December 04, 2007

Modal semantics in National Intelligence Estimate

The page I've copied below the cut (page 5 from the summary National Intelligence Estimate, Iran: Nuclear Intentions and Capabilities , as reproduced in NYTimes Dec 3, 2007) struck me as remarkable in showing how important modal semantics is in the real world -- the authors of the summary used a full page out of their 9 total to give an exegesis on their use of modal adverbials and modal verbs and on the distinction between circumstantial modals ("assessed likelihood") and epistemic modals ("level of confidence we ascribe to the judgment"). Given the historical-political context, it's clear that these modals really matter. The authors are not linguists, but it seems to me that they've done a very good job of explaining the language and the concepts behind it in clear non-technical terms - I doubt that a linguist could have done better.

What We Mean When We Say: An Explanation of Estimative Language

We use phrases such as we judge, we assess, and we estimate -- and probabilistic terms such as probably and likely -- to convey analytical assessments and judgments. Such statements are not facts, proof, or knowledge. These assessments and judgments generally are based on collected information, which often is incomplete or fragmentary. Some assessments are built on previous judgments. In all cases, assessments and judgments are not intended to imply that we have "proof" that shows something to be a fact or that definitively links two items or issues.

In addition to conveying judgments rather than certainty, our estimative language also often conveys 1) our assessed likelihood or probability of an event; and 2) the level of confidence we ascribe to the judgment.

Estimates of Likelihood. Because analytical judgments are not certain, we use probabilistic language to reflect the Community's estimates of the likelihood of developments or events. Terms such as probably, likely, very likely, or almost certainly indicate a greater than even chance. The terms unlikely and remote indicate a less then even chance that an event will occur; they do not imply that an event will not occur. Terms such as might or may reflect situations in which we are unable to assess the likelihood, generally because relevant information is unavailable, sketchy, or fragmented. Terms such as we cannot dismiss, we cannot rule out, or we cannot discount reflect an unlikely, improbable, or remote event whose consequences are such that it warrants mentioning. The chart provides a rough idea of the relationship of some of these terms to each other.

Remote < Very unlikely < Unlikely < Even chance < Probably (Likely) < Very likely < Almost certainly

Confidence in Assessments. Our assessments and estimates are supported by information that varies in scope, quality and sourcing. Consequently, we ascribe high, moderate, or low levels of confidence to our assessments, as follows:

High confidence generally indicates that our judgments are based on high-quality information, and/or that the nature of the issue makes it possible to render a solid judgment. A "high confidence" judgment is not a fact or a certainty, however, and such judgments still carry a risk of being wrong.

Moderate confidence generally means that the information is credibly sourced and plausible but not of sufficient quality or corroborated sufficiently to warrant a higher level of confidence.

Low confidence generally means that the information's credibility and/or plausibility is questionable, or that the information is too fragmented or poorly corroborated to make solid analytic inferences, or that we have significant concerns or problems with the sources.

Update from Kai von Fintel:

This kind of exercise in determining a graded list of expressions of probability/confidence is not uncommon among professionals who need to give such assessments. There is an interesting article in the February 1990 issue of the journal "Statistical Science" with peer commentary by Herb Clark among others. (The link works only if you have access to JSTOR).

Other articles in this vein here, here, and here.

I'm sure there's plenty more where these came from. These are just what I came across a while back (I don't remember how I got there -- probably had to do with Herb Clark somehow).

Oh, and while we're at it, here's a legal take on the meaning of 'must'.

Update from Roger Shuy:

I encountered the somewhat similar attempt to quantify uncertainty a while back. In 1990 Mosteller and Yountz wrote an article in Statistical Science, offering their word definitions about such things. If you're interested, you can find it linked in my LL post here .

Posted by Barbara Partee at December 4, 2007 02:16 AM