions with a coargumental subject. We will begin in Sect. 3.1.1 by introducing Ahn's (2015) empirical observations regarding the prosodic behavior of English reflexives. This allows us to distinguish two cases of descriptively plain English anaphors (i.e. anaphors that overtly obey Condition A): those that are strong, and those that are weak. We also summarize Ahn's account of this distinction, which will serve as the starting point for our analysis of LBEs. In Sect. 3.1.2, we motivate a proposal that incorporates Ahn's insights with the general competition principle of Cardinaletti and Starke (1999). As we will show, this combination of ingredients allows us to account for LBEs in English without restricting Condition A or appealing to homophony.

Two types of plain anaphors in English
3.1.1.1 Ahn's (2015) empirical discovery To understand why logophoric binding is blocked in examples like (67b), we will reexamine the behavior of anaphors through a different lens than Condition A, namely prosody, as done in Ahn (2015) (cf. Spathas 2010: Chap. 3 34 ). Specifically, Ahn (2015) observes that English plain reflexives descriptively fall into two classes: those that exhibit exceptional prosodic behaviors, and those that do not. To test the prosodic behavior of reflexives, Ahn examines them in positions where other elements bear nuclear phrasal stress in maximally broad-focus contexts (i.e. contexts in which they are neither given nor contrastively focused), as illustrated in (70) Ahn (2015) and Spathas (2010) independently establish very similar empirical generalizations about the prosody of English reflexives. While these facts lead Ahn (2015) to posit a reflexive voice (as we will explain in Sect. 3.1.1.2), Spathas (2010) concludes from them that English (vs. Greek) plain reflexives (under PBTs' theory of the plain/exempt distinction) should be treated as reflexivizing functions rather than variables. Although both theories correctly derive the main prosodic facts, we here adopt a modified version of Ahn's theory because it is compatible with the logophoric A-binder hypothesis (see Ahn 2015: 186-188 for further advantages of Ahn's theory over Spathas'). Spathas' theory, on the other hand, explicitly posits homophony between coargument bound reflexives and the other reflexives since this analysis implies that they must have a different semantics. 35 In these examples, we follow Ahn's (2015) notation in indicating the locus of phrasal stress with bolded underlined italics, and infelicity due to information structure (question-answer congruence) with the sign #. Ahn  On the basis of anaphora-independent data, Ahn (2015) demonstrates that, in neutral contexts, English phrasal stress is received by the most deeply embedded constituent in a spellout domain. 36 Whereas nuclear stress therefore typically falls on the direct object in basic transitive sentences (e.g., on Marie in (70)), it may not fall on himself in (71), or else the sentence is rendered infelicitous. Ahn calls anaphors that exhibit such exceptional prosodic behavior extrametrical; we will refer to them as weak. Importantly, not all anaphors are weak, as illustrated in (72), in which conjoined himself bears nuclear stress; such reflexives are strong, patterning with referential DPs. This immediately excludes the null hypothesis that reflexives never bear phrasal stress because they are given as a result of necessitating an antecedent.
Here we can already notice a link between prosodic behavior and logophoric behavior: the configurations that license logophoric binding in English (e.g. (67a)) resemble the configurations that exclude weak anaphors (e.g. (72)); conversely, the configurations that exclude logophoric binding (e.g. (67b)) resemble the configurations that can host weak anaphors (e.g. (71)). Ahn (2015) does not discuss logophoric anaphors and, in fact, leads to wrong predictions about them. We will nevertheless see that Ahn's proposal provides the crucial clue to derive LBEs in English: logophorically bound anaphors cannot be weak.

Ahn's (2015) account: Movement to reflexive Voice
This section offers an overview of Ahn's (2015) account of the prosodic facts introduced above. While we will ultimately depart from certain details of Ahn's proposal, two aspects will carry over to the present proposal: the appeal to movement, and the appeal to a covert reflexivizing head distinct from the anaphor.
As a first step to analyze the contrast between (71) and (72), Ahn (2015) identifies three syntactic configurations in which anaphors are strong. 37 First, English anaphors 36 More precisely, Ahn (2015: 294) claims that a syntactic object, X, is more deeply embedded than some other syntactic object, Y, provided that no copy of X c-commands all copies of Y. To support the hypothesis that structure, not linear order, is the input for determining the locus of phrasal stress, Ahn (2015: 78-81) examines several phenomena independent of reflexivity that show dissociation between linearization and phrasal stress. 37 Another apparent case of strong reflexives occurs in double object constructions, where subject-bound direct object reflexives bear phrasal stress in broad focus contexts. But Ahn (2015: 127-129, 194-195) argues that such reflexives should in fact be analyzed as weak reflexives (i.e. they move to Reflexive Voice, see Sect. 3.1.1.2), but prosody-specific constraints impose stress on them in those cases (because the direct object of a double object construction forms its own prosodic domain). bear stress when they are separated from their antecedent by an island boundary. This was the case for himself in (72), which occurs in a coordinated structure; this is also the case for himself in (73), which appears in another type of island. To account for these findings, Ahn (2015) is inspired by the similar behavior of the French reflexive clitic se and Sportiche's (2014) analysis of it (as will become clearer in Sect. 3.1.2.1). Specifically, Ahn posits the presence of a reflexive Voice head (REFL), which is endowed with an EPP feature that obligatorily attracts a reflexive argument. In a sentence like (71b), himself undergoes "covert overt" movement 38 to the specifier of the REFL, as shown in the simplified representation in (77) Ahn's appeal to movement directly derives the prosodic weakness of himself in sentences like (71b): because himself undergoes movement to the specifier of VoiceP, it is not the most deeply embedded constituent in its spellout domain and, hence, does not bear nuclear stress. Movement also derives the fact that weak reflexives cannot be separated from their antecedent by an island boundary, as seen in (72)-(73): because himself occurs in an island, it cannot move to VoiceP, as shown below in the simplified representation of (72).
As we will detail in Sect. 3.1.1.3, Ahn assumes in such cases (i.e. when the presence of a reflexive Voice yields ungrammaticality) that reflexives can be licensed in a derivation without reflexive Voice. In the absence of a reflexive Voice, himself in (72)-(73) remains the most deeply embedded constituent of its spellout domain, leaving it the target for nuclear stress. Furthermore, association of reflexive movement with reflexive Voice head accounts for why weak reflexives cannot be anteceded by a derived subject. For example, the passive and raising constructions in (74)-(75) rely on a Voice other than the reflexive Voice (i.e. passive, raising; see Ahn 2015: 106-108). Given that only one Voice can be merged per clause, there is no reflexive Voice to trigger movement of himself in these examples; thus, Ahn here assumes a derivation without reflexive Voice, where himself remains low in the spellout domain and therefore bears nuclear stress. Note that the height of the reflexive Voice further derives the obligatory subject orientation of weak reflexives, under the assumption that the subject is the only DP within the relevant spellout domain that outscopes the reflexive Voice projection. 39 Thus, the reflexive in (76) can only take the subject Liz as antecedent after movement to VoiceP; binding of himself by the object Danny is only possible in a derivation without reflexive Voice.
In short, Ahn (2015) distinguishes between two types of reflexive constructions, which can be diagnosed by their prosody in maximally broad-focus contexts in positions normally bearing nuclear phrasal stress: those that involve a reflexive Voice head (where the reflexive is weak) and those that do not (where the reflexive is strong). The former are only possible in contexts of local subject-oriented reflexivity (or LSOR, in Ahn's terms), whereas the latter are possible in the complement set of these contexts. Crucially, the dividing line between LSOR and non-LSOR constructions does not rely on the position occupied by reflexives. Certainly, there are positions that systematically exclude LSOR reflexives (e.g. within an island excluding the antecedent, as in (72)) and others that systematically exclude non-LSOR reflexives (e.g. direct object of a verb, as in (71)). But there are also positions that are compatible with both LSOR or non-LSOR reflexives, such as the indirect object position in configurations like (76); in those cases, the difference between LSOR and non-LSOR anaphors is interpretive: weak reflexives are subject-oriented, while strong reflexives are objectoriented.

Ahn's overgeneration of logophoric anaphors
Returning now to our main goal of accounting for LBEs, Ahn's (2015) analysis makes specific predictions about logophoric anaphors, even if they are not discussed. Central to this is Ahn's (2015: 290) Rule J, a pragmatic competition principle according to which reflexive Voice must be merged if (i) its presence is grammatically possible, and (ii) its presence does not change the interpretation. Clause (i) correctly predicts the complementary distribution of LSOR and non-LSOR reflexives in configurations like (71)  Ahn's hypotheses imply that logophorically bound reflexives in examples like (67a) (repeated below as (79a)) are strong, as they are not local-subject oriented. As shown in (79b), such constructions are correctly permitted under Rule J: reflexive Voice cannot merge since, just like in (72a), the reflexive sits in an island (the coordinated structure) excluding its antecedent. However, the predictions Rule J makes are too weak to rule out (67b) (repeated below as (80a)). As shown in (80b), the derivation with reflexive Voice is grammatical. But as per clause (ii), it does not exclude the derivation without reflexive Voice in (a) because it does not trigger the same interpretation (cf. (76b-76c)). Contrary to facts, himself in (80a) is thus predicted to be acceptable under neutral phrasal stress. This option is not explored in Ahn (2015), which limits discussion to derivations that overtly comply with Condition A. Specifically, in the examples considered in Ahn's analysis, 40 if a reflexive Voice is merged, only the local subject can bind the anaphor because it is the only DP above the reflexive Voice that sits within the spellout domain; and if the reflexive Voice cannot be merged, only subjects and objects within the spellout domain are potential antecedents. But crucially, the logophoric A-binder hypothesis adds a further option by introducing pro log as a new potential binder within the spellout domain. Given that in configurations like (80), antecedence by pro log and antecedence by the local subject yield different interpretations, both are predicted to be available, since Rule J does not force the merging of the reflexive Voice if a derivation without it does not give rise to the same meaning. 41 Thus, Ahn's (2015) account cannot derive LBEs arising in sentences like (80a). Though Ahn's Rule J incorrectly predicts the distribution of logophoric anaphors in English, it suggests an important ingredient to solving the LBE issue. In particular, we will argue in the next section for a different kind of competition that derives both the distribution of LSOR reflexives and LBEs in English, namely a general competition between strong and weak forms.

The weak/strong competition hypothesis
In Sect. 3.1.1 we introduced Ahn's (2015) proposal distinguishing between weak and strong reflexives, which we will henceforth note 'erself and herself, respectively. This distinction was attributed not to the reflexives themselves, but to the syntactic configuration in which they occur: weak 'erself is associated with a reflexive Voice head to which it is attracted, but strong herself is not associated with any head. 42 Moreover, the availability of weak 'erself was argued to block the presence of strong herself under the same interpretation. In this section, we will explain how this distinction can help us understand why the distribution of logophoric anaphors is restricted in English. Specifically, we will propose that competition arises not only between strong and weak reflexives but, rather, between strong and weak pronominals more generally. To motivate this proposal, we will compare the English facts to the facts in French, where competition between strong and weak forms is morphologically apparent.

Insight from French
Just as English distinguishes between weak 'erself and strong herself, French distinguishes between a weak reflexive form, se (a clitic), and a strong reflexive form, elle-même. Recall from Sect. 3.1.1.2 that Ahn's proposal 41 An alternative way to integrate the logophoric A-binder hypothesis to Ahn's (2015) account would be to hypothesize that pro log sits above the reflexive voice and can thus serve as antecedent for weak reflexives. This would also wrongly predict (80a) to be grammatical. We only discuss the option under which logophorically bound herself is not associated with the reflexive voice because this is ultimately the correct one: as we will see, logophorically bound reflexives are strong. What needs to be refined is the kind of competition introduced by Rule J. 42 Recall that this contrast was diagnosed based on the prosody of reflexives in positions normally bearing nuclear stress in broad focus contexts. Otherwise, note that strong herself can surface without phrasal stress when it is not the most embedded element of its spellout domain or when it is given, while weak 'erself can bear stress under narrow focus (see Ahn 2015: Chap. 4 on REAFR) or when it is the direct object in a double object construction (see fn. 37). Ahn's account correctly captures the fact that the prosodic difference between herself and 'erself only surfaces in some configurations: their actual difference is structural.
for English 'erself was inspired by parallels with French se. Indeed, the distribution of French se is strikingly similar to that of English 'erself since it, too, requires a local, deep subject as antecedent. It is therefore ungrammatical in the three types of configuration identified in Ahn (2015) to require herself (vs. 'erself ). 43 First, se cannot occur in islands excluding its antecedent as in (81) (85) and (71)) and (ib) (to be compared with (84) and (76) The presence of se here supports the hypothesis that the reflexive Voice is involved in English in these cases, even if narrow focus induces strong prosody on 'erself. Further note that clitic doubling is not obligatory in (ib) (vs. ia). According to Kayne (2000), the version without clitic doubling is an instance of topicalization (which is more available with datives than with accusatives) and is therefore irrelevant to our purposes (see further discussion in fn. 54). 44 To facilitate presentation, we gloss the strong reflexive lui-même as himself and the reflexive clitic as SE, and we place an index on se as if it stood for an argument (but we do not take a stand on its actual role, see Labelle 2008; Sportiche 2014, i.a. for discussion). Further note that the presence of se obligatorily induces the auxiliary be. 45 In (82b), the addition of the parenthesis is meant to make sure that the sentence is interpreted as a passive (an alternative, irrelevant interpretation is otherwise available, i.e. Liz assigned herself ); also, the choice of the auxiliary reflects the only possible combination of auxiliaries possible with se (see Charnavel 2008). Furthermore, the experiencer is right-extraposed in (83a) because this is the only position where it can be acceptable in French; it is not perfect though because non-clitic experiencers are generally degraded in French. Notice in (84a-84b) above, as well as in (85) below, that the availability of the reflexive clitic se blocks the strong reflexive elle-même in broad focus contexts (see Charnavel and Sportiche 2016: 54-55), just as the availability of 'erself blocks herself in English (cf. (71)- (72) As we mentioned, the striking parallel between the distribution of se and 'erself inspired Ahn (2015) to analyze 'erself as undergoing a clitic-like movement to a reflexive Voice. Similarly, the competition between elle-même and se should serve as a basis, we suggest (cf. Charnavel and Sportiche 2016: 57-58), for analyzing the division of roles between herself and 'erself. Specifically, Charnavel and Sportiche (2016: 53-57) claims that the blocking of elle-même by se falls under a generalization proposed by Cardinaletti and Starke (1999): all else (relevant) being equal, if a weaker form of the target element is available, it must be used (and thus blocks the use of a stronger form). We thus hypothesize that strong herself is blocked by weak 'erself, recasting Ahn's (2015) Rule J as one particular subcase of a very general principle of economy driven competition between weaker and stronger forms.
This kind of competition is fully independent of Condition A; in fact, it does not only affect reflexive elements, but also pronouns and adverbs across languages. For instance, Charnavel and Sportiche (2016: 55-56) show that in French, strong pronouns like elle are also blocked by clitics, just like strong reflexives. Crucially for our purposes, Charnavel and Sportiche (2016) further show that elle-même is not only blocked by the reflexive clitic se, but also by accusative and dative clitics. 46 Thus, lui-même in (86b) (cf. himself in (67b)) is not ruled out by the availability of se, which yields a different interpretation (86c), but by that of the accusative clitic le (86a). With this insight into the nature of strong/weak pronominal competition, we are finally in a position to provide an account for English LBEs, to which we now turn.

Accounting for LBEs
As foreshadowed above, we propose that a general strong/weak competition principle derives both the distribution of descriptively plain herself vs. 'erself and the distribution of descriptively exempt (i.e. logophoric) herself. Only assuming competition between herself and 'erself, as implied by Ahn's (2015) Rule J, remains insufficient under Cardinaletti and Starke (1999). They do not explicitly take into account interpretation, but their proposal implies that competition should only arise if the weak form can induce the same interpretation as the strong form. The unavailability of logophoric herself in (80) thus requires enrichment of the set of competitors involved.
The data in (86) revealed that, in French, competition applies not only between strong and weak reflexive forms, but between strong and weak pronominals generally. We propose to extend this line of analysis to English: the strong reflexive himself in (67b) 47 (repeated below as (88b)) is not blocked by the weak reflexive, which yields a different interpretation (see (88c) repeating (80b)), but by the weak pronoun 'im as shown in (88a) (cf. French (86a)). Just as non-reflexive clitics appear in the same environments as reflexive clitics in French, weak pronouns in English have indeed been shown to be confined to the environments that host weak reflexives (cf. Zwicky 1986; Wallenberg 2007; i.a.).

(88)
What happened at the party? a. Max i boasted that the queen invited 'im i for a drink. b. *Max i boasted that the queen invited himself i for a drink. c. Max i boasted that [the queen] k invited 'erself k for a drink.
However, the strong reflexive himself is acceptable in (67a) (repeated as (89a)) (89), insofar as the phonologically reduced reflexive 'emzelf (though not the weak pronoun 'em) is permitted in the equivalent construction. However, a proper comparison between Dutch and English would require a very careful investigation of the relevant properties of Dutch; in particular, note that languages like Dutch that exhibit more types of pronouns and anaphors than English (i.e. not only hem(zelf ) and 'm(zelf ), but also zich(zelf )) raise further questions regarding strong/weak competition that go beyond the scope of this article (see e.g. Reuland 2011;Rooryck and Vanden Wyngaerd 2011;Broekhuis 2022 for specific discussion about Dutch anaphors). Thus, the addition of the weak pronoun, alongside the weak reflexive, as a competitor to the strong reflexive, solves the LBE issue in the verbal domain. Before coming back to the nominal domain and PNAs, some clarifications are in order. First, it is worth noting that the reflexive morpheme (i.e. French -même, English -self ) does not seem to be taken into account when evaluating the weight of competing elements. Strong reflexives (French elle-même; English herself ) behave like strong pronouns (French elle; English her) with respect to weak/strong competition: as shown in Table 4, strong forms do not block each other, and each is blocked by clitics or weak pronouns (see Charnavel and Sportiche 2016: 55-56). This is unsurprising under the movement-based account we adopt, which yields a distinction between forms that move (clitics and weak pronominals) and forms that do not move (the other pronominal forms). We will nevertheless not fully derive this hypothesis here, which is not directly relevant to our purposes: through morphological distinctions, French unambiguously shows that elle-même and elle behave similarly with respect to competition; it is sufficient to extend this empirical generalization to English.
Second, it is also worth noting that while adopting the same parsimony-based approach as in the previous section, we have done so in reverse. In Sect. 2, we started from the widespread observation that plain and exempt anaphors exhibit some systematic differences in their distribution, which have led some to postulate homophony; we then unified them by hypothesizing that the differences in their behavior do not come from their lexical entries, but derive from the nature of their binder (esp. overt binders vs. implicit pro log ). In this section, we conversely started with the null hypothesis that English plain reflexives form a homogeneous class; based on Ahn's (2015) discovery, we then showed that they actually exhibit systematic differences in their prosody. Just as before, we do not want to imply that herself and 'erself are homophonous: the particular behavior of 'erself comes from movement triggered by the presence of another implicit element (Ahn's reflexive Voice), which, due to its properties, can only occur in some syntactic configurations in which it can yield prosodic effects. Thus, we assume that neither plain herself and exempt herself, nor weak 'erself and strong herself, have different lexical entries. This does not mean though that other languages cannot display different lexical entries for marking similar differences. Morphological distinction between plain and exempt anaphors seems to be documented in some languages (e.g. zichzelf vs. hemzelf in Dutch, see Rooryck and Vanden Wyngaerd 2011), which implies that some anaphors may be lexically specified with respect to (non)logophoricity (see Charnavel 2020aCharnavel , 2020b. We take the morphological distinction between French se and French elle-même to suggest a similar pattern: association with the reflexive Voice is lexically marked in French se, unlike in English 'erself. Finally, as should have been clear, we use the weak/strong terminology (and the 'er(self )/her(self ) notation) in the same way with pronouns and with reflexives. The prosodic difference between the pronouns her and 'er, just like that between herself and 'erself, only surfaces when they are in a syntactic position requiring phrasal stress in neutral contexts. Otherwise, strong her, like herself, can surface as weak (e.g. when it is given); conversely, weak 'er, like 'erself, can surface as strong (e.g. when it is focused). 49 This hypothesis raises the question of the source of the difference between her and 'er. It cannot rely on movement to a specific Voice head as Ahn (2015) argues for reflexives, but we similarly hypothesize that the difference is structural and can be specified in light of French. In French, even if only se is associated with a reflexive voice, accusative la and dative lui also qualify as clitics. We likewise assume (cf. Wallenberg 2007) that weak 'er, unlike strong her but like 'erself, undergoes some clitic-like movement corresponding to the kind of movement that Cardinaletti and Starke (1999) hypothesize for weak pronouns; they indeed propose a three-way distinction between strong pronouns, weak pronouns and clitics, which correlates with a three-way distinction between various movement spans. 50 Fully motivating this hypothesis is beyond the scope of this article, but we will provide some additional support for it in Sect. 3.2.1.
In sum, LBEs arise from the fact that logophorically bound herself cannot occur in positions that can host weak pronominal elements-just like French logophorically bound elle-même cannot occupy cliticizable positions (cf. Italian se in Napoli 1979). And this fact derives from the hypothesis-fully independent from Condition A-that due to a general principle of competition between weaker and stronger forms, herself is systematically excluded from configurations in which a weaker form ('erself or 'er) is acceptable and can yield the same interpretation.
Due to Condition B ruling out pronouns in some local configurations, plain herself (cf. French elle-même) can sometimes appear in positions that can in principle host weak elements, when none of the weak elements can give rise to the relevant interpretation; this is for example the case when herself is an indirect object bound by the object, as shown in (76c) repeated as (90a).
(90) a. Liz i assigned Danny k to himself k . b. *Liz i assigned Dánny k to 'imself k . c. *Liz i assigned Dánny k to 'im k .
Logophorically bound herself, however, can never appear in such positions, as shown in (91), because her and 'er cover the full range of non-local interpretations that exclude descriptively plain herself : an anaphor whose only potential local binder is 49 Cardinaletti and Starke (1999: 49-50) similarly mention that deficient pronouns can bear contrastive stress as long as they refer to an entity that is already prominent in the discourse. 50 This three-way distinction is relevant to French itself, in which deficient subject pronouns seem to be weak pronouns rather than clitics (see Cardinaletti and Starke 1999).
pro log can always alternate with a pronoun. 51 When the alternating pronoun is weak (as in (91c)), the logophorically bound anaphor is excluded. Therefore, logophorically bound anaphors are acceptable only in positions excluding weak elements.
(91) a. *Liz i said that they assigned Danny k to herself i . b. *Liz i said that they assigned Danny k to 'erself i . c. Liz i said that they assigned Danny k to 'er i .

The logophoric blocking effect in the nominal domain
Recall that we have made a detour and delved into LBEs in the verbal domain in order to ultimately return to our main goal consisting in explaining the behavior of PNAs. In this section, we will examine the consequences of the weak/strong competition in the nominal domain, which will allow us to provide a full picture of the PNA behavior. By further specifying the set of positions that exclude logophorically bound herself, we will break the similarity between French and English and (descriptively) reintegrate the notion of coargumenthood into the picture.

Refining the reflexive projection associated with 'erself
In the previous section, we concluded that logophorically bound anaphors are blocked by clitic-like elements under the same interpretation. Using again French as a clue for understanding English, it seems that such blocking effects should be irrelevant in the nominal domain. Reflexive, accusative and dative clitics are indeed all banned from the nominal domain as they need to move to a position in the tense field. Assuming that weak pronominal elements in English are similar to French clitics, as suggested by the facts so far, it should follow that neither 'erself, nor 'er should be licensed within DPs, thus never blocking logophorically bound herself. As shown in Ahn (2015: 129-132), this is incorrect: even when they are the most deeply embedded element in a broad focus context, DP-internal anaphors need not bear phrasal stress (cf. Helke 1970: 114, 126-128). This fact alone is not sufficient to question the comparison between French and English though, as shown by (92) In (92), 'erself is acceptable within the picture noun phrase, unlike the French reflexive clitic in (93b). But this does not necessarily imply that 'erself moves to a DP-internal reflexive voice. Given that 'erself does not sit in an island, as shown by (94) below, movement to the verbal reflexive voice is possible and it is predicted to induce weak prosody on 'erself. In other words, the clitic counterpart to a DP-internal weak reflexive should not necessarily appear within the DP, but could also be a clitic appearing in its standard position. The reason why this option is also unavailable in French, as shown by (93c), is due to an independent difference between French and English pertaining to constraints on extraction illustrated in (94) (96a) cannot involve movement to the main clause VoiceP since the binder of the reflexive is the object, not the subject; a derivation including the verbal reflexive Voice is also excluded in (96b) since it is a passive construction. 52 Although he leaves for further research a more detailed investigation of DP-internal reflexives, Ahn (2015: 131) concludes from these facts that the binder of 'imself must be some DP-internal local subject here, and that a reflexive Voice can therefore be present within DPs. This hypothesis is further supported, Ahn suggests, by the interpretation of DPinternal weak reflexives: for instance, (97) entails Jack writing the letter. To account for the distributional differences between English 'erself and French se, Ahn (2015) in effect proposes that the reflexive Voice has a broader distribution in English than in French. According to him, facts such as (96)-(97) above entail that the reflexive Voice can occur within DPs; facts in (98)-(99) below further imply that it can appear within small clauses, even non-verbal ones unlike in French (see (100)- (101)).

brûler. burn
In both (98)-(99), the English reflexive, which is bound by the small clause subject, is weak. However, French se is only available within verbal small clauses such as (101). Other types of small clauses, such as the adjectival small clause in (100), can only host strong reflexives. The parallel between French se and English 'erself thus seems to break down here. Instead of assuming an idiosyncratic distribution for the English reflexive Voice, we take these facts to reveal that the head attracting 'erself can't literally be a Voice head, which associates only with verbs; it must be a reflexive head compatible with adjectives and nouns. Such a head is more transparently relevant in another English construction. Reflexivity within nominals or within small clauses does not always require the weak reflexive 'erself as in (96)-(98) above, but in some lexically restricted cases, it can be expressed with a different kind of element, namely self-, as illustrated in (102)-(103). We thus hypothesize that the head attracting 'erself is similar to that associated with self -and we will henceforth call it SELF. 53 Movement to SELF within nominals is illustrated in (104) representing (97). 53 Our account may also be compatible with the hypothesis that self itself moves (see Charnavel and Sportiche to appear). We here prefer to adopt the hypothesis that 'erself moves to an abstract head SELF to retain the similarity with Ahn's account, which can directly explain the prosodic facts. Further note that self -movement has also been proposed by e.g. Reinhart and Reuland (1991) and Reuland (2011) in order to derive Condition A. This is crucially different from Ahn's and our proposals, according to which only a subset of plain reflexives (i.e. weak reflexives) undergo movement. Movement to SELF is thus not meant to derive Condition A; assuming so would incorrectly predict that strong reflexives are always logophoric, but we showed that inanimate reflexives can appear in strong configurations (see e.g. (18) or (20)).

(102)
What happened during the writer's meeting yesterday? Jenna made Patrice self-critical.
(103) What happened in art class yesterday? Lucie showed Pete his self-portrait.
Our hypothesis is that 'erself must move to an abstract head SELF, which like self-, can be present in verbal, nominal and adjectival domains, and unlike self-, is not restricted lexically. In other words, our head SELF remains similar to Ahn's (2015) reflexive Voice in many respects, but it is crucially not a Voice head, which accounts for its broader distribution than French se outside the verbal domain. At the same time, we argue that by replacing REFL Voice with SELF, we do not lose the explanation for the restricted distribution of 'erself in the verbal domain, namely its deep subject orientation as well as its exclusion from islands (see (73)-(76)). To understand why, comparison with French is again revealing. French also has a head that, like self -, can express reflexivity within nominals or within adjectives in some lexically restricted cases, namely auto-(see Labelle 2008; Sportiche 2014; i.a.) as illustrated in (105) Therefore, all distributional properties observed with se directly apply to auto-, as illustrated in (108) We do not here aim at explaining such obligatory association of auto-with se, which is beyond the scope of this article (for some discussion, see Labelle 2008; Sportiche 2014; i.a.). But we take it as a clue explaining the distribution of English 'erself : 'erself exhibits a distribution that is similar to se in the verbal domain, but broader otherwise, because it must move to a head, SELF, that is not a Voice head itself, but directly interacts with Voice in the verbal domain. We leave the details for future research, but will henceforth assume, as summarized below, that SELF is similar to (se)auto in entailing local deep subject orientation and not being restricted to the verbal domain. This hypothesis has a further welcome consequence. Recall that under our hypothesis, herself competes not only with 'erself, but also with 'er, so that understanding the distribution of weak pronouns is crucial to predict the distribution of logophorically bound herself. As a first pass, we compared 'er with French accusative and dative clitics la/lui, whose distribution is similar to that of se in the sense that they all originate from non-nominative structurally case-marked positions. 54 As suggested by our examples so far, this generalization applies to 'er in the verbal domain. But crucially, the distribution of 'er in small clauses and DPs raises the same issue as 'erself as shown in (112) Kayne (2000) strengthens this generalization by claiming that in French, pronominal arguments that are structurally case-marked must be doubled by a clitic, whether they are silent (as in examples of the text) or not (see fn. 43). He further suggests that this approach could replace Cardinaletti and Starke's (1999) economy-based approach, which requires comparing derivations. To explain why pronominal dative elements do not always require a clitic, whether they are stressed or not (see (84)), Kayne hypothesizes that the configurations without clitics are instances of topicalization, which independently applies more readily to datives than to accusatives. The issue of this hypothesis is to explain why in the absence of contrastive focus, such topicalization is only available (or at least much better) when there is no cliticdoubled version yielding the same interpretation (see (84d) vs. (84b)). It thus seems that assuming some kind of competition is after all necessary to capture all the facts. While la/lui are excluded from non-verbal small clauses (e.g. (114)) and from DPs (e.g. (115)), 'er is available in both (e.g. (112)-(113)). The distribution of 'er is thus similar to that of 'erself in displaying the same kind of distribution as clitics in the verbal domains, but in also appearing outside the verbal domain. Of course, the solution cannot rely on SELF itself, which is only associated with reflexive interpretations, just like the distribution of la/lui does not rely on the reflexive voice associated with se. But we assume that 'er belongs to the same paradigm as 'erself just like la/lui belong to the same paradigm as se, and each paradigm is associated with a certain type of movement: clitic movement for se/la/lui, weak pronoun movement for 'er/'erself. Recall indeed from Sect. 3.1.2.2 that Cardinaletti and Starke (1999) assumes a three-way distinction between strong pronouns, weak pronouns and clitics, and propose that weak pronouns undergo a shorter movement than clitics. This hypothesis is fully compatible with the English facts reviewed so far (cf. Zwicky 1986;Wallenberg 2007), and allows us to understand both the similarities and the differences between French la/lui/se and English 'er/'erself. In sum, we hypothesize that 'er and 'erself are weak forms rather than clitics, and thus undergo shorter movements, which explains why they can occur not only in the verbal domain, but also in nominal and adjectival domains. 55 But in all domains, we observe that the distribution of 'er/'erself is similar to that of la/lui/se in being restricted to non-nominative structurally case-marked positions (where, for our descriptive purposes, we take genitive positions to be structurally case-marked).
(116) Weak vs. strong pronouns in English: a. English pronouns divide into strong forms-her and herself -and weak forms-'er and 'erself. b. Unlike strong forms, weak forms undergo short movements, which restricts them to non-nominative structurally case-marked positions, by hypothesis.
We are thus now in a position to partly reintegrate the notion of coargumenthood into the picture. Recall from Sect. 2 that defining Condition A and exemption from it based on the notion of coargumenthood, as PBTs do, is not tenable. But the discussion above reveals that at least descriptively, the notion of coargumenthood (with a subject, in particular) is partly relevant to understanding contrasts like (67a-67b), which motivated PBTs. Logophorically bound herself is blocked by 'er/'erself, which can only appear in non-nominative structurally case-marked positions. It follows that logophorically bound herself is excluded from positions with syntactic coargumental subjects, since positions with a syntactic coargumental subject are necessarily nonnominative structurally case-marked positions. 56 The relevance of coargumenthood to reflexives is thus twofold: on the one hand, plain 'erself must be bound by a syntactic coargumental subject; on the other hand, exempt herself is ruled out in the presence of a syntactic coargumental subject. Crucially though, the relevance of coargumenthood to the distribution of reflexives is fully independent of Condition A and exemption from it, unlike what PBTs argue. First, the coargumental subject orientation of 'erself certainly entails compliance with Condition A, but it is distinct from Condition A, since plain herself (unlike 'erself ) is not subject to this requirement, but only to Condition A. Second, the blocking of exempt herself by the presence of a coargumental subject derives from competition with weak forms, which falls under a general principle of competition fully independent of Condition A. 57 These observations confirm that the notion of coargumenthood as such is irrelevant to the distribution of reflexives, even if it is descriptively useful in stating the generalizations pertaining to the distribution of 'erself and logophorically bound herself summarized in (117).

(117)
The descriptive relevance of coargumenthood for English reflexives: a. LSORs: 'erself must be bound by a syntactic coargument subject. b. LBEs: logophorically bound herself is ruled out in the presence of a syntactic coargument subject. 56 Non-nominative structurally-marked positions, however, do not necessarily have a coargumental subject: this is not the case with verbs like seem or bother that lack a subject. We correctly predict that logophorically bound herself is excluded from complements of such verbs as shown in (i). But we will not further delve into such cases as nominal counterparts of such verbs (e.g. possibility in fn. 63) do not descriptively qualify as picture noun phrases.
(i) *He i thinks it bothered himself i that S. (Chomsky 1981: 214) 57 French shows even more transparently that the notion of coargumenthood is only descriptively and only partly relevant to the distribution of reflexives (see details in Charnavel and Sportiche 2016): the binder of se is always a syntactic coargument, but it is neither the case that all syntactic coarguments qualify as binders of se (see (i) below), nor that all positions with syntactic coargumental subjects exclude logophorically bound elle-même (see (ii) below).  (Charnavel and Sportiche 2016: 52) The facts are different in English: the reflexive is weak in the English counterpart of (i) and logophorically bound herself is excluded in the counterpart of (ii). This difference correlates with different constraints on extraction in French and in English (cf. (94)-(95)).

Consequences for possessed PNAs (first pass)
Now that we have established a precise generalization about LBEs and showed that they derive from a competition principle fully independent of Condition A, we can explore the consequences for the distribution of PNAs. In Sect. 2, we have shown that PNAs, just like any other anaphor, can be bound by any local binder, including the possessor (if present) and pro log (under the appropriate discourse conditions) as exemplified in (118) In our view, the controversy about the status of the possessor results from a confusion. As stated in (122), the so-called possessor, which we will henceforth call genitive to avoid any further confusion, can correspond to various underlying positions: the subject of NP or any other source (object of NP, possessor, etc.), which end up in the same surface position in English (see Stowell 1989;Giorgi and Longobardi 1991;Longobardi 2001;i.a.).

(122)
Ambiguity of the genitive in English: In English, a genitive DP realizes the subject of NP or other types of (quasi)arguments (complements, possessor).
Only the subject of NP qualifies as a coargumental subject for PNAs. The relevant question is thus to determine whether the genitive counts as the subject of NP or not. The interpretation of the genitive, we hypothesize, can provide a crucial clue: if it specifically depends on the denotation of the noun (just like the interpretation of a verbal subject depends on the interpretation of the verb), it can be a subject of NP; it cannot if it stands in some other relation (e.g. possession) to the noun.
For nominalizations (e.g. examination), this hypothesis (inspired by Grimshaw 1990 58 ) predicts that when the nominalization denotes a process (complex event nominal), the genitive is construed as the specifically relevant actor of the process (e.g. the examiner) and thus qualifies as an argument, i.e. the subject, of the noun; but when the nominalization denotes a result (result nominal), the genitive is compatible with several modifier readings (e.g. the possessor, author or taker of the exam) and thus does not necessarily qualify as the subject of the noun. This hypothesis accounts for LBEs in examples like (31b)/(64) or (33a)/(65) repeated below.
(123) *The fact that Mary k 's description of himself i was flawless was believed to be disturbing John i .
(124) *Jill i found Matt k 's fear of herself i surprising.
In (123), Mary is intended to be interpreted as the agent of the act of describing John. Under our hypothesis, this implies that Mary originates as the subject of NP, thus blocking the logophoric binding of himself. 59 Similarly, Matt in (124) is construed as the experiencer of the feeling of fear and thus counts as the subject of NP, which gives rise to a LBE for herself. Conversely, PNAs in result nominals are predicted to be licensed by logophoric binding as confirmed by (125), which contrasts with (123).

(125)
John confessed that the media's descriptions of himself are always disturbing to him.
In (125), the noun description is pluralized, which according to Grimshaw, is incompatible with the process reading. This implies that the NP need not have a subject here, so that logophoric binding is possible. As its antecedent John is appropriately construed as the logophoric center in (125) (vs. (123); see fn. 59), himself is thus correctly predicted to be licensed. Returning now to picture NPs in the narrow sense (recall fn. 3), we hypothesize in (126) that the genitive can qualify as the subject of a picture NP when it is interpreted as the creator (cf. Chomsky 1986;Asudeh and Keller 2001;Davies and Dubinsky 2003;Jaeger 2004;Ahn 2015;i.a.). By creator, we mean the agent responsible for the informational content (of the entity) denoted by the noun, such as a photographer or painter (in the case of picture or portrait) or an author, writer or teller (in the case of book or story), for example. We further hypothesize (pace Davies and Dubinsky 2003;Runner 2007) that this holds whether the picture noun is interpreted as concrete or result (see discussion in fn. 65). Like Runner (2007) (inspired by Davies and Dubinsky 2003; see fn. 21), we thus distinguish between two possible types of argument structure for picture nouns (i.e. picture nouns with or without subjects), but unlike 58 Unlike Grimshaw (1990), we assume that result nominals can have a subject, thus divorcing differences in argument structure from the complex event/result distinction (cf. Runner 2007;Reuland 2011;see fn. 21). 59 (123) is degraded also under a result reading because two factors disfavor the logophoric construal of John: the passive was believed to and the possessor Mary introduce two potentially intervening logophoric centers. By contrast, the logophoric construal of John is favored in (125) by the use of the attitude verb confessed and the non-specificity of the possessor the media. Runner (2007), we do not take this difference to correlate with the concrete/result distinction (see further details in the next section).

(126)
Subject of picture nouns: The subject of a picture noun phrase (when there is one) must be interpreted as the creator of informational content of the entity denoted by the noun.
This hypothesis is not sufficient to account for the contrast in (61)  The creator interpretation of Harry is possible only in (127a) given that the creator must be Joe in (127b) due to the presence of the creation verb. Under our current hypothesis, we would thus expect the contrast to go in the other direction: the logophoric interpretation of himself as Joe could be blocked by the subject interpretation of Harry only in (127a), not in (127b). As we will see in Sect. 3.2.4, the actual contrast is due to several additional factors interacting with our hypothesis: the possible implicitness of subjects of NP, the obligatoriness vs. optionality of subjects of NP depending on the environment, and the fact that in English, a subject of NP and a possessor cannot be realized simultaneously if they are disjoint. To explain these factors, we first need to examine the consequences of the generalizations in (117) for possessorless PNAs.

Consequences for possessorless PNAs
At the beginning of Sect. 3, we mentioned that we would start with LBEs in the verbal domain because the nominal domain presents additional complexities. One is the ambiguity of the genitive (as subject of NP or not) discussed in the previous section. A second one is the availability of silent subjects in NPs, which has no counterpart in the finite verbal domain in English. Recall that we have so far circumvented the issue (which we briefly discussed in connection with Chomsky's (1986) PRO-based hypothesis at the beginning of Sect. 2.1.1.1) by avoiding agentive interpretations of possessorless PNAs, i.e. interpretations under which they refer to the creator of the entity denoted by the picture noun. We are now in a position to tackle the issue. Under the hypothesis that the subject of NP can be covert-which we will henceforth call pro subj 60 -generalizations (117a-117b) entail that in the presence of pro subj , pro subj must bind the PNA, thus blocking any other binding, in particular by pro log . 61 In other words, the presence of pro subj forces the PNA to be weak 'erself re- 60 We remain agnostic about the precise identity of pro subj , i.e. whether it should be treated as PRO or as pro (see e.g. Sichel 2009 andreview in Landau 2013: 208-213). Nothing hinges on this issue in our argumentation. 61 Recall that under our hypothesis, such blocking is due to the fact that logophorically bound herself cannot occur in positions that can host weak elements because it competes with 'erself and 'er under identical interpretations. Here, herself is blocked by 'er as 'erself would yield a different interpretation. This relies ferring to the creator of the informational content of the entity denoted by the picture noun; a possessorless PNA can only be strong herself in the absence of pro subj .
Thus, generalization (117a) first predicts that a possessorless PNA that is descriptively exempt can be acceptable even if it is not logophoric, as long as it refers to the creator of its picture noun. This prediction is borne out in example (128) (112)) that object pronouns must be weak (i.e. undergo weak pronoun movement) when there is a coargumental subject. 62 This implies that apparent exemption is not always due to logophoricity as generally assumed (see references mentioned in Sect. 2). Recall that this assumption forms the basis of Charnavel and Sportiche's (2016) inanimacy-based tool, used in Sect. 2. But we avoided the issue by excluding non-agentive interpretations in that section. Further note that fortunately, the claims made in Charnavel and Sportiche (2016) and earlier work are not affected either as on closer inspection of their examples, it turns out that agentive interpretations are usually not intended. 63 Most previous arguments of the literature were made on the basis of Condition C (see (i)) and control (see (ii)).
(i) a. The PRO *i/k knowledge that John i might fail bothered him. b.
The possibility that John might fail bothered him. (Chomsky 1986: 167;cf. Ross 1969: 195;Williams 1985: 298) (ii) a. the PRO destruction of the city PRO to prove a point b. *the city's destruction to prove a point (Roeper 1987: 280;cf. Chomsky 1986: 123) In (ia), the impossibility of attributing the knowledge to John arguably results from a Condition C effect due to the presence of an implicit subject of knowledge; no such effect arises in (ib) given that possibility (like be possible) does not license a subject. In (iia), possible control into the adjunct clause suggests the presence of an implicit subject of destruction; the ungrammaticality of (iib), which involves nominal passivization, further suggests that this subject is syntactically represented: under the assumption that there is only one genitive position in English (see (122)), possessivization of the object is incompatible with the presence of a subject (see further discussion in Giorgi and Longobardi 1991;Landau 2013;i.a.). Other arguments involve secondary predicates Landau 2013;i.a.) or agreement facts (Landau 2013;i.a.). Of course, Condition A has also been used as an argument for the presence of an implicit subject in nominals (see Stowell 1989;Landau 2013;i.a.), but the argument is usually confounded by the lack Under this hypothesis, generalization (117b) furthermore predicts LBEs within nominals. But the details of the prediction are complicated by another difficulty specific to the nominal domain: it is often assumed that subjects of NP do not systematically project syntactically. For example, Chomsky's (1986) argument about (3)-(4) implies that pro subj projects only optionally in expressions like hear stories about (i.e. hear (pro subj ) stories about). Under our hypothesis, LBEs only arise for possessorless PNAs in configurations where pro subj obligatorily projects; logophoric binding should still be possible if pro subj is only optionally present. The distribution of LBEs thus has the potential to clarify the conditions under which pro subj is present.
In fact, the experimental findings presented in Bryant and Charnavel (2020) reveal two configurations forcing the projection of pro subj : nouns with goal arguments (e.g. letter to) and complements of creation verbs (e.g. write a book about) as illustrated in (130)-(131) Bryant and Charnavel 2020: 12) In (130), herself is only acceptable if the referent of its antecedent, Lea, is interpreted as the creator, namely if Lea wrote the note; whether she is interpreted as the logophoric center or not is irrelevant (both contexts are compatible with this interpretation but do not force it). This follows from generalization (117b) if pro subj obligatorily projects when the picture noun takes a goal argument as shown in (133a-133b) representing (130a-130b) respectively. 65 of control for logophoricity or competition, as should be clear from our argumentation in the main text. Finally, note that the various arguments about the presence of an implicit subject in nominals are often made without controlling for the type of nominals (e.g. complex event vs. result nominals), which as we saw complicates matters. 64 Contrasts in Bryant and Charnavel (2020) reflect statistically significant differences in grammaticality judgments made by 108 native English speakers. We here slightly adjust some of their examples or contexts to make their pairs more minimal without affecting the relevant factors. 65 The contrast holds even though note is construed as a concrete noun here (denoting the physical object). This observation argues against Runner's (2007) and Reuland's (2011) interpretation of Davies and Du-(133) a. The water Lea i splashed smeared [the (pro log-i/k ) pro subj-i note to 'erself i ]. b. *The water Lea i splashed smeared [the (pro log-i ) pro subj-k note to herself i ].
By contrast, herself is acceptable in (132a-132b) under a logophoric reading whether or not the antecedent is construed as the creator, as shown in (134a-134b) (representing (132a-132b) respectively). This follows from generalization (117b) if pro subj only optionally projects when the picture noun takes a theme argument as represented.
(134) a. The water Lea i splashed smeared [the (pro log-i/k ) (pro subj-i ) picture of (h)erself i ]. b. The water Lea i splashed smeared [the pro log-i (*pro subj-k ) picture of herself i ].
The hypothesis of a difference between nouns with theme arguments and those with goal arguments with respect to pro subj projection is further supported by the contrast between (135) and (136) 66 Furthermore, this hypothesis correctly predicts that apparent exemption of inanimates is always possible when they express the goal of a noun if their antecedent can be construed as the subject of that noun. This is illustrated in example (i) adapted from google hits (cf. (128)).
(i) Examine [the requested page] i to be sure the link to itself i is displayed.
of pro subj turns the picture noun phrase into the binding domain, Ellis is in fact not a possible local binder for himself ; only pro subj and pro log are. But due to generalization (117b), logophoric binding of himself is blocked. Thus, himself is only acceptable if it refers to the letter writer.
(137) a. Ellis i enjoyed [the (pro log-i/k ) pro subj-i letter to 'imself i ]. b. *Ellis i enjoyed [the (pro log-i/k ) pro subj-k letter to himself i ].
However, himself in (136) is acceptable whether or not its antecedent is interpreted as the creator as expected under representations (138a-138b).
(138) a. Ellis i liked [the (pro log-i/k ) (pro subj-i ) picture of (h)imself i ]. b. Ellis i liked [the (pro log-i/k ) (*pro subj-k ) picture of himself i ].
Example (131a-131b) above (represented in (139a-139b) below) further reveals that optionality vs. obligatoriness of pro subj projection does not only depend on the type of complement taken by the noun, but also on the broader syntactic context. Under our hypothesis, the contrast between (131a) and (131b)  The reason why the PNA here must refer to Lea's brother is that the picture noun phrase obligatorily involves pro subj anteceded by the subject of the creation verb painted. Given generalizations (117a-117b), the PNA must therefore be bound by pro subj ('imself in (139b)); it cannot be logophorically bound (herself in (139a)). 67 In sum, the results of our examination of LBEs in the verbal domain have allowed us to refine the empirical generalizations that we made about possessorless PNAs in Sect. 2: we have added one further possible local binder for PNAs, namely pro subj , and we have restricted the availability of logophoric binding (or any binding by a binder different from pro subj ) to configurations lacking pro subj . We have thereby shed further light on the conditions under which subjects of NPs syntactically project in English as summarized in (140) 67 This also explains the contrast in (i) below discussed in Chomsky (1995: 206) and Runner (2002): under the idiomatic reading of take a picture in (b), himself can only refer to Bill. Under our hypothesis, this follows from the obligatory presence of pro subj under that reading (where take is a creation verb), which is thus the only possible binder of himself given generalizations (117a-117b). b. Subjects of picture NPs must be syntactically represented in English when the NP is the complement of a creation verb (e.g. write a book about). c. Otherwise, the syntactic projection of subjects of picture NPs is not obligatory.
Exploring the referential constraints on PNAs thus provides a new probe into the argument structure of NPs. Space limits do not allow us to further use this probe here, but we hope that it will be done in future research. 68 We also have to leave for further investigation the analysis of the generalizations established in (140)-(141). But to close the examination of possessorless PNAs, we would like to show how to further test these generalizations based on Conditions B and C by illustrating one case: experimental evidence from Conditions B and C independently supports the generalizations regarding nouns taking a goal argument ( (140) and (141a)). 69 Under any definition of Condition B (i.e. under any version of Chomskian or predicate-based theories), the subject of NP is sufficiently local to its coarguments to potentially trigger Condition B effects. Generalization (141a) thus predicts that, if the goal of the noun is expressed by a pronoun, Condition B effects will occur if the pronoun also refers to the creator and, hence, corefers with the subject of NP. The contrast between (142a) and (142b) shows that this prediction is borne out: while the pronoun him can corefer with the matrix subject Jack in a context in which Jack did not write the letter (142b), coreference is not possible when Jack did write the letter and, thus, is syntactically represented as subject of NP (142a).
(142) a. Context: When Jack was young, he wrote a letter to his future self. His mom, Faye, kept the letter in her scrapbook of family mementos. Over the holidays, Jack looked through the scrapbook. * Jack i tore up the letter to him i . 68 Among other issues, it would be interesting to use this probe to investigate other types of nouns, nouns with multiple objects, or nominal passives. Examination of the interaction between prosody and interpretation in DP-internal reflexives is also an important avenue for future research: in light of Ahn (2015), we would predict only prosodically weak reflexives and pronouns to be available in NPs containing a subject in maximally broad focus contexts. 69 The contrasts in (142) and (143) reflect statistically significant differences in grammaticality judgments obtained in a survey involving 61 native English speakers recruited through Prolific. In the presentation and analysis of this survey, we used the same methodology as Bryant and Charnavel (2020), except that we added follow-up comprehension questions after each example in order to make sure that participants took into account the preceding context. Specifically, the participants were presented with context-sentence pairs as in (142) and (143), where target sentences always contained a picture NP with goal argument, and were asked to rate the naturalness of the target sentences on a continuous sliding scale from 1 (very unnatural) to 7 (very natural). Ratings for indefinite and definite PNPs with pronoun and proper name goals were analyzed in R using a linear mixed effects model (lme4; Bates et al. 2015), revealing a significant effect of antecedent role (p < 0.05): sentences for which the pronoun or proper name referred to the author were rated significantly lower than sentences for which the pronoun or proper name did not refer to the author. For example, sentences like (142a) received a mean rating of 2.4/7, whereas sentences like (142b) received a mean rating of 4.5/7. Similarly, sentences like (143a) received a mean rating of 3.3/7, whereas sentences like (143b) received a mean rating of 5.5/7. b. Context: When Faye was young, she wrote a letter to her brother Jack. Their mom kept the letter in her scrapbook of family mementos. Over the holidays, Jack looked through the scrapbook. Jack i tore up the letter to him i .
Similarly, if the goal of the noun is expressed by a proper name, Condition C effects are predicted to arise if the name refers to the creator (e.g. letter writer) and, hence, corefers with the subject of NP. As shown by the contrast between (143a) and (143b), this prediction is likewise borne out.
(143) a. Context: When Jack was young, he wrote a letter to his future self. His mom, Faye, kept the letter in her scrapbook of family mementos. Over the holidays, Faye looked through the scrapbook. *Faye tore up the letter to Jack. b. Context: When Faye was young, she wrote a letter to her brother Jack.
Their mom, kept the letter in her scrapbook of family mementos. Over the holidays, Faye looked through the scrapbook. Faye tore up the letter to Jack.
To wrap up, possessorless PNAs are licensed in the absence of an overt local binder in two cases: when they are bound by pro subj (and are thus interpreted as the creator) or when they are bound by pro log (and are thus interpreted as the logophoric center); these two cases exclude each other as the presence of pro subj forces the PNA to be bound by it.

Returning to possessed PNAs
All pieces are now in place to solve the remaining LBE issues in the nominal domain. Recall from Sect. 3.2.2. that the hypothesis that the subject of picture NPs must be construed as the creator is not sufficient to explain contrasts like (144a) vs. b (repeating (127a-127b)).
(144) a. /?Joe i destroyed Harry k 's book about himself i . b. ?/*Joe i wrote Harry k 's book about himself i .
But the conclusion we reached in (141b) about the obligatory projection of the subject in nouns complements of creation verbs provides the missing piece to the solution.
Given the presence of the creation verb wrote in (144b), (141b) entails that book must have a subject denoting the creator, namely Joe (the subject of wrote) as represented in (145). Given that the genitive position is occupied by Harry, this gives rise to a conflict due to the availability of only one genitive position in English (see (122); cf. Stowell 1989;Giorgi and Longobardi 1991;Longobardi 2001;i.a.). Unlike what Runner (2007) and Reuland (2011) argue (see Sect. 2.1.2.1), the ungrammaticality of (145) is thus not due to the restriction of exemption to concrete nouns (cf. fn. 65), but to the conflict between the creator and the possessor. In fact, this constraint also explains the ungrammaticality of examples like (62a) (repeated as (146)), which do not involve any reflexive.
(146) *John i took Mary's pictures of him i . (Williams 1987: 156) This hypothesis is further supported by the fact that himself in (145) cannot be bound by the genitive either, as shown by the similar example in (147). 70 (147) *Gordon i wrote Faye k 's pro subj-i book about herself k . (Bryant and Charnavel 2020: 12) By contrast, himself can be logophorically bound in (144a) (represented in (148)) because in the absence of a creation verb, the noun book need not have a subject. Consequently, himself can also be bound by Harry, whether it is interpreted as the creator or not. 71 (148) Joe i destroyed [pro log-i Harry k 's book about himself i/k ].
The same reasoning can be applied to derive the unacceptability of examples like (66) (specified below in (149a-149b)) involving goal arguments.
(149) a. Context: Chelsey gave Brandon a letter from her journal that she had written when they started dating. *Chelsey i found Brandon k 's letter to herself i . b. Context: As the first leg of a scavenger hunt designed for his daughter, Chelsey, Brandon hid a letter of instructions he wrote in the back of the pantry. *Chelsey i found Brandon k 's letter to herself i . 70 Nevertheless, we further observe in Bryant and Charnavel (2020: 13) that in the absence of a creation verb, logophoric binding is easier in concrete nouns, e.g. (ib) (cf. (148)), than in result nouns, e.g. (ia).
(i) Context: For a school assignment, Olivia took a series of photos depicting her everyday life. Afterward, she gave one of the photos to her boyfriend, Patrick.
a. ?Olivia is proud of Patrick's photo of herself. b. Olivia shredded Patrick's photo of herself.
Given the lack of contrast in both (132) and the example discussed in fn. 65, the contrast in (i) above cannot be due to the presence of pro subj in (a), which we saw is optional in the absence of creation verb or goal argument. In fact (ia), although degraded, remains crucially better than (145). The contrast in (i) may instead suggest that in the case of result nouns, speakers tend to interpret the genitive as a creator (which is incompatible with the context here). Interestingly in this respect, most examples used in experimental studies to show that possessed PNAs can be bound from outside their DP include concrete nouns (see e.g. (32a)). 71 Coreference of himself and Joe does not entail that Harry cannot be interpreted as the creator. As argued by Grimshaw (1990) for the case of nominalizations (see discussion in Sect. 3.2.2), the possible creator interpretation of the genitive does not necessarily imply that it is an argumental subject: the modifier reading can include the argument reading. Therefore, logophoric binding of himself in (144a) remains possible under the creator interpretation of Harry (i.e. Harry wrote the book) as long as Harry does not originate as the subject of NP. More generally, (126) only entails that the subject of a picture noun phrase must be interpreted as a creator, not that a genitive DP interpreted as a creator must have originated as subject of NP.
In context (a) implying that Brandon did not write the letter, the ungrammaticality of (149) results from a conflict between the genitive Brandon and pro subj , which must be syntactically represented given (141a) and refer to the letter writer (i.e. Chelsey) given (126), as represented in (150a). In context (b) in which Brandon did write the letter, no such conflict arises as Brandon originates as subject of NP, but logophoric binding of herself is blocked by the presence of the subject of NP as per generalization (117b).
(150) a. *Chelsey i found [(pro log-i/j ) Brandon k 's pro subj-i letter to 'erself i ]. b. *Chelsey i found [pro log-i Brandon k 's Brandon subj-k letter to herself i ].
(151) a. Context: Chelsey wrote a letter to Brandon when they started dating. *Chelsey i found Brandon k 's letter to himself k . b. Context: Brandon wrote a letter to himself without telling his girlfriend Chelsey. Chelsey i found Brandon k 's letter to himself k . (cf. Bryant and Charnavel 2020: 11) (152) a. *Chelsey i found Brandon k 's pro subj-i letter to himself k . b. Chelsey i found Brandon k 's Brandon subj-k letter to himself k .
Thus, several factors must be taken into account to predict the acceptability of possessed PNAs: the underlying position of the genitive (subject of NP or not), the obligatoriness or optionality of the subject of NP, the interpretation of the reflexive (logophoric or not, creator or not). Logophoric binding is blocked if the genitive originates as subject of NP, and any type of binding is ungrammatical if the genitive is not the creator in configurations requiring a subject of NP.

Conclusion
In sum, so-called Picture Noun Anaphors do not form a natural class: they are neither special, nor exceptional, but just like any anaphor, they systematically obey Condition A, which is a fully general principle. The reason why some instances of English PNAs-and other anaphors-seem to be exempt from it is that Condition A can be satisfied covertly. Furthermore, PNAs appear to exhibit a particularly irregular behavior because the availability of implicit binders in picture nouns is conditioned upon various interacting factors, some of which are specific to the nominal domain. Specifically, the illusion of PNA exemption results both from the possible implicitness of subjects in NPs, which must serve as binders when co-occurring with PNAs, and from the possible absence of subjects in NPs, which licenses binding by an implicit logophoric pronoun. Such apparent complementarity between subject and logophoric binding is due to a general binding-independent principle of competition between weaker and stronger forms, which regulates the availability of various potential bindees. By restricting the scope of logophoric binding, this principle obscures binding behaviors, especially in English where the weak reflexive form, which requires a coargumental subject as binder, is morphologically identical to the strong reflexive form. The interaction of all these factors gives rise to a complex set of binding possibilities for English PNAs summarized in (153). In our view, the failure of previous theories results from overlooking at least one of these factors. In particular, ignoring the possible binding by a logophoric pronoun (pro log ) led many to incorrectly assume exemption or long distance binding; ignoring the obligatory binding by the subject of NP (overt DP or pro subj ) resulting from the weak/strong competition principle led PBTs to wrongly build coargumenthood into Condition A.

(153)
Binding possibilities of English PNAs: a. If there is no genitive (whether overt or covert), the PNA can be bound by pro log or by any other DP that is not separated from the PNA by a subject or a tense boundary: In identifying all the various factors at play and specifying how they interact with each other, we hope to have solved the PNA puzzle in English without compromising on parsimony. For our investigation, we have used a toolkit inspired from the results of various recent works: for example, Charnavel and Sportiche's (2016) inanimacy-based tool, Ahn's (2015) prosodic diagnostics, Charnavel's (2020aCharnavel's ( , 2020b logophoric tests, Bryant and Charnavel's (2020) contextual control of genitive interpretations. While we have restricted our focus to English, we believe that this toolkit could benefit future crosslinguistic studies about binding theory and beyond. As we have shown, principles of binding interact with other properties of language, and, importantly, these properties can be language-specific: 72 for instance, the effect of competition between strong and weak forms will depend on the inventory of 72 See Charnavel (2020a: Chap. 5) for more specific discussion about how to reconcile a universal Condition A with apparent binding specificities in various languages. The discussion includes precise diagnostics to test the hypothesis that cross-linguistically, apparently exempt anaphors are logophorically bound (applied to Mandarin, Icelandic and Norwegian partly based on experimental data) and suggestions as to how to explain apparent binding domain variability under a universal Condition A.
anaphoric and pronominal expressions available in the language, and languages may vary with respect to the structure of DPs. Hence, a careful examination of PNAs in languages other than English would first require identification of the relevant factors. Due to the number of factors relevant to the solution, our exploration had to leave many questions for further investigation. In particular, the pro log hypothesis has many consequences for other perspectival elements beyond anaphors that would be interesting to explore. The competition hypothesis raises several issues related to the analysis of weak pronouns and reflexives. The conclusions we reach about the conditions on subject projection in NPs would be worth further testing on the basis of anaphoraindependent evidence. In sum, it seems to us that Picture Noun Anaphors still deserve specific attention, not as an exceptional class of elements, but as a probe into various questions such as the grammatical representation of perspective, the typology of pronominal elements, or the argument structure of nouns.