Annotated table of contents

Knowledge in the image of inquiry

1. Finding our way to knowledge

Jonathan Schaffer (Knowledge in the image of assertion) has argued that we can learn something about the structure of knowledge by investigating the nature of assertion, to which knowledge bears a substantive connection. Knowledge also bears a substantive connection to inquiry, and I maintain that we can similarly learn something about the nature of knowledge by investigating the nature of inquiry.

2. The nature of inquiry
• Deduction

Issac Levi develops an account of inquiry in terms of revision of information states. He takes the principal mandate of proper inquiry to be to avoid importing error into one’s state of information. Since refusing to alter one’s information state guarantees that no additional error will be imported, the maximally conservative inquirer is perfectly rational by the lights of Levi’s account. An information state on this model is merely a set of worlds, representing the space of genuine (open) possibilities according to the inquiring agent.

• Abduction

But we aren’t satisfied to remain cognitively stationary. In C. S. Peirce’s (Fixation of belief) words, we are constantly subject to the irritation of doubt. We can incorporate this impetus to revise our information state into a formal account of inquiry by way of William James’ (Will to believe) dual inquiry mandate: avoid error and acquire truth. The impetus to revise stems from the second goal, and the rational inquirer strives for truth in addition to avoiding error.

To model this extension of the nature of inquiry, we add additional structure to the information states that are the objects of inquiry. We partition the space of possibilities based on the issue the agent is striving to resolve.

• Impetus to revise
• Will to believe
• Induction

An inquiry is not simple transition from one state to the next. It is a dilated process, developing over the course of a number of steps, some of which, perhaps, involve false progress and require the inquirer to backtrack. The process of having reached a dead end in the inquiry often provides valuable information to the inquirer.

Plans require bookkeeping both in the form of preparing multiple lines of progress the inquiry might entertain as it runs its course as well as a record of past steps taken. To account for this, we must expand the information state structure again, modeling this bookkeeping operation.

John Stuart Mill (On Liberty) maintained that the ability to defend one’s beliefs against detractors is an essential element of the justification of those beliefs. As the principal justifying process of belief, inquiry thus carries the commitment to engage the best arguments one can find in pursuit of issue resolution. The result is that proper inquiry is essentially a joint activity – one in which dispute plays a progressive role.

To model joint, disputative inquiry, we need a staging ground upon which attempted information state revisions can be proposed and vetted. I call this platorm a sandbox, and it is essentially a mini-information state embedded within the primary one. It allows for extended investigation of issues without potentially infecting the primary information state with false tries.

• Coherence
• Correction
• Implementation

We end up with a picture of inquiry as the productive evolution of an articulated information state. Productive evolution can take a variety of forms. These forms correspond to different types of contribution. Inquiry contribution types have direct analogues in discourse. There are possibility restricting assertions and state partitioning questions with which we are familiar. But there is also a contribution type corresponding to conjecture, which can be modeled by sandbox-level investigations. Conjecture, along with its natural pair correction, provides a distinctive means of advancing the inquiry.

3. What do we know?
• Solitary knowing

Those who do not belong to a community of inquirers have no one against whom to test their beliefs, but surely they are capable of possessing knowledge. Proper inquiry is doubly qualified to account for solitary knowledge: (i) an inquirer need only possess the ability to defend against challenges, and (ii) she need only defend against the best arguments available. Degenerative inquiry, in which the best available arguments are trivially defended, is still inquiry, its outputs still knowledge.

• Knowing the truth

Everything known is true, but the inquiry-first account posits complex objects that are not identical to truths. Fortunately, the factivity of knowledge is secured by any object that entails truth, as proper inquiry does. Unfortunately, proper inquiry does so only in the long run. But intermediate-stage inquiry still produces truth, in context.

4. The value of knowledge
• Belief stability
• Success and credit
• Justifying processes

Linda Zagzebski argues that many theories of knowledge – notably machine-product accounts of knowledge, on which knowledge is taken to be a state that is the output of some state generating process – are subject to a swamping problem. Possessing knowledge is more valuable than possessing merely true belief, but these accounts of knowledge have no explanation for the distinctive value that knowledge provides over and above the truth of output they produce. But the problem isn’t with the machine-product model itself. The problem comes from focusing on the product and ignoring the process. The value of knwoledge is in grounding the credit the agent deserves for the belief. Knowledge provides, as Socrates suggested to Meno, an account of the product’s genesis. So long as the process is seen as external to the product, the product will carry no value of its own. But the insight of the connection between inquiry and knowledge is that they need not be independent. The product encodes the process. The inquiry itself is the object of knowledge.

5. Tracing back through the breadcrumbs

Collaborative discourse dynamics

1. Scoreboards and best systems
• Constructing an interpretation

David Lewis’ conception of the conversational scoreboard is grounded in his theory of radical interpretation. The scoreboard represents our theory and its bells and pulleys represent the theoretical posits and regularities. The structure we give the scoreboard depends on what serves as the most simple and robust representation of our shared linguistic knowledge, relative to domain specific constraints.

• Generativity and usability

One domain specific constraint on theory building is that inguistic interpretations be finitely specifiable. But this only makes sense as a constraint on language structure under the presumption that the theory is something possessed by human agents. And if that is our motivation, we need a stiffer constraint.

• Anaphora and coherence

In Index, Context, and Content, Lewis contends that an adequate characterization of our shared linguistic knowledge must provide us with truth values for sentences to account for our practice of using language to impart information. But this isn’t all we do with language! We use it to resolve issues in the course of inquiry. And to capture this, our characterization must provide us with a means of resolving anaphora and determining coherence structure.

• Honesty and trust

In Language and Languages, Lewis suggests that linguistic communities abide by a convention of truth and trust, and that this is what secures the link between what they say and what they mean. But communities that use language to realize joint inquiry require a more integrative convention – one in which speakers are honest and both speakers and hearers trust in the contributions of their interlocutors.

2. General update semantics
• Discourse referents and conditions

Standard update semantics divides information added to the conversational platform into two categories: referential information – functioning primarily to populate the universe of the discourse – and propositional information – placing constraints on the set of models that represents the discourse.

• Content at different dimensions

But there is room to further articulate information states to capture more refined distinctions between the functions of additions to the conversational platform. Sarah Murray and Anderbois, Brasoveanu, and Farkas have developed accounts to capture the difference between the at-issue content and not-at-issue content expressed by utterances. But these proposals hit a wall when we attempt to apply them to extended discourse.

3. Collaborative contributions
• Discourse pairs

Herb Clark argues that conversational contributions can be decomposed into an initiation phase and a completion phase, and that the completion is generally provided by someone other than the initiator. Contributions to discourse require not just cooperation, but collaboration.

• Accuracy and precision

Standard discourse pairs are the assertion-assessment pair and the question-answer pair. These pairs earn their keep in virtue of the distinctive role they play in advancing inquiry. They can also be categorized within a two-dimensional concept space in terms of the accuracy and precision to which they commit the contributors.

• Conjecture and correction

Conjecture-correction is another discourse pair that earns its keep as a distinctive contribution to discourse. Conjecture differs from both assertion and interrogation in terms of its position on the accuracy/precision spectrum.

4. Collaborative discourse dynamics
• Sandbox discourse

Discourse pairs are the basic evaluative units discourse, but utterances aren’t made in paralell. To address this worry, we need to introduce a staging area into the information state – a place where information expressed by utterances can be held and manipulated without committing it to the primary information state. I call this staging area a sandbox.

• Checkout, commit, merge

Collaborative update involves manipulation of information within a sandbox. In order to model collaborative activity of this sort, I incorporate concepts from version control in computer programming. Initiating utterances open up a new sandbox and populate it with needed information from the primary information state; this is known as a checkout. Continuation utterances add to sandbox and subject its contents to manipulation; these procedures involve a variety of commits. Finally, a completion utterance merges the finished sandbox product back into the primary information state.

• Defaults and content evolution

It isn’t strictly true that individual utterances are inevaluable in the absence of their complement utterance. Instead, context frequently determines a default saturation to allow for initial evaluation of the utterance. Defaults can be overridden by explicit completion utterances. This suggests that what a speaker expresses can change as the discourse progresses. Meaning is, in a very real sense, out of their hands, determined by the way in which their interlocutors choose to make use of their utterance.

5. Contextualism and relativism
• Disagreement as linguistic data

A now common line of argument in the philosophy of language starts by pointing to an intuition that parties to a particular kind of discourse are engaged in a genuine disagreement. From there it is claimed that explaining this intuition requires a semantic theory that represents their utterances as expressing conflicting semantic contents. From here it is maintained that only one of contextualist or relativist accounts of the semantic value of certain expressions provides the requisite content.

Plunkett and Sundell have argued that the restriction of disagreement securing content to specifically semantic content is artificial. There is no principled reason that pragmatically generated content can’t serve to ground intuitions of disagreement. And perhaps the disparaged semantic account can revive its status by hitching its wagon to such a pragmatic story.

• Commitment and contribution

Our discussion of the structure of inquiry suggests that the restriction of disagreement securing contributions to assertive contributions is itself unmotivated. Dispute is a central element of healthy inquiry, and as such it can potentially be grounded in any number of types of contribution to inquiry.

• The nature of relativism

John MacFarlane characterizes as relativistic any semantic theory that makes essential reference to the context of the assessor of an utterance. And much of the contextualism/relativism debate has centered on who’s context matters. But Peter Lasersohn provided an interestingly different take on relativism. His support of relativism was not so much that certain expressions demand that the assessor’s context be included but that whose context is important to evaluation is not determined at the point of utterance.

• Joint inquiry and relativism

Joint inquiry, modeled by sandbox-level investigation, suggests exactly this. An individuals utterance is incomplete in an important way, and it calls for completion from one’s interlocutors. Thus, evaluation is delayed until the inquiry progresses. Knowledge, it turns out, is relative; just not in the way we first suspected.

Dispute in discourse

1. Disputative discourse

Disputative discourse is rich with interpretive detail. I present a number of observations about a short dialogue involving disputative discourse. I take accounting for these observations to be a criterion of adequacy for an account of the discourse dynamics involved.

2. Correction as contrastive topic
• Focus and alternatives

A salient feature of our example dialogue is the presence of focus in the response. An attempt to capture the observations by applying an account of focus interpretation is elaborated. Mats Rooth’s alternative semantics for focus is introduced.

• Questions under discussion

Craige Roberts’ QUD model of discourse representation is introduced. Daniel Büring’s extension of QUD to account for contrastive topic is explained.

• Advancement or accommodation

A question is raised as to the amount of reliance the QUD account places on accommodation.
3. Denial as downdate
• Denial and LDRT

Geurts and Maier’s Layered discourse representation theory is introduced. It is explained how this account avoids swamping worries for denial.

• Denial and information state revision

Spenader and Maier extend LDRT with a downdate operation to account for the effect of contrast in discourse.

• Binding retractions

While LDRT can avoid the binding problem that plagued Karttunen and Peters, it can’t get around all interaction effects between kinds of content. The susceptibility of the denial as downdate account to this extended class of binding-style worries is due to a mis-diagnosis of the nature of interaction between content types.

4. Contrast as collaboration
• Strictness, shifts, and structure

Both counterfactual conditionals and bare plural sentences have simple, strict analyses in terms of set-inclusion. But both face sticky situations that threaten to undermine the simple analysis. Each can be extracted from its sticky situation by better understanding the pragmatic principles in play.

• Pragmatics of plurals

Bare plural sentences carry a presupposition to the effect that the subject class is homogeneous with respect to the predicate. This is fairly well accepted; the novel bit is a representation of this presupposition as imposing a partition on the predicate class against which the predication is evaluated.

• Plural predication and contrast, informally

An informal characterization of how the information state evolves through the course of our example dialogue is presented.

• Plural predication and contrast, formally

First, an explication of the general collaborative update processes involved in the discourse is provided. This framework is then extended with collaborative update operations that are specific to bare plurals and contrastive focus. The story about plurals is couched in the language of relational algebra.

5. Outcome and observations obtained

The collaborative update account exctracts a strict analysis of bare plurals from its sticky situation. It also provides a satisfying account of the observations about disputative discourse.