Saturday, 24 December 2016

What Are My Problems Now?

This is a follow-up to What Was My Problem?.

1. The Basis of Puzzlement about Modality

One line of investigation I would like to pursue now is into what might be called the basis of the puzzlement about modality. And as suggested by my experience of vaguely wrestling with a bunch of problems, before realizing that my strongest leading ideas for my thesis were really about some of these problems rather than others, I think this line of investigation may itself call for the distinguishing of various problems.

One locus of puzzlement about modality is the notion of metaphysical or subjunctive necessity as it applies to propositions. And one question about this notion is whether, and how, meaning comes into the picture. Also, just the question of how this notion relates to other notions, and the extent to which it can be analyzed (not necessarily in non-modal terms). Those problems are addressed, properly I hope, by the account in my thesis. But lots of what I was wrestling with at the beginning of my research remains, and does not attach specifically to the notion of subjunctive necessity de dicto; there is a lot that is puzzling about modality that my thesis does not address.

One puzzling thing which borders directly on my thesis work, and does have to do with the notion of subjunctive necessity de dicto, is the question of how this relates to de re modal constructions and quantification into modal contexts. But I have been very frustrated in my research here, and to be honest I have come to feel like it is a bit of a minor, abstruse issue compared to some of the more fundamental problems about modality (although I have no doubt that very interesting work could be done on that issue, and have a couple of ideas).

A more fundamental area I would like to work on is indicated by the question: Why is modality puzzling at all? But here too there are probably several puzzling things to distinguish. One thing I am not primarily thinking of, although it may end up becoming relevant to the problems I am grasping at, are questions about modality in an extremely general sense. For instance, the question of what unifies all uses of expressions which we call modal, or which we say are about possibility or whatever - including 'You can come in if you like', 'It could be that John is on his way', 'It is impossible for two colours to be in the same place', '"Hesperus is Phosphorus" is necessarily true', 'I can lift this weight', 'This apparatus has four possible configurations'. Also, questions about what generalizations can be made covering all or at least a great diversity of such uses, for instance about logical implication relations between them.

Rather, I am interested in trying to get at the basis of our puzzlement about what may be called objective modality. What does 'objective modality' mean? Well, one clear thing it does is exclude epistemic modals, like 'It could be that John is on his way' in natural uses. These are to be put to one side - at least initially - in the line of investigation I want to pursue. Likewise with uses of modal language having to do with permission. Within the puzzlement attaching primarily to objective modality as opposed to these set-aside kinds, important distinctions may have to be made. For instance, there may be a need to distinguish between more down-to-earth uses of modal language, for instance 'I can lift this weight', from what may be called more metaphysical uses - but not 'metaphysical' in the sense often used in modal philosophy, to mean either something like 'objective' or something a bit more specific, like picking out what I pick out with 'subjunctive'. Rather, by 'more metaphysical uses' I mean uses which are so to speak puzzling from the start. That is, where there isn't as much non-problematic, clearly useful use as in the case of 'I can lift this weight' and the like. E.g. 'The world could have been otherwise', 'Aristotle is essentially human'.

One way forward in this line of investigation would be to look critically and closely at philosophers' attempts to give a sense of the puzzlement about (objective) modality, often as a preliminary to some account or a survey of accounts. For instance, Sider's remarks on the subject in 'Reductive Theories of Modality'. But I think it will also be important to look within, so to speak, and keep seriously asking myself 'What is it that puzzles me about this?'.

2. Propositions and Meaning, Language Systems, and Our Expectations

Another line of investigation I would like to pursue has to do with the account of propositions and meaning sketched in chapter 6 of my thesis. That account appeals to a notion of an expression's internal meaning, cashed out in terms of the expression's role in the language system to which it belongs. This may raise questions about the nature of the system, and how we should think of it and describe it. In my thesis, I tried to remain quite open about this, emphasizing that I was offering a sketch, and that different fillings in of the detail here may be possible.

It was hard to avoid striking a false note here. For I do not think this is the whole story about my sketch, and the middle-Wittgenstein idea about role-in-system which it takes over; it may not be quite right to just think about it as a sketch of a theory, where some aspects are not filled in. For the very idea of what needs filling in, and how, should I think be scrutinized. It is not that I am advocating quietism, or defeatism, about questions about the 'language system' I appeal to. But I think that some of our expectations here may be in need of examination.

A curious thing happens in this territory - it is easy to become disoriented, and wonder what the problem was and what is needed now. Maybe sometimes in philosophy, as we solve problems, they slip from our grasp. Sometimes there is a strange feeling where we wonder something like: how could there be a solution here which is given in mere words? How could that ever do? We feel we still need to be taught something, or shown something. Could it be something practical, so to speak? I.e. something we could get through practice?

In the new year, I intend to use this blog to try to make some inroads into these and related problems.


  1. Your comment in your penultimate paragraph points in very fruitful directions: "[H]ow could there be a solution here which is given in mere words? How could that ever do? We feel we still need to be taught something, or shown something. Could it be something practical, so to speak? I.e. something we could get through practice?"

    There are two developments outside Philosophy that can illuminate these questions.

    First, statistical learning (AKA "neural networks," "deep learning," etc. etc.). "Good old fashioned AI" attempted to capture meaning in "mere words" -- increasingly sophisticated versions of code, logic, "frames", etc. -- all attempts to encode propositions. This approach failed.

    More recently statistical learning has been quite successful at many tasks including ones closely related to language understanding. Google just rolled out version 2 of their machine translation service, their first to be built purely on machine learning. It has been dramatically more successful than previous attempts -- it is now good enough to fool some sophisticated observers some of the time.

    This doesn't of course imply "real understanding" (whatever that would be) on the part of the system. But its success relative to the old approaches does confirm that statistical learning is a good path and propositional encoding is a bad path.

    Statistical learning is based on practice in various forms. With video games and in some cases human games like the recent Go tournaments, "pure" practice -- just trying and getting feedback on success or failure -- is sufficient. When teaching its system to translate, Google used existing translated texts, and then feedback on attempted translations. In all cases of statistical learning extensive practice is essential, and the practice is essentially never accompanied by detailed feedback to the system on why it succeeded or failed.

    Not surprisingly given this emphasis on practice, statistical learning leads to tacit knowledge -- the system knows how to perform tasks but doesn't typically have that knowledge in a form that's externally comprehensible or at all modular. Research is underway on converting this tacit knowledge to explicit (explicable, modular) forms -- which would get us back to propositional encoding of at least some of the knowledge by an entirely different route. The jury is out on whether and how completely this can be done.

    I'll put my second point in another comment.

  2. A second point about your question "[H]ow could there be a solution here which is given in mere words?"

    The current best conceptual framework for understanding natural language semantics is relevance theory ("relevance semantics" would be better but unfortunately is used for something completely different). To greatly oversimplify, it posits a convergence on shared meaning between two or more language users trying to anticipate each others' interpretation of what is said.

    This framework easily accommodates both propositional content and tacit knowledge and skills that can't be made propositional by the participants.

    To the best of my knowledge relevance theory hasn't been thoroughly formalized or implemented in a computational form. However I think it could be formalized, and it is the best candidate for a more abstract framework to complement the more concrete approach of statistical learning, which doesn't yield a conceptual framework for thinking about natural language semantics.

    A recent comprehensive review can be found at

  3. Also... looking for useful accounts of Relevance Theory I came across the original article laying out the conduit metaphor for communication (which is the default in English and deeply embedded in the language). The article can be read at ( but the appendix is missing.

    I think the expectation that we can capture all meaning in propositional form, and even more the expectation that propositions can be captured in a formal language, are inevitable results of basing a theory of meaning on the conduit metaphor.

    Since the conduit metaphor is demonstrably a bad account of human language, it isn't surprising that these attempts fail. My previous comments point towards alternatives. I can't quite imagine how analytic philosophy would look if it took advantage of alternatives like these, but I'd like to see.

  4. Glad you found the post interesting.