Skip to main content

11. The Title

Most of the field will read your title and nothing else. It appears in the proceedings, in citations, in a reviewer's bidding list, in search results, in tweets, and on slides other people make about your work. You will write it in ninety seconds and live with it for ten years.

What a good *ACL title does​

  • It names the contribution. "Neural Machine Translation" names a topic that a thousand papers share. Attention Is All You Need is a claim only one paper can make.
  • It is searchable. Reviewers, future students and retrieval systems search by keyword. A paper on retrieval-augmented generation for biomedical QA should carry those words, or close variants, in its title. Nobody cites a paper they cannot find.
  • It is honest. Do not promise Universal or General-Purpose if you tested one language and two datasets. Reviewers see through it by the abstract. The claim ages badly once someone runs the obvious counterexample.
  • It survives being read aloud. People say titles aloud in talks, in introductions and in hallway recommendations. If you stumble reading yours, rewrite it.
  • It sets expectations, and the paper meets them. The title makes a promise. You keep it in paragraph 4 of your introduction, the results preview (see the Introduction chapter).

The dominant patterns​

A handful of shapes cover most *ACL titles. None is mandatory. Each fits a kind of paper.

PatternExampleWhen it fits
Declarative claimAttention Is All You NeedYou have one clean claim the paper delivers
Name + colon expansionBERT: Pre-training of Deep Bidirectional Transformers…A reusable system, model, or dataset others will cite by name
Resource + scopeSQuAD: 100,000+ Questions for Machine Comprehension of TextA dataset or benchmark; the scope numbers are the selling point
DescriptiveNeural Machine Translation by Jointly Learning to Align and TranslateThe work is foundational and the plain description is already interesting
QuestionWhy are Sensitive Functions Hard for Transformers?The paper investigates a question and ends with an answer
Hook + payoffWith Little Power Comes Great ResponsibilityThe hook is memorable and the second half says the real topic
ImperativeDon't Stop PretrainingA finding sharp enough to phrase as advice

The phrasebook below gives real, highly cited titles for each shape.

Phrasebook: titles by pattern​

Each subsection collects real *ACL titles, plus a few field-defining ones from nearby venues, grouped by surface shape. Each entry names the move, what the title does, so you can copy the move without copying the words.

Study the shapes; do not fill them in. Quoting a title carries no plagiarism risk, unlike the abstract and introduction phrasebooks (Chapter 12, Chapter 13). A title is its citation, so quoting it in full names the paper. The risk here is cliché. These shapes are so worn that copying one without thought produces the ten-thousandth X: A Novel Framework for Y. Read a subsection, name the move ("name-and-expand," "hook that still tells the topic," "the honest question"), then write your own. A paper can fit several patterns; we file each title under the shape it shows most clearly. The larger patterns show two titles and fold the rest behind Show more. The shorter ones list every title.

Name + colon expansion (the workhorse)​

This is the most common shape in modern *ACL: a short, pronounceable name, a colon, then a precise expansion. Use it when you ship a system, model, or method others will cite by name. The name must be sayable, and the expansion must be honest about scope.

Show 11 more

Resource and benchmark titles (name + scope)​

This variant of the name-and-expand pattern suits datasets, benchmarks, and shared-task resources. The expansion sells the scope: how many examples, how many languages, what task. Put the number in the title. It is the most reused fact about the resource.

Show 5 more

The declarative claim​

The title is a full sentence that asserts something. It is memorable and quotable, but the paper must earn the claim or reviewers will use the title against you. Reserve it for one clean, defensible result.

The descriptive workhorse​

The plain title says what the paper did. Most *ACL papers use it, and most should. It is never wrong and always searchable. Make it your default when the contribution is solid but does not reduce to one quotable claim. The craft lies in choosing the right nouns and verbs: precise, keyword-rich, no filler.

Show 12 more

The question title​

Use a question only when the paper investigates and answers it, and only when readers want the answer too. A question with an obvious answer reads as filler. An unanswered one reads as a blog post.

Hook + payoff (two-part, colon-joined)​

A memorable first half, a colon, then a half that names the topic. Authors love this shape and abuse it. The part after the colon must let a stranger find and understand the paper without the hook. Delete the cute half and the title should still work. If deleting the informative half leaves the title meaningless, the hook is doing too much.

Show 6 more

The imperative​

An imperative title gives advice. It works only when the finding is sharp enough to be advice and the paper has earned the right to tell the reader what to do.

The "On the …" framing​

The "On the …" opening signals a position paper, an analysis, or a survey: a paper about a question, with no system that solves it. The tone is formal and essayistic. Use it when you contribute an argument or a measurement instead of an artifact.

  • On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? The "On the Dangers of" framing announces a position paper, and the metaphor ("stochastic parrots") became field vocabulary. It is not *ACL (FAccT 2021), but study it for how a title can name a concept the field then adopts. (Bender, Gebru, McMillan-Major & Mitchell, FAccT 2021)
  • On the Cross-lingual Transferability of Monolingual Representations: "On the … Transferability of …" frames the paper as the study of a property. It proposes no new model. (Artetxe, Ruder & Yogatama, ACL 2020)
  • Universal Adversarial Triggers for Attacking and Analyzing NLP: not literally "On the," but the same essayistic register. The title offers a phenomenon to understand instead of a leaderboard to top. (Wallace et al., EMNLP 2019)

Common mistakes​

  • The kitchen-sink title. A Novel Multi-Task Cross-Lingual Contrastive Pre-Training Framework for Low-Resource Question Answering with Knowledge Distillation. Every modifier you add halves the weight of the others. Pick one or two ideas; cut the rest.
  • The mystery title. On the Nature of Things. It is cute, but nobody can search for it or cite it. No reviewer looking for your topic will find it.
  • The buzzword title. Stacking LLM, agentic, and novel in one line tells reviewers you are trying too hard.
  • "Novel." Delete it. Every paper claims novelty. Saying so in the title wastes a word and dares the reviewer to disagree.
  • The over-promise. Solving Mathematical Reasoning with Transformers had better solve it. A title that outruns the results is the first thing a skeptical reviewer quotes back at you.
  • "Towards" as a hedge. Towards X can be honest for early work, but reviewers often read it as "we did not quite get to X." If you got there, drop the "Towards." If you did not, ask whether the paper is ready.

Process tip​

Write five titles. Show them to two colleagues with no context. Ask which paper they would rather read and what they expect it to contain. Fix the gap between what they expect and what your paper delivers. Pick the title that closes it, then sleep on it. Titles improve overnight more reliably than almost any other part of the paper.

Five examples worth studying​

  1. Attention Is All You Need (Vaswani et al., NeurIPS 2017). The declarative claim at maximum risk, and earned.
  2. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding (Devlin et al., NAACL 2019). The name-and-expand template that the field has copied ever since.
  3. SQuAD: 100,000+ Questions for Machine Comprehension of Text (Rajpurkar et al., EMNLP 2016). The resource title that puts its scale in the name.
  4. Mission: Impossible Language Models (Kallini et al., ACL 2024 Best Paper). The hook that pays off in the abstract and the experiments.
  5. With Little Power Comes Great Responsibility (Card et al., EMNLP 2020). Playful on the surface, but "power" is the literal topic. The joke and the keyword are one word.

Further reading​