Semantic resources project/Use Cases/Meeting Notes/SWAN/Meeting 02032010

PC and TD interviewed GW by Skype at 11am 02/03/2010

= The SWAN Annotation Process =

and discussion, and from the research findings. She tries to capture the citations for these claims, but she doesn't take them all; modulo materials and methods sections. swan:hasPathogenicNarrativeQualifier swan:hasEvidencetypeQualifier swan:researchStatementQualifier
 * GW: She's faced with phrasing that contains lots of qualifications or conditions; these need to be combined in some way.
 * I try to adhere to one subject-one verb-one object, but scientists don’t write that way.
 * TD: Describe your annotation process. How do you decide between calling it a comment, a question or a statement (claim or hypothesis)?  Comments correspond to web-comments? Do you ever choose to call a comment something that didn't come from the web?  How do you choose between claim and hypothesis?
 * Comment: There are two sources of comments: those submitted directly to SWAN, which may be mediated or introduced or aroused by the SWAN editors, and in the future they will solicit comments on *all* hypotheses. They can also be curated directly from AlzForum. [Web comments are previously published on a website, so a link is made.]
 * Hypothesis: an adaptation of the title of the article itself, usually. [titles turned into statements]
 * Claim: generally, the big findings, added to the key introductory background findings. -- from the intro
 * Question: Often the question "at the end of the article," often setting up the next article, or a difficult question.
 * GW: Questions also from comments.
 * TD: Where do the title and description come from?
 * GW: In the hypotheses we curate from articles (90% of them) -- almost all of the descriptions start out with "This hypothesis suggests..." along with a statement of difference, backed up with key describing sentences. "The take home message," along with a description of a research method. But the description is not culled from a *single* location within the article.
 * TD: How do you determine evidence type or pathogenic narrative qualifiers?
 * Judgement-based description.

= Containment of Discourse Elements =

(2) Paolo described a process of annotating *parts* of a paper, as well as the entire statement. Describe how discourse elements *contain* other elements. (Can there ever be a cycle?)

(Paolo says: no cycles should be present.) [But I’d like one cycle, where we can say a Research Question is motivated by… and then link possible claims or hypotheses that address the question, not necessarily “In response to” but could be “Addressed by” relationship.]

Hypothesis on top, and then claims inside? She wants to represent the "bundled" set of claims. "Containment" is a logical relation between claims, *not* a physical relation. "Summary statements of data that support a framework claim." Every piece of claim 1 is either a piece of claim 2 or a piece of claim 3 -- this means that 1 contains 2 and 3. [This is incorrect – Claim 1 is the summary statement, Claims 2 and 3 are specific examples of the general veracity of claim 1, so you have it the other way around.]

"In a pubmed abstract, you've got multiple sentences, and each one is a very general statement. Sometimes that statement is a claim in a hypothesis, but in and of itself it's not that well supported.  But what the specific findings are in the paper that are annotated in the claim are actually evidence for the claim."

When claims are *ordered*, that reflects a *logical* structure -- from general to specific. "Every figure in a paper, I want to capture a claim."

These are *never* put on claims/hypotheses from the *same* paper. They are trying to capture discourse from multiple members of the research community. "We don't draw any relations between claims of the same hypothesis."

= Annotation of Discourse Elements =

(3) Which relations are you assigning between elements to reflect the author's intent? Which ones reflect your judgement?

swan:inResponseTo -- comments are "in response to" a hypothesis. swan:motivatedBy -- a special relationship we use for questions, which are motivated by a set of findings. Either the questions came from the paper itself, or from a separate author.

swan:discusses swan:alternativeTo (these are now called "consistent/inconsistent?") swan:refutes → Inconsistent swan:supports → Consistent

There's a question of whether they use "discusses" without knowing what it means, too vague, without a "clear logic flow." Discusses has been partly abandoned. Gwen wants to go back and replace "discusses" with either inconsistent, consistent, or alternative-to.

Look at the SWAN tutorial. View Slide # 17. Flowchart -- from a claim of Author A to a claim or hypothesis of Author B. "Finding" is data of author B.  "Interpretation" is the claim of author B. If the data was wrong but the conclusion was right, or if the data was good but the conclusion was wrong, we put it into AlternativeTo.

TODO : send gwen a list of claims that still use the discusses relationship.

= Citations =


 * Third set of citations: meeting abstracts. Inserted hyperlinks in text -- in the swan:description.
 * GW: See Morse hypothesis or Stutzman comment for examples.
 * How do you determine, in general, whether something is cited? Is it only cited in the "region" of the paper/comment that you quote? Or can it be mentioned elsewhere?
 * (Paolo: Both, it is arbitrary and people can add citations after that can also not be part of the paper. We discussed for a long time this point and gwen decided to model the second case as a comment listing the new citation. We have been working on an evolution model for distinguishing what was said in the original paper and what after. I pushed a lot for it but at the beginning they were not feeling the need when they recognized the need was the end of the grant.)


 * If A contains B, and B cites C, then (can I assume that) A cites C?
 * (Paolo: there is no precise rule for that.)
 * GW: yes, this is true, but won't be explicitly added. This is always true.  But we don’t pin all citations to the hypothesis title.
 * PC: Except if the claim is a strawman, and then we can't infer the super-citation.
 * So, no:
 * we can't infer the citation supporting/refuting polarity from containment, because a strawman isn't given an explicit relationship with the other sub-claims.
 * GW: Strawman argument example: cited evidence listed as “Discussing evidence” not Supportive Evidence.

LSEs
swan:citesLifeScienceEntity
 * What qualifies as an LSE citation? When is a protein, or a gene, mentioned? How are organisms used?


 * Transferring citations across organisms is a huge and difficult problem -- figuring out how to not partition by species would be a huge help.
 * Right now, Gwen is often flagging *both* human and mouse versions of the same protein where applicable. (Can we automatically infer groups from this?)

Papers
swan:derivedFrom
 * Are you copying citation information from the paper itself? Or looking it up on your own?
 * (Paolo: both.)
 * GW: They only add citation information that wasn't in the original paper with the explicit guidance of the author -- and this "happens all the time." [It happens all the time with comments and online submitted hypotheses.  It happens occasionally through author feedback in review process.]
 * GW: If a hypothesis is derived from a specific article, with a PMID.


 * Is it only explicit citations? Or do you ever "infer" a citation?
 * (Paolo: both.)


 * How do you determine the distinction between "Supporting" "Refuting" and "Discussing" evidence?

swan:citesAsDiscussingEvidence


 * "Papers, written by other people, talking about 'alternative' hypotheses, that are really just slight modifications." [They can be slight modifications, but the main point is that this occurs when an author presents a strawman argument.]

swan:citesAsRefutingEvidence swan:citesAsSupportingEvidence


 * Do these relations interact at all with the "consistent" (supports) or "inconsistent" (refutes) relations between elements themselves? [Not sure which elements you are referring to here.]