User guide

How to run a debate analysis

DebateLens takes a televised debate and applies the framework to it automatically. You supply the recording; it returns a draft transcript, coded observations across five domains, gaze and repetition measurements, and — once those records are reviewed — a comparative interpretation of the corpus. The comparative interpretation reports which participant's discourse is denser in the framework evidence counted for this corpus. It is not a statement of who won the debate, a ranking of candidates, or a prediction of an election. This page walks through the whole path, from account to report.

1. 1

   Create your account

   Choose Sign up in the top bar. Enter your name, email, and a password of at least eight characters, or continue with Google.

   • If email confirmation is required, open the link sent to your inbox before signing in.
   • Everything you create — studies, recordings, transcripts, observations — is private to your account.
2. 2

   Start a study

   Go to Studies and create one study per debate. Give it the debate title, the date, and the broadcaster if you know them.

   • A study is the container for one debate: its recording, its transcript, its coded observations, and its report.
   • Add the participants (for example, the two candidates and the moderator) so turns can be attributed correctly.
   • Set the language of the material in the Overview tab before you run the analysis. The coding then reads the debate within its own rhetorical tradition — honorifics, forms of address, register and dialect shifts, culturally specific idiom — and keeps every quotation in the original language.
3. 3

   Add the material

   Open the Automatic analysis tab and supply whatever you have: a video file, an audio file, or an existing transcript.

   • Video gives the fullest analysis, because eye-gaze and speech state can only be read from the picture.
   • Audio alone still yields transcript, turns, and all verbal domains.
   • A transcript alone yields the verbal domains without paralinguistics.
   • Uploads go to private storage; only you can open them.
4. 4

   Run the analysis

   Press Start analysis. DebateLens works through the stages in order and shows progress as it goes.

   • Transcription and speaker separation, then turn-by-turn segmentation.
   • Coding across the five framework domains: cohesion and coherence, pragmatics, paralinguistics, ideology and bipartisanship, culturemic and pragmemic.
   • Instruments: the repetition matrix, Gricean maxim breaches, lexical fields.
   • Eye-gaze and speech state read across the whole recording, first second to last.
   • The run is resumable. If it stops, press Resume and it continues from the last completed stage rather than starting again.
5. 5

   Inspect the evidence

   Use the workspace tabs to check the material behind every figure before you rely on it.

   • Transcript — the turns as segmented, editable if a speaker label needs correcting.
   • Observations — each coded record with its observable evidence kept separate from its interpretation.
   • Repetition — the matrix computed from the links between turns.
   • Lexical trio — the LCT panel: reference and transitional signals, repetition links, and academic lexical phrases per participant, plus the Hoey ranking that flags the debate's central utterances and a full turn-by-turn link matrix.
   • Gaze — time-based q-gaze and a-gaze totals per speaker, with denominators and coverage shown.
   • Pragmemic matrix — every pragmatic act laid out against the turn sequence, so you can see how illocutionary force shifts across the debate; click any cell to reach the verbatim evidence behind it.
6. 6

   Test the reading before you publish it

   Three tabs exist so a reviewer's objections can be answered with figures rather than assurances.

   • Reliability — draws a reproducible slice of the corpus with a seeded shuffle and lets you re-code it blind, then reports Cohen's kappa between the two readings for domain and category.
   • Coverage — how much of the recording carries evidence, which turns carry none, which domains are sparse, and exactly what was excluded and why.
   • Sensitivity — the same comparative reading recomputed under different settings (bond cut-off 2 to 5, qualified records in or out, gaze in or out), so you can see whether the leading participant survives a change of assumptions.
   • Validation — a cross-study page comparing each study's reading with the independently documented result of that debate, with its source and date.
7. 7

   Read the comparative interpretation

   The comparative interpretation reports which participant's discourse is denser in the framework evidence counted for this corpus. It is not a statement of who won the debate, a ranking of candidates, or a prediction of an election.

   • The tab shows the figure with its confidence, its coverage, and the reasoning that produced it.
   • The panel lists the evidence behind the figure, the confidence level, the caveats, and any gaps in the data.
   • A narrow margin is flagged as narrow. Treat it as such.
8. 8

   Take away the report

   The Report tab assembles the whole study: framework findings, instruments, gaze proportions, and the projection, each anchored to the transcript.

   • The Export tab produces the material you need for publication: comma-separated matrices for SPSS or R, the full dataset as a structured file, scalable vector figures of the gaze and repetition matrices, and a pre-formatted analytical appendix in APA or MLA style.
   • Every export is built from counted records only, so an exported table matches the figures shown on screen.