[Feature] Alignment over longer spans to assess priming decay

[fusaroli]

Several papers have pointed to differential patterns and effects of alignment over longer spans:

• Reitter and Moore 2017: syntactic alignment at 20 turns of distance (but not shorter) predicts performance
• children with ASD show similar patterns of immediate lexical priming to TD children, though they show different patterns over longer timecourses (Harper-Hill, Copland, & Arnott, 2014; Henderson, Clarke, & Snowling, 2011)

We should implement the possibility to measure alignment over longer spans (not adjacent turns, but at a distance of n). This would also require measuring self-alignment at increasing spans to control for that.

[fusaroli]

The fix could look like adding a loop around the TurnByTurnAnalysis function looping from delay 1 to delay n, and adding a delay column to the output.
As far as I can see there is no check as to whether it's self or other alignment that is being measured, but the direction column is keeping track of that (e.g. A > B vs. A>A).
The interest of adding self-alignment is that it allows us to control whether the temporal changes in alignment to the interlocutor over longer delays are confounded by the interlocutor repeating themselves.
Not implementing this yet, so we can keep the package as-is during the review process

[jseale-asapp]

The way delay is implemented now, is it the case that, if delay is say, 20, then the alignment will be measured between the prime and the one target that is exactly 20 turns later?

[jseale-asapp]

I'm seeing different patterns in comprehension-production and production-production priming in my data, so a self-alignment option would be great.

[fusaroli]

Yes, exactly that

On Tue, 13 Aug 2019 at 16.41, jseale-asapp ***@***.***> wrote: The way delay is implemented now, is it the case that, if delay is say, 20, then the alignment will be measured between the prime and the one target that is exactly 20 turns later? — You are receiving this because you were assigned. Reply to this email directly, view it on GitHub <#24?email_source=notifications&email_token=ABOTIJJ4QNLCKT7RHKP2CQ3QELBZXA5CNFSM4E7GBWFKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD4F4CWQ#issuecomment-520864090>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABOTIJNXOKVA7E3EIXM4XBLQELBZXANCNFSM4E7GBWFA> .

--

…

-- Riccardo Fusaroli Associate Professor in Cognitive Science Interacting Minds Centre & School of Communication and Culture Aarhus University

[jseale-asapp]

Would it be of interest to create a window of turns from the delay-indicated turn to, say, the end of the dialogue?

There may be some short-term priming effects within the window that would need to be taken into consideration. For example, using 20 as the delay, if a prime construction is repeated in turn 21 and turn 22, then the probability that is repeated in turn 22 is bolstered by its presence in 21. However, it seems as if the larger window may help to detect the presence of longer term priming—perhaps if the prime-target pairs are restricted to one target per prime within the window, instead of allowing for multiple targets, that would give something of a binary indication that the priming is occurring.

Yes, exactly that
On Tue, 13 Aug 2019 at 16.41, jseale-asapp @.***> wrote: The way delay is implemented now, is it the case that, if delay is say, 20, then the alignment will be measured between the prime and the one target that is exactly 20 turns later? — You are receiving this because you were assigned. Reply to this email directly, view it on GitHub <#24?email_source=notifications&email_token=ABOTIJJ4QNLCKT7RHKP2CQ3QELBZXA5CNFSM4E7GBWFKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD4F4CWQ#issuecomment-520864090>, or mute the thread https://github.com/notifications/unsubscribe-auth/ABOTIJNXOKVA7E3EIXM4XBLQELBZXANCNFSM4E7GBWFA .

…
-- Riccardo Fusaroli Associate Professor in Cognitive Science Interacting Minds Centre & School of Communication and Culture Aarhus University