Skip to main content

12. Background / Preliminaries

Not every paper needs a separate background section. Some do. Use one when:

  • The paper builds on a specific formalism the reader needs (a particular decoding algorithm, a probability model, a measure).
  • You are introducing notation that will be used heavily later.
  • The audience genuinely spans subfields and you need to level the floor.

If your work is on a topic everyone at ACL already understands (e.g., autoregressive language modeling), skip this section and fold any necessary notation into Methods.

New authors blur these two sections, and their adjacent placement — §2 and §3 — does not help. They do different jobs. Related work is an argument: it positions your paper against competitors and names the gap you fill (see Chapter 11). Background is scaffolding: it sets up the notation and formalism your Methods section depends on, and it cites the origin of a technique rather than its rivals.

A simple test separates them. You can delete a related-work paragraph and the method still stands; delete a background paragraph and the notation in §Methods no longer parses.

Some papers do combine the two under one heading, usually Background and Related Work, when the technical setup is light and the positioning is short. That is a fine choice — but combine them on purpose, knowing you are putting two distinct moves under one header. The names vary too: "Related Work" or "Prior Work"; "Background" or "Preliminaries."

What goes in​

  • Notation. Define symbols once, use them consistently. A small notation table can be worth its weight in pages.
  • The core formalism. State it cleanly, in a few equations if needed. Cite the original source.
  • The specific prior result you build on. If your method modifies a published algorithm, write the original algorithm here so the reader can see what changes in §Methods.

Common mistakes​

  • Lecture-mode. Re-deriving softmax for two paragraphs. The ACL audience knows softmax.
  • No clear function. The background section exists to support the methods. If a paragraph here is not used later, cut it.
  • Inconsistent notation. If h_t is the hidden state in §3, do not switch to h^(t) in §5.

A useful test​

Before you keep a paragraph in the background section, ask: will I refer to this content later in the paper? If not, it does not belong here. Background exists in service of methods. Trivia about the history of the field is not background — it is filler.

Five examples​

  1. Attention Is All You Need (Vaswani et al., 2017). The "Background" subsection is short, focused, and used.
  2. Neural Machine Translation by Jointly Learning to Align and Translate (Bahdanau, Cho, Bengio, ICLR 2015). Sets up the encoder-decoder framework cleanly before introducing attention.
  3. Lexically Constrained Decoding for Sequence Generation Using Grid Beam Search (Hokamp & Liu, ACL 2017). Background on beam search that sets up the modification crisply.
  4. Get To The Point: Summarization with Pointer-Generator Networks (See et al., ACL 2017). Exemplary preliminaries that prepare the reader for the model contribution.
  5. Latent Retrieval for Weakly Supervised Open Domain Question Answering (Lee, Chang, Toutanova, ACL 2019). The preliminaries are tightly scoped to what the model needs.

Further reading​

  • The methods sections of textbooks (Jurafsky & Martin's Speech and Language Processing, free online) for the level of explanation you should aim for.