Why is Language Unlimited? David Adger’s book’s take on possessive ‘s

Everyone agrees that language is unlimited. But nobody agrees why.

What do we mean when we say that language is unlimited? Basically, just that. There is no limit to the number of different things you can say. Take a simple sentence like John is over there. You could just as well say That man is over there, That terrible man is over there, That terrible man who plays the piano is over there, Sue’s brother is over there, Claire’s sister’s brother is over there, Pete’s wife’s sister’s brother is over there and so on ad infinitum. 

So why is language unlimited? If you asked me, and probably most other members of the LuCiD team, I’d say it’s because the structure of language reflects the structure of the world. John, That terrible man, Claire’s sister’s brother and so on refer to exactly the same person. As a result, anything your language can do with John – such as saying is over there – it can do with That terrible man, Claire’s sister’s brother and so on. 

As one piece of evidence for this view, I might point to deaf children born to hearing, but non-signing, parents. In the absence of any language around them, these children make up their own. And these made-up languages come to have exactly this property I’ve been talking about. Given the world, and the things that people want to say about it, it just makes communicative sense to come up with a language in which different phrases that refer to exactly the same thing (John, That terrible man, Claire’s sister’s brother etc) work in exactly the same way. At the risk of belabouring the point, it would be sheer folly to invent a language that requires you to put JOHN in a different sentence position each time depending on the particular string of words you choose to use to refer to him (e.g., John is over there vs Is over there that terrible man). And communicative efficiency is something that languages – by which I mean their speakers – seem to care about a lot

In Language Unlimited, David Adger takes exactly the same phenomenon and draws exactly the opposite conclusion; that we have “an inner self that can’t help but impose hierarchical structure” (p.6); that we have “Universal Grammar, an innate and particularly human capacity, rather than just an ability to extract patterns and generalize them (p.60). 

What I’d like to do in this (sort of, but not really) book review is to zero in on one of the phenomena that Adger takes as evidence for his position and argue that “an ability to extract patterns and generalize them” actually explains the phenomenon better than a Universal Grammar that imposes hierarchical structure1.

The phenomenon is one that I’ve already mentioned in passing: possessive ‘s, as in Sue’s brother, Claire’s sister’s brother, Pete’s wife’s sister’s brother and so on. In fact, as Adger points out (p.184) there is no limit on the number of ‘ss at all: 

Arnie’s sister’s neighbour’s friend’s cat’s tail’s tip’s colour’s brightness blinded me. 

Adger argues that, without Universal Grammar, a child armed with just an ability to extract and generalize patterns could never learn the unlimited nature of possessive ‘s, given that – in the language that children hear – adults never use more than two of them. That is, adults do use phrases like Arnie’s sister’s neighbour but never (at least never in a 13-million-word sample) phrases like Arnie’s sister’s neigbour’s friend, let alone phrases like those in the example above, which has no fewer than eight ‘ss.

But Adger is missing a crucial point about the pattern-extraction-and-generalization approach: These patterns (what linguists like Adele Goldberg and Bill Croft call “constructions”) are generalized on the basis of similar relations. What do we mean by relation here? Handily, Adger gives a neat example just a few pages earlier (p.167). 

…think of some socks. Now think of some feet. There’s a relationship between those. Now think of hands. If we keep the relation between feet and socks in mind, then hands bear the same relation to gloves.

The relationship between socks and feet, or between gloves and hands can be summed up by the following pattern:


Once we’ve noticed this pattern, we can easily come up with new examples: hats and heads, say, or tea-cosies and teapots. Coming up with new examples of a pattern is called generalization. But notice how this generalization works: We generalize the pattern by coming up with new pairs that share the same relationship: [COVERER] [THING COVERED]. Notice that socks, hats and tea-cosies are not similar to one another in and of themselves. Neither are feet, heads and teapots. Socks, hats and tea-cosies are similar to one another only in the relationship they share with feet, heads and teapots.

Now, back to possessive ‘s. Suppose that child hears a lot of phrases with just one of them, phrases like:

Arnie’s sister

The cat’s tail

John’s wife

The book’s title 

By pattern-extracting across these phrases, the child forms a construction2 – or slot-and-frame-pattern – that looks something like this:


Notice that, just like with the [COVERER] [THING COVERED] example, the slots [POSESSOR] and [THING POSSESSED] are defined by their relationship to one another, not by their general similarity: Arnie isn’t similar to a cat or a book in any way, except that each can play the first role in a  [POSSESSOR] [THING POSSESSED] relationship.

Once the child has formed this construction, she can create new phrases at will by inserting new possessors into the [POSSESSOR] slot and new things into the [THING POSSESED] slot:

[Simon]’s [daughter]

[The girl]’s [bag]

Just like with the [COVERER] [THING COVERED] example, generalization happens on the basis not of general similarity, but similarity of relationships. A girl isn’t similar to a book. A bag isn’t similar to a title. But the relationship between a girl and her bag is similar to the relationship between a book and its title – one of possession3– and it’s this similarity that allows the construction to be extended to new [POSESSOR][THING POSSESSED] pairs4.

You can probably see where I’m going with this. Arnie’s sister, as a living breathing human is perfectly capable of possessing a cat. Therefore, we can stick Arnie’s sister into the [POSSESSOR] slot, just as easily as we can stick cat into the [THING POSSESSED] slot:

[Arnie’s sister]’s [cat]

Crucially, the fact that the phrase Arnie’s sister already contains a possessive ‘s doesn’t enter into consideration. Why not? Generalization, slot-filling, is based purely on relationships. Arnie’s sister + cat is a perfectly good example of a [POSSESSOR]+[THING POSSESSED] relationship. So Arnie’s sister can go into the [POSSESSOR] slot, just as easily if we’d decided to refer to her by her name (Jane, let’s say).

Now you can definitely see where I’m going. Suppose I want to bring Arnie’s Dad, Pete, into all of this. First I use my [POSSESSOR]’s [THING POSSESSED] construction to generate the phrase:

[Pete]’s [son]

No problem here because Pete is certainly capable of possessing a son. Now I simply plug this phrase, Pete’s son, into the [POSSESSOR]’s [THING POSSESSED] construction, giving:

[Pete’s son]’s [sister]

Again, there’s no problem. Pete’s son can “possess” a sister, so into the construction it goes. And, of course, Pete’s son’s sister can possess a cat:

 [Pete’s son’s sister]’s [cat]

And so, on ad infinitum.

How does all of this work under Adger’s claim of “an inner self that can’t help but impose hierarchical structure”? The idea is that the Universal Grammar we’re born with contains the knowledge that structures can be recursively combined; i.e., that the output of one rule – traditionally formulated as something like “add ‘s to a NOUN PHRASE” – can then be used as the input to that same rule. All I’ve done, of course, is to reformulate this rule in terms of relationships. I agree that possessive ‘s can be recursively applied, but to strings that can stand in a [POSSESSOR] relationship with a particular [THING POSSESSED] rather than to strings that meet some syntactic criteria for NOUN PHRASE.

A quick aside: Adger notes that while many languages (he cites English, Hebrew and Mandarin) allow an unlimited number of possessive ‘ss, others (he cites German and Pirahã) allow only one. How does my relationship-based account, which – by its very nature – is unlimited, explain learning of this latter type of language? Simple, statistical preemption: If a speaker clearly means something like Pete’s son’s sister but instead says something like The sister of Pete’s son – and this happens many times across many different scenarios – the speaker infers that the heard form, and not the unheard form, is the conventional way of expressing “possessed possession” in the relevant language. Adele Goldberg recently wrote a book about this phenomenon (it’s called Explain me this; a phrase that sounds ungrammatical to our ears precisely because we always hear the preempting alternative Explain [THING] to me). I’ve also provided evidence for statistical preemption in quite a few different studies, in particular this one and this one

Now, I promised earlier that my relationship-based account wouldn’t just equal Adger’s in terms of explaining the data, but actually beat it. How so? Well, because the fillers of the slots have to be able to form a plausible [POSSESSOR] [THING POSSESSED] relationship, my account avoids generating all kinds of borderline garbage5 like:

The book’s wife

Pete’s bag’s tail

The neighbour’s tip’s brightness

By contrast, a possessive ‘s rule formulated purely in terms of NOUN PHRASEs


generates all of this freely (and more, ad infinitum). Adger would presumably argue that this is a good thing because these types of phrases, while strange, are nevertheless grammatical. He would say, I imagine, that the fact that books don’t usually have wives; bags, tails; or neighbours, bright tips is not part of our knowledge of language, but of the world.

And if he did, I would say this: Given that we already have detailed real-world knowledge of what can possess what, the relationship-based construction [POSSESSOR]’s [THING POSSESSED] is all we need to generate infinitely long strings of possessive ‘ss, while avoiding borderline garbage. In contrast, the [NOUN PHRASE]’s [NOUN PHRASE] construction doesn’t restrict our utterances to those that make sense. To do so, it needs to bring back the notion of a plausible [POSSESSOR]-[THING POSSESSED] relationship: the very thing that – compared with my account – it throws out.

So, is language unlimited? Of course. Why? Because slot-and-frame constructions are infinitely accommodating, provided, that is, the slot-fillers never forget their relationship goals.



[1] I set out a similar argument for complex NOUN PHRASES (things like That terrible man who plays the piano) in this LuCiD paper. I also did my best to test this argument with an experiment. Unfortunately, the results were pretty inconclusive (this isn’t a euphemism for “supported the other side” – they really could be argued to be consistent with either side).

[2] Researchers who are broadly sympathetic to this idea of a construction differ as to whether they think these constructions are actually stored in memory, or just a metaphor for a kind of generalization that speakers make across stored utterances. I recently wrote a very long article arguing for this latter point of view, but don’t worry about it now – it doesn’t matter either way for our purposes here.


[3] Of course, the sense in which a girl possesses a book is quite different to the sense in which a book possesses a title or Arnie possesses a sister, but that’s no problem – we just have to define possession sufficiently broadly as to include more-or-less metaphorical types.


[4] If you’d like experimental evidence for the idea that construction slots are filled on the basis of relations, I invite you to check out my papers on the English and Indonesian passive constructions here and here. In this case, the relevant relationship is between [PERSON/THING AFFECTED] and [ACTION/EVENT AFFECTING THEM].

[5] Of course, we can get some kind of poetic or abstract meaning from these borderline-garbage phrases. But this precisely because we can squeeze them – at a push – into our relationship-based [POSSESSOR]’s [THING POSSESSED] construction.


Leave a Comment

* Indicates fields are required