1 The phenomenon

This paper has both a descriptive and an analytical goal. First, we describe, perhaps for the first time, the properties of a kind of relative clause found in the English of some speakers (possibly limited to younger generations), which we will call a wh-which relative, exemplified by (1):

We will argue for an analysis of this relative-clause type that bears on Cable’s (2010a; 2010b) proposed explanation for apparent pied-piping in wh-questions. Building on these proposals and Coon’s (2009) extension of them to Ch’ol (and some other Mayan languages), we propose an analysis of these English relative clauses almost identical to Cable’s and Coon’s proposals for wh-questions.

In particular, we argue that instances of apparent pied-piping seemingly motivated by the presence of a non-head relative wh word are actually mundane instances of movement of the maximal projection of a different element, which is the actual head of the moving constituent and the real reason for movement — just as Cable and Coon suggested for apparent pied-piping in wh-questions. But we will also suggest that this might not be the true characterization of all movements that appear to involve pied-piping, the exciting prospect that emerged from their work. We will suggest one analytic path that might eliminate the problem, but leave a fuller resolution of this puzzle open as a topic for future research.

One surprising property of the relative clause in (1) when compared with the more commonly described relative clauses characteristic of “Standard” or “Mainstream” English varieties (henceforth “non-wh-which English”), is, of course, the presence of a second wh-form which. We are not surprised by the presence of who(m) in a relative clause whose head is [+human] (including vacationing snowmen), but relative clauses in non-wh-which English introduced by which are otherwise generally limited to those that modify inanimate heads. We examine several variants of this construction below, not all of them introduced by whom — but they do all contain an instance of which seemingly doubling the more expected wh-form.¹ The optional presence of of also needs an explanation, of course. We will argue that its nature and properties hold the key to the construction as a whole. We will suggest that a phonologically null variant of this element is present in wh-which relatives even when of is not pronounced. We will therefore refer to it as R_of (which we will suggest is a variant of an R-element that heads the Ā-moved phrase in non-wh-which English relative clauses as well), and will provide a generalization concerning the circumstances under which it may be overt. We will ultimately propose that what distinguishes R_of from other instances of English R is the fact that it triggers movement of a wh-headed phrase to its specifier (with which marking the original position of the moved element) — at which point the link to Cable’s and Coon’s proposals will be clear.

We have not formally investigated the distribution of wh-which relatives among English speakers. The construction seems perfectly natural to the first author of this paper, a speaker in her early 20s from Hawai’i, and to some but by no means all of her peers — but it is completely alien to the second author and to many other speakers. We have not been able to determine whether the construction is more prevalent among younger than among older speakers (though we suspect this is the case), nor have we discerned any particular geography to its distribution. The construction seems to be most prominent in casual discourse, and online attestations are easiest to find in social media posts and other texts of a similarly informal nature — but are not limited to such environments. Though we suspect most of these attestations are written by North American speakers, some appear to come from the UK and Ireland as well. In this connection, it is striking (and unfortunately not a property for which we will have an explanation) that the form whom is strongly preferred to who, despite the fact that whom is otherwise regarded as a more formal option than who.

Throughout this paper, we offer examples from online sources wherever we can — mainly because the very existence of this construction comes as a surprise to many speakers. (All internet citations have been verified on June 27, 2023.) Examples not credited to an online source reflect the judgments of the first author and four speakers that she consulted who accept examples like sentence (1).²

A sampling of relative clauses from online sources introduced by whom of which is given below, followed by examples introduced by whom which without overt R_of.

That wh-which relatives are not slips or typing errors is suggested not only by the speakers’ intuitions that we report, but also by several prescriptivist websites on which readers have asked about the propriety of whom of which relatives (and are uniformly advised to spurn the construction as non-standard and redundant). For example:

“I recently encountered this expression and I’m pretty much stumped. People seem to be using it in place of ‘who’. Example:

[…] they were developed by non-medical professionals whom of which have applied their skills to help solve a health care/medical problem they had faced.

“Another example:

As well as Dave, special thanks goes to the ‘Girls in Pink’ for providing a high service throughout the whole day, as well as club chairman, Bob Thomsett and Chrissie whom of which manned the kitchen which was busy all day.

“I can’t quite wrap my head around the expression, but I figured it might be one of the many idiosyncrasies of the English language. Is this a valid expression or some newfangled bastardisation?”

(https://english.stackexchange.com/questions/106180/whom-of-which-a-valid-expression)

Some typical responses:

“No, it is not a valid expression. It’s simply bad English”

“They’re wrong. Who is all you need and what you want. And there’s more wrong with those sentences besides ‘whom of which”’

“I’m voting to close this question as off-topic because it is based on a grammatical mistake and no definitive answer could be made.”

A similar extended discussion of the construction can be found at https://www.usingenglish.com/forum/threads/using-whom-of-which.159982. The website Grammarhow, which describes itself as “on a mission to help you become better at English”, devotes an entire page to this construction — followed by multiple examples of appropriate substitutions for it.

“In English, there can be plenty of times when new words or phrases seem to be created as idiosyncrasies. Phrases such as “whom of which” seem to develop over time, even though they don’t make all that much sense. This article will explore whether it works grammatically.

“Is ‘Whom of Which’ Correct?

“‘Whom of which’ is never grammatically correct. It never makes sense to write this phrase as it is a long-winded version of the simple ‘who.’

“There is never a situation where ‘whom of which’ makes sense. It’s not something that you should be comfortable with using. Most native speakers will look at it and think it sounds bizarre.

The only time that people seem to use it is when they’re trying to be overly wordy. They might use it when they feel like they need to ‘pad out’ the word count in their writing. Even in these cases, ‘whom of which’ is never correct.”

(https://web.archive.org/web/20221219114332/https://grammarhow.com/whom-of-which/)

There is of course no better sign that a syntactic construction forms part of the real grammar of real speakers than prescriptivists protesting against its use.

Apart from the contents of its left periphery, the syntax of wh-which relatives is that of non-wh-which relative clauses, involving Ā-movement from within the relative clause. For example, this movement may cross a clause boundary, but obeys islands, as shown in (4); and can license a parasitic gap, as shown in (5):

As examples throughout this paper demonstrate, the construction is available in both restrictive and non-restrictive relatives, as well as relative clauses with a variety of modified heads — definite, indefinite, quantificational, and pronominal. Throughout our investigation of this construction, we have come across a general feeling on the part of speakers that the restrictive relatives are somehow “more restrictive” than their non-wh-which counterparts, but we have not been able to make this feeling precise — and the use of the construction in non-restrictive contexts makes it clear that there is no restrictivity requirement imposed by the construction.

We will thus assume, possibly wrongly, that wh-which relatives are interpreted semantically just like their non-wh-which English counterparts. This favors analyses in which the main differences between the two types of relative clauses can be shown to be purely syntactic in nature, ideally invoking parameters of variation familiar from other puzzles with a similar character. Our proposal will have that property.

Turning now to the left periphery of the construction, we note first that a whom-(of)-which relative can only modify a [+human] nominal, just like non-wh-which English who-relatives. If the head is inanimate, once might expect a which-(of)-which relative to be possible. These are rejected by the speakers we consulted, possibly to avoid repetition — but online attestations, though seemingly rare, do occur:³

Where and when are also possible as the initial wh-form in a wh-which relative. Speakers may feel slightly less comfortable with these than with examples whose first element is whom, and attestations seem to be correspondingly sparser:

2 More than just which: towards an analysis

Since the transformation of a wh-word into a complementizer is a well-known diachronic path (Heine & Kuteva 2007: 242–244; Kuteva et al. 2019: 354–355), one might wonder whether which is a complementizer in the variant of the construction lacking overt R_of , in which case whom could be viewed as a specifier of CP in a “doubly-filled COMP” configuration.⁴ This proposal makes it hard to find a place for R_of in the variant that includes it, however, and runs afoul of constituency evidence from coordination that we present in section 5. While this introductory section might be the optimal place to present that evidence, creation of the relevant examples will require us to exploit certain resources of the construction that we are not yet in a position to introduce.

We argue instead for a different proposal, according to which the sequence wh-(of)-which seen in the examples presented above forms a constituent Ā-moved in its entirety to the specifier of CP by the same process familiar from non-wh-which English relative clauses like the person who you met. We propose an internal structure for this constituent like that seen in (8) in which “wh” occupies a specifier position internal to the moved constituent and R_of is the constituent’s head (in the examples seen so far). The reason we have surrounded both “wh” and “which” with sets of three dots will become clear as we proceed.

In the non-wh-which English counterparts of all the examples seen so far, what is relativized would be a single word, e.g. the snowmen who(m) the children loved (though of course more complex analyses of words like where are imaginable). When a phrasal PP is relativized — a case where non-wh-which English appears to show pied-piping (e.g. the person with whom you disagree) — one of the outcomes for speakers who command the wh-which construction sandwiches the preposition between the initial wh and which, as seen in (9) and (10) below.⁵

The presence of a preposition between wh and which as in (9) and (10) blocks the presence of overt R_of :

This fact might seem to suggest that a preposition sandwiched between wh and which occupies the same position as that occupied by R_of . Note, however, that these sandwiched prepositions head PPs that are all licensed in the clause-internal position from which relativizing Ā-movement was launched. In this respect they contrast with R_of , which is not licensed in the clause-internal relativized position (e.g. the children loved (*of) the snowmen.) This fact already militates against the proposal that R_of occupies the same position as the sandwiched prepositions of (9) and (10).

We propose instead that the sandwiched prepositions in (9) and (10) are heads of a PP complement of the position occupied by R_of , and that the obligatory silencing of R_of should be explained in some other way (to which we turn below). We thus propose a structure like (12) below (with P thus belonging to the three dots preceding which in (8)):

In fact, a PP containing which as a subconstituent is not the only kind of full phrase that can occupy the position we are identifying as the complement of silent R_of . A nominal containing which as a subconstituent can also occur in this position:⁸

Strikingly, overt of is disallowed in such examples as well.⁹ A variant of (13f) with of, for example (e.g. *barbarians whom of stories about which are told to scare children) is not acceptable, and we have not found online examples with this configuration. This fact supports our view that the complementary distribution of overt of and the boldfaced prepositions seen in (9) and (10) is not a sign that they occupy the same position. Instead, we propose the following rule governing the pronunciation of R_of :

3 Proposal

There is a simple way to convert any wh-which relative clause into a non-wh-which relative clause: replace which with the element we have been calling wh, replace whom with who as appropriate, and do not pronounce R_of . This procedure will convert the snowmen whom (of) which the children loved into the snowmen who the children loved; those whom with which you disagree into those with whom you disagree; and barbarians whom stories about which are told to scare children into barbarians stories about whom are told to scare children. In this fact, we suggest, may lie the key to the correct analysis of the wh-which construction: the initial wh originates in the position of which, and reaches its surface position by movement to the specifier of R_of . Our preliminary proposal for the analysis of a wh-which relative is thus the following:¹⁰

In the simplest examples with which this paper began, the complement of R_of moves within RP_of to form its specifier, and the entire RP raises to spec,CP, as shown in (16a).¹¹ Examples like those in (9) and (10) involve preposition-stranding subextraction of a wh element to form a specifier of RP, as shown in (16b). Examples like those in (13) involve comparable subextraction of wh from a nominal, as shown in (16c):

If this analysis of wh-which relatives is correct, what looks like wh-movement in a wh-which relative —i.e. movement triggered by a [+wh]-probe on C — is actually RP_of movement. The fact that R_of is sometimes silent but (silent or not) requires a wh creates the illusion that it is C that needs [+wh] — but the relationship is actually indirect: C interacts with RP_of and R_of interacts with wh. This analysis can immediately be extended to non-wh-which English relatives like the snowmen who the children love that lack the distinctive properties of wh-which relatives, if these involve RP movement as well. All that is necessary is to posit a distinct flavor of R for these relatives that does not trigger movement of its wh goal (unlike R_of ) and is always silent.¹²

An attractive feature of this proposal, we believe, is its extreme unoriginality. It is in fact a faithful analogue for relativization of Cable’s (2010a; 2010b) theory of pied-piping phenomena in wh-questions.¹³ Our proposal is especially close to Coon’s (2009) analysis of the syntax of questions in the Mayan language Ch’ol that built on Cable’s work.

Following work by Hagstrom (1998) and others on similar phenomena in other languages, Cable observed what looks like interrogative wh-movement in Tlingit always includes a particle sá at the right periphery of the phrase that undergoes Ā-movement to C, a particle that he assigned to the category Q. Since Tlingit is head-final in all syntactic categories, Cable concluded that sá is the head of the moved constituent. Consequently, what interrogative C probes for is Q, and what it causes to move is its maximal projection QP. Internal to QP, Tlingit Q in turn requires the presence of a wh element within its complement, just as we have suggested for R_of internal to RP_of . When a question involves what has traditionally been viewed as pied-piping, Tlingit makes clear that we are simply observing a situation in which Q has merged with a constituent larger than the wh-word itself, and what moves is QP:¹⁴

When an island boundary separates the wh-element from interrogative C in Tlingit, sá must follow the entire island. Certain locality conditions also hold between sá and the wh that it is looking for.

Cable argued that Tlingit holds the key to the proper analysis of interrogative Ā-movement and pied-piping cross-linguistically. In a language like English, in which we see no overt Q in wh-questions, Cable proposed, Q is still present but phonologically null — and its interaction with wh can still be detected in locality conditions that limit the syntactic distance between the wh-word and the edge of the constituent that Ā-moves.

Coon used Cable’s proposal to explain a puzzling pattern of pied-piping phenomena in Ch’ol (earlier explored in the closely related language Tzotil by Aissen 1996). Unlike Tlingit, but like English, Ch’ol does not display an overt Q-morpheme at the edge of the phrase that Ā-moves in wh-questions. However, the wh-word of the phrase obligatorily occupies its left-periphery. Thus, though a non-wh possessor always follows the possessum in Ch’ol, as seen in (18) —

— a wh-possessor will always be found to the left of the possessum, with the options seen in (19a-c), which also displays Coon’s analysis. As our diagrams show, Coon proposed that this pattern is a consequence of a null Q like that posited by Cable for English questions, but with a twist: the wh-probe on Ch’ol Q (unlike English and Tlingit Q) triggers movement of its wh-goal to its specifier. In a structure with possessor recursion and a wh-element as the innermost possessor, the size of the constituent that moves to spec,CP is determined by how high Q attached within the nominal. Wherever Q attaches, it attracts the wh-element to its specifier. As a consequence, the Ā-moved interrogative phrase comes in a variety of sizes, depending on how big a constituent Q merged with — but its leftmost element is always the wh-word, as illustrated in the examples of (19), where the arrow marks the movement of who to form a specifier of Q, and the QP as a whole has moved to the specifier of CP from the trace position indicated with t_QP.¹⁵

Our analysis of wh-which relatives thus proposes a head in relative clauses that plays a syntactic role analogous to that played by Q in Cable and Coon’s proposals — which is why with malice aforethought we have called the head found in wh-which relatives R_of (R being the letter after Q, as well as the first letter of “relative”), and treat R_of as an instance of R. Like English and Ch’ol Q but unlike Tlingit Q, English R can be silent — but like Tlingit Q, it can also be overt for speakers whose grammar permits wh-which relatives. Like Ch’ol Q but unlike English and Tlingit Q, English R triggers movement of a wh element in the domain that it c-commands.

It is also a general cross-linguistic fact about the QP proposed by Cable for questions — and equally well must be a fact about RP in relative clauses under the analysis developed here — that in all of its non-Ā-positions, its external syntax is identical to the external syntax of the complement of its head. If R_of or Q merges with a DP, for example, the resulting RP_of or QP behaves as a nominal behaves — satisfying selectional requirements that only a DP can otherwise satisfy, triggering agreement as a DP does, and undergoing A-movement as if the Q or R_of that merged with it were not there. Likewise, if R_of or Q merges with a PP, the external syntax of its maximal projection below the CP level is identical to the syntax of that PP. Though Cable (2010b: 62 ff.) attributes this phenomenon in the case of QP to the semantic type transparency of Q, we will assume that QP and RP are syntactically transparent as well, inheriting syntactic features from the complements of their heads as well as from the head itself, exemplifying Grimshaw’s (2000; 2005) notion of “Extended Projection” (as proposed for Q and other similar particles by Branan & Erlewine to appear: section 4.2).¹⁶ We will continue to notate phrases headed by R and Q as “RP” and “QP” in this paper for the sake of brevity — assuming throughout, however, that the features of their complement also form part of the labels of these phrases — so that an RP whose head takes a DP or PP as its complement, for example, also bears the features of that DP or PP.

The only actual novelty of wh-which relatives that we have seen so far besides the pronunciation condition in (14) is the fact that the original position of the wh-element is obligatorily marked by an overt element which rather than a gap. Crucially, the relation between which and the wh-element in the specifier of RP_of obeys islands — a fact that strongly argues for the movement that we are proposing, but also suggests that which is not a resumptive pronoun (which one might expect to ameliorate island violations), but perhaps a syntactic companion of the wh-element obligatorily stranded by movement.¹⁷ The unstarred examples below, which do not violate islands, are judged awkward and stylistically peculiar — but contrast clearly in acceptability with the starred examples:

We do not have an explanatory account of the obligatory use of which, rather than a gap (or some other element) as a marker of the extraction site, unfortunately. So we stipulate this fact and leave it as a topic for future investigation.¹⁸

Finally, we should note at this point that there does not appear to be any form of interrogative for the wh-which speakers we have consulted that is comparable to a wh-which relative.¹⁹ This suggests that even for for wh-which speakers, Q contrasts with R in lacking a pronunciation as of and does not attract wh as its Ch’ol counterpart does.

4 More than just wh: is pied-piping a concept after all?

An important consequence of the Q theory of interrogative phrase movement is its potential to explain pied-piping phenomena without the need for a special theory of pied-piping. If one views wh-question or wh-relative clause formation as a result of a wh-probe on C, movement of a phrase headed by a wh-word is expected — but not movement of a phrase that merely contains a wh-word that is not its head. If mere containment is sufficient (i.e. if apparent pied-piping turns out to be real), the theory must include some special mechanism such as “feature percolation” that copies the wh feature from a non-head to a node that dominates it. If, on the other hand, the probe on C is actually looking for Q or for R — with Q and R being the only heads with an actual interest in wh — then what looks like pied-piping can be understood instead as garden-variety movement of the maximal projection of the head bearing the feature C is actually looking for. If this is the end of the story, then, as Cable (2010b: 9) writes, “the concept of ‘pied-piping’ may be eliminated from the theory of grammar”.

It is in fact important to the logic of Coon’s discussion that neither of the movements she discusses in her analysis of Ch’ol wh-questions requires anything beyond movement of the maximal projection of the head bearing the feature sought by the probes that she posits on C and on Q. Indeed, the central observation that motivated her paper is the fact that there is no fourth word-order pattern alongside the three seen in (19). Just as there is no need to posit a special pied-piping mechanism for movement of an interrogative phrase to C, there is also no pattern that would require us to posit pied-piping in the movement of wh to Q. Q attracts a wh-headed constituent — and is not interested in other QPs or any phrase that merely contains wh but is not headed by it. As a consequence, we never find any sign that the phrase ‘whose dog’ moves internal to the QP in examples like those in (19). What moves to the specifier position of CP never looks like ‘[whose dog] plate’ or ‘[dog (of) whom] plate’.

Troublingly, the situation looks different in wh-which relatives, at least under the analysis we have developed here. Constructions seemingly analogous to the missing fourth pattern in Ch’ol do exist. In wh-which relatives it is possible for a phrase that merely contains wh but is not headed by it to occupy the initial position in this construction. This raises the possibility that though Q and R might be real, and though their properties might indeed permit us to dispense with pied-piping as an account of specific examples, something like pied-piping and the special mechanisms needed to account for it is also real after all.

The examples in (22) show movement of a nominal that contains a non-head wh to the specifier of R, with and without overt of. The more complex examples in (23) show movement of similar nominals stranding a preposition.²⁰

Clearly, if our overall approach to the construction is correct, examples of this kind pose a threat to the prospects for the elimination of pied-piping mechanisms as a component of the grammar. It is worth noting, however, that examples like these can at least be fit into a pigeonhole compatible with our overall analysis that removes the threat, albeit by purely technical means. A centerpiece of Coon’s account of the Ch’ol pattern is a sharp distinction between C, which attracts only the maximal projection of Q, and Q itself, which attracts only the maximal projection of wh. If QPs were permitted to contain QPs within them — and if, crucially, Q were permitted to attract another QP (or if Q and wh were not distinct features in the first place) — we might expect a “roll-up” movement pattern to be acceptable, in which wh raises to the specifier of a QP that then raises to form the specifier of a higher Q within the same interrogative phrase. Such a derivation is in fact forbidden in Ch’ol:

Let us suppose, however, that wh-which English dialects permit an analog for relative clauses to what Ch’ol forbids for questions: a probe on R that attracts another RP rather than wh. If so, examples like those in (22) and (23) can be derived as instances of RP movement forming a specifier of a higher RP:

If RP recursion is possible, then we predict the possibility of “snowballing” movement whenever a movement-inducing R_of embeds a constituent that contains another movement-inducing R_of (with which present at each extraction site). In fact, wh-which recursion of exactly this sort does appear to be possible:²²

This pattern has a precedent outside English. As explored by Huhmarniemi (2012: 62, 223) and by Huhmarniemi & Brattico (2013), a comparable pattern of “snowballing” movement in Ā-constructions is observed in Finnish. This pattern is exemplified in (28), where interrogative ‘what’ has moved within the PP ‘towards what’, which itself has moved within the phrase that undergoes clause-level Ā-movement. Note that the movement within the phrase that moves to the specifier of CP leaves an ordinary gap rather than an element comparable to which (whose presence in the wh-which relatives we have in any case not successfully explained):

The absence of an overt Q (or R) element in Finnish has made it possible to envision proposals to explain this phenomena that do not involve an interrogative or relativizing head as the trigger for what they call “secondary movement” within the phrase that moves to the specifier of CP (the path that Huhmarniemi 2012 and Huhmarniemi & Brattico 2013 in fact take). If we are correct in analyzing the overt of in (some) English wh-which relatives as just such an element, however, then we may seek an analysis of the Finnish patterns along similar lines to that proposed here.

At the same time, we must regard our proposal in this section as a speculative pigeonhole, a proof-of-concept demonstration that Cable’s most ambitious claim might not have to be weakened after all, if Q and R may embed and attract other instances of QP and RP. Whether this proposal is actually correct is a question we must leave open, pending stronger evidence for the possibility of RP recursion.

One obvious issue to resolve concerns the semantic contribution, if any, of R and Q, if RP and QP recursion is possible. One of Cable’s important contributions was a semantics for questions that assigned a specific role to Q, explaining its obligatory presence in wh-questions. The extent to which QP and RP recursion can be accommodated into this theory remains an open question that we will not attempt to answer here. Even if R bears a movement-inducing probe that searches for RP in wh-which dialects of English, it must also be required to find a wh-phrase in its c-command domain (as in Cable’s and Coon’s proposals) — perhaps as a consequence of a distinct probe, or perhaps as a necessity imposed by semantic interpretation.

Another issue concerns the nature and scope of cross-linguistic variation. As (25) suggests, on our analysis of the English side of the puzzle discussed in this section, movement-inducing R_of must be permitted to embed not only a phrase headed by a second occurrence of movement-inducing R_of , but also a phrase headed by plain non-movement-inducing R. This fact correctly predicts the existence of variants of examples like those in (26) in which no movement takes place within the most embedded RP:

This option is not available in Finnish, where Huhmarniemi & Brattico (2013) observe that deviations from snowballing pattern seen in (28) are forbidden or possible only with an echo-question reading.

Taking stock: we have seen languages without movement-inducing Q or R (Tlingit; non-wh-which English), a language in which Q induces movement of wh alone (Ch’ol), and languages in which Q or R may induce movement of QP or RP (wh-which English dialects optionally in relative clauses; Finnish obligatorily). Against this backdrop of limited variation, what the Q theory of wh-questions and the analogous R theory of relative clauses continue to offer is a uniform and parsimonious explanation of why pied-piping phenomena should exist cross-linguistically in the first place — reinforced by the documented overtness of Q and, we have argued here, R in some languages, in just the positions where we expect to find them. We leave for future work the systematic charting of the spectrum of variation.

5 Constituency tests

Finally, returning to the English wh-which construction, let us redeem a promissory note from the introductory section of this paper. The availability of the complex phrasal specifiers of RP_of seen in (22) and (23) and the availability of the complex phrasal complements to R_of seen in (9) and (13) now permit us to construct examples that use coordination to verify the overall constituency for wh-which relatives that we have proposed throughout.

The examples in (31) below support the internal constituency of RP_of proposed here. As shown in (31a–b), it is possible to coordinate contrasting phrases occupying both the specifier and complement positions of RP_of , though some judgments are reported as slightly uncertain. Note as well that the acceptability of a complex phrase in both the position of whom and the position of which in initial examples such as (1) already eliminated the possibility (which one might entertain at first) that either element is itself a head in the left periphery of the clausal spine. What might remain at stake if one is searching for alternatives to our analysis is the constituency of the wh-which sequence as a whole, for which (31c) below is relevant.

Similarly, the examples in (32) below verify the constituent structure proposed here for the CP within which Ā-movement of RP_of takes place. Example (32a) coordinates two relative clause CPs as a baseline. The remaining examples shave off words one by one, and show that coordination cannot target any smaller sequence of words until what is coordinated is a TP (example (32a)), as predicted:²⁴

The wider range of configurations discussed in this section also raises questions about the constituency of acceptable and attested wh-which relatives whose leftmost element is a P:

In principle, the leading preposition in these examples might belong to the specifier of RP_of that underwent Ā-movement — in which case, they represent instances of the same phenomenon as the examples in (23), except that it is a PP that moved within RP_of . If so, then it is a point of interest that which can mark the location from which a PP has moved, not just a nominal. On the “proof-of-concept” analysis just sketched, the structures in question once again would involve a plain RP whose head does not trigger movement attracted to specifier position by a higher R_of .

Alternatively, the leading preposition might be external to the RP_of within which movement of a wh-element took place, in which case these examples represent something new in this section: a non-wh-which English RP embedding an RP_of :

If relative where and when are themselves prepositional, then the examples in (7) may have already taught us that the first of these parses is a possibility. Whether the second parse is also available could in principle be tested by coordinating wh-(of)-which following a single P, e.g. people with [whose friends of which] and [whose enemies of which] I am equally comfortable. The judgment is less than clear, however, so we will leave this question open.

6 Conclusions

If our analysis of wh-which relatives is correct, it provides yet one more instance of the “unity in diversity” of syntax structures across the world’s languages. In the present case, we have suggested that ideas developed to explain Tlingit interrogatives, which have already illuminated puzzles in Ch’ol, might also illuminate the novel set of puzzles discussed here — puzzles posed by relative clauses in a (possibly innovative) dialect of English, some of which are strongly similar to phenomena earlier observed in Finnish.

Of course, the puzzles posed by English wh-which relatives have not been fully solved here, so it is possible that an alternative view of the construction (with different connections to phenomena in other languages) will fare better with the remaining problems. In particular, though we have argued that the relation between which and the pre-R_of phrase is a movement relation, we do not know why the original position of this relation should be marked by which. In addition, the surprising preference for whom over who remains unexplained.²⁶

Appendix: Wh-which free relatives

Gary Thoms (personal communication) has pointed out to us that though there is no interrogative counterpart to wh-which relatives, free-choice free relatives (FC-FRs; see Giannakidou & Cheng 2006) do participate in the wh-which construction. They occur as both arguments and adjuncts. The examples in (35) show wh-which FC-FRs as free adjuncts and appositives:

When functioning as arguments, however, wh-which FC-FRs present an interesting puzzle concerning its external and internal syntax. As we noted when first discussing Q and R, the external syntax of their maximal projections (apart from their susceptibility to Ā-movement) is identical to that of the their complements. A phrase headed by Q or R with a DP or PP complement has the A-syntax of a DP or PP respectively. In response, we proposed that the features of the complement of Q and R are shared by the maximal projections of Q and R, adopting Grimshaw’s (2000; 2005) notion of “Extended Projection”.

This is true of argumental wh-which FC-FRs in their pre-Ā-movement positions as well — but with a twist concerning the “Matching Effect” characteristic of free relatives (Bresnan & Grimshaw 1987; Groos & Riemsdijk 1981). In argumental wh-which FR-FCs, it is not the syntactic behavior of the specifier of CP that determines the behavior of of the free relative as a whole (which is the Matching Effect as normally described). but rather the specifier of the specifier of CP. Thus, in (36a–36c), though R_of has merged with a PP, and behaves like that PP in its original position, it is the nominal whom in Spec,RP_of that determines the syntactic behavior of the free relative clause as a whole, which thus behaves as a nominal (not a PP) within the higher clause:

The Matching Effect in free relatives is sometimes analyzed as the result of the apparent specifier of CP being permitted to project, rather than C (Larson 1998; Iatridou et al. 2001: 224). We might analyze wh-which FC-FRs in the same spirit. It cannot be the Ā-moved of RP in Spec,CP that provides the head for the free relative, so it cannot be this instance of movement that is followed by projection of the moved element. Instead there must be an additional movement step after this initial instance of Ā-movement that extracts the internal specifier from the RP specifier of CP — and this is the element that projects after movement. This second movement step might be triggered by a higher silent head (perhaps relevant to the semantics of free relatives), or it might be self-motivated “Münchausen movement” of the sort proposed by Fanselow (2003) and others. The absence of a Matching Effect in non-argumental FR-FRs like (35) can then be attributed to the absence of this second movement step.

Why should it the specifier of the specifier of CP, rather than the specifier of CP itself that raises in this fashion? One avenue to explore might be the possibility that the specifier of CP is a criterial position, frozen in place to permit it to type the clause as a relative or interrogative — while the specifier of RP and QP might not have that status. What raises and projects to form the FC-FR is therefore the highest maximal projection that does not have this property.

We will not develop this proposal further here, as our main goal in this appendix has been to document the puzzle and note its interest for the analysis of the Matching Effect.

Abbreviations

ACC = accusative [Finnish], A3 = 3rd person ergative/genitive [Ch’ol], CL = proper name clitic [Ch’ol]; ESSA = ‘while doing’ [Finnish], ITV = intransitive verb suffix [Ch’ol], NOM = nominative [Finnish], PART = partitive [Finnish], PRFV = perfective aspect [Ch’ol], PX = agreement inflection (possessive suffix) [Finnish]

Competing interests

The authors have no competing interests to declare.

Author contributions

The authors contributed equally to this article, and their names are listed alphabetically.