Skip to main content

5. Working with advisors and co-authors

A first *ACL paper is almost never solo work. You will have an advisor, and usually one or more co-authors, and the human side of that — who does what, whose name goes where, how to give and take feedback, how to disagree — is rarely taught and quietly decides whether the project is a good experience or a miserable one. The science is only half the job. This chapter is the other half.

Who earns a place on the byline​

Authorship is not a reward for seniority, funding, or friendship. The widely-used standard, from the ICMJE, is four conditions, all required: a substantial contribution to the work; drafting it or revising it critically; approving the final version; and agreeing to be accountable for it. By that bar, someone who only secured the grant, ran the lab, or proofread the English does not qualify — such people "should not be listed as authors, but they should be acknowledged." A language model, likewise, is never an author (Chapter 21): it cannot be accountable for anything.

Two failure modes have names (Chapter 3): gift (or guest) authorship — adding someone who did not contribute, to flatter them or look impressive — and ghost authorship — omitting someone who did. Both are misconduct. Note that in NLP the advisor is normally a genuine author: they shaped the idea and revised the paper, which clears the bar. The principle is that they are an author by contribution, not by default.

Author order, and what it signals​

NLP and ML read author order as a signal, so it is worth getting right:

  • First author — the primary contributor, usually the student who did most of the work and writing.
  • Last author — the senior researcher who supervised: your advisor or PI.
  • Middle authors — meaningful contributions in between, roughly in decreasing order.
  • Co-first authorship — when two people contributed equally, both are marked first with a footnote ("* Equal contribution"). Legitimate and common; use it rather than forcing a false ranking.

Conventions vary — a few communities order strictly alphabetically — so ask your advisor what your group does. The one rule that holds everywhere: decide order early, out loud, and revisit it if contributions shift. Disputes over authorship order are common, bitter, and almost entirely preventable by having the conversation before submission rather than on deadline day.

Make contributions explicit​

Many venues now ask for an author-contribution statement, and the standard vocabulary is CRediT, an ANSI/NISO taxonomy of 14 roles — conceptualization, methodology, software, data curation, validation, investigation, writing – original draft, writing – review & editing, supervision, and so on. Even when a venue does not require it, writing a CRediT-style statement early in the project is the cheapest insurance against a later credit dispute: it forces everyone to say who is doing what while there is still time to rebalance the load.

Working with your advisor​

Advisors vary enormously — some want a meeting a week and a draft a fortnight out, some you will see twice a semester. Diagnose yours and adapt rather than resent.

  • Ask how they work. Meeting frequency, how much detail they want, when they expect drafts. Guessing wastes both your time and theirs.
  • Come prepared. Bring a specific question, a result, or a decision to be made — not "what should I do next?" A meeting is for the things you cannot resolve alone.
  • Manage up on time. Your advisor is busy; a draft sent the night before the deadline gets a rushed read and worse feedback. Send it with real lead time, and say what kind of feedback you want.
  • Disagree with evidence. "I tried X; it failed because Y; I propose Z" is a contribution, and a good advisor welcomes it. Silent compliance and silent defiance are both worse than a reasoned case.

Giving and taking feedback​

Feedback on your writing is about the writing, not about you — separate your ego from the draft and the whole process gets easier.

  • Address every comment. Fix it, or push back with a reason; never silently ignore a co-author's note.
  • A harsh internal review is a gift. Far better that a co-author finds the hole now than Reviewer 2 finds it later. Invite the toughest reading you can get before submission, not after.
  • When you are the one giving feedback, be specific and kind — the same standard the reviewing chapter asks of reviewers. These are your collaborators, not your adversaries.

Co-writing logistics​

  • Keep one source of truth. Agree on Overleaf, or git-plus-LaTeX, and stop emailing versions around before someone overwrites someone else's edits.
  • Divide by expertise, unify by one voice. Let people draft the sections they know best, but have a single author own the final pass so the paper reads in one voice (Chapter 6) rather than as a patchwork of five styles.
  • Share the bibliography. One .bib file, one reference manager, and every entry verified (Chapter 22).
  • Write down who owns what, with dates. A shared task list prevents the deadline-week discovery that nobody wrote the related work.

When things go wrong​

  • Authorship disputes: raise them early and in person, with the contributions statement as your evidence. If you cannot resolve it between you, bring in a neutral senior person before submission, not after.
  • An unresponsive collaborator: set explicit deadlines and follow up; if a co-author goes dark near a deadline, tell your advisor early rather than absorbing their work in silence.
  • Serious cases — credit theft, coercion, or misconduct — are exactly what the ACL Policy on Publication Ethics and its ethics committee exist for. You are allowed to use them.

Common mistakes​

  • Leaving the authorship and order conversation until submission day. The single most preventable source of conflict in a research group.
  • Gift authorship to flatter a senior, or ghosting a junior who did real work. Both are misconduct, not generosity or oversight.
  • The midnight draft. Sending your advisor the paper hours before the deadline and expecting a careful read.
  • Taking writing feedback personally, or ignoring comments without a word of reply.
  • No single owner for the final voice, producing a paper visibly stitched together from different hands.
  • Treating an internal critique as an attack instead of the cheapest, kindest review you will ever get.

Further reading​