---
kind: concept
slug: anaphora-resolution
title: Anaphora resolution
summary: >-
  Anaphora resolution links a pronoun or other reference to an earlier item in the text. Options run from hand-written discourse rules to end-to-end neural models, which the survey reports at about 68.8 CoNLL F1 on English. The scores are author-reported and the metrics differ, so compare only within one metric and setup.
tags: [nlp, coreference, anaphora, discourse, evaluation-metrics]
papers: ["1805.11824"]
related: [coreference-resolution, sentiment-analysis, benchmark-validity]
status: draft
updated_at: 2026-11-09
---

## Overview

A reference, such as a pronoun, points back to an earlier item in the text.
Resolving it means finding that item. The task is often mixed up with
coreference resolution, which groups all mentions of one entity. The survey
keeps the two apart: anaphora links a reference to an antecedent, coreference
clusters mentions [arxiv:1805.11824 §1]. The survey covers English NLP only
[arxiv:1805.11824 §1].

The invariant all variants share: each mention either links to one earlier
mention or starts a new entity. Systems differ in how they find mentions, how
they score candidate antecedents, and what features or learned representations
they use [arxiv:1805.11824 §8].

This page rests on one survey card, [Anaphora and Coreference Resolution: A Review](/papers/1805.11824).
It describes the systems from their original papers, which have no cards
here. Check those originals before implementing any of them.

## Variants

### Rule-based systems

Early systems apply discourse properties and syntactic and semantic
constraints. The survey names the BFP centering algorithm as an early
discourse-based method for pronouns [arxiv:1805.11824 §8.1]. It also names a
multi-sieve deterministic system as the best rule-based system on CoNLL 2012
[arxiv:1805.11824 §8.1]. The survey's table gives the rule-based row (Lee et al.
2013) a CoNLL F1 of 54.82 [arxiv:1805.11824 tab. 7]. Card: [Anaphora and Coreference Resolution: A Review](/papers/1805.11824).

### Statistical and machine-learning systems

These use mention-pair or mention-ranking models trained on hand-built
features. The survey also lists joint inference with integer linear
programming and graph partitioning with conditional random fields
[arxiv:1805.11824 §8.2] [arxiv:1805.11824 tab. 5]. The survey's CoNLL F1 for
this group runs from 61.40 (Durrett and Klein 2013) to 63.02 (Clark and
Manning 2015, entity-centric) [arxiv:1805.11824 tab. 7]. Card: [Anaphora and Coreference Resolution: A Review](/papers/1805.11824).

### Neural mention-ranking systems

A feed-forward or recurrent network scores each candidate antecedent. The
survey groups Wiseman et al. 2015 and 2016, and Clark and Manning 2016 in
their "improving" and "deep" forms [arxiv:1805.11824 §8.3]. The second Clark
and Manning variant replaces the heuristic loss with a reinforcement-learning
objective [arxiv:1805.11824 §8.3]. Their CoNLL F1 runs from 63.39 to 65.73
in the survey's table [arxiv:1805.11824 tab. 7]. Card: [Anaphora and Coreference Resolution: A Review](/papers/1805.11824).

### End-to-end neural systems

Lee et al. 2017 detects mentions and links them in one model. A bi-LSTM reads
the words, every span up to width 10 is scored, the top-scoring spans are kept,
and each is paired with the preceding 250 mentions [arxiv:1805.11824 §8.3]
[arxiv:1805.11824 tab. 6]. The survey reports 68.80 CoNLL F1 for it, the best
figure in its table [arxiv:1805.11824 tab. 7]. The survey also says this was
the best system it knew of, as of its 2018 posting [arxiv:1805.11824 §8.3].
Card: [Anaphora and Coreference Resolution: A Review](/papers/1805.11824).

## When to use which

| Situation | Use | Why |
|---|---|---|
| Target domain differs from training data, or features are costly to build | End-to-end neural | The survey says hand-written features transfer poorly across domains [arxiv:1805.11824 §1] [arxiv:1805.11824 §8.3]. This is an inference from that claim. |
| You need the highest reported CoNLL 2012 English score | End-to-end neural (Lee et al. 2017) | Highest figure in the survey's table, 68.80 CoNLL F1 [arxiv:1805.11824 tab. 7]. The survey does not rerun it, so the figure is author-reported. |
| You already have a trusted mention detector | Mention-ranking neural | The early neural systems take mentions from Berkeley and Stanford tools [arxiv:1805.11824 tab. 6]. Lee et al. 2017 detects mentions itself [arxiv:1805.11824 §8.3]. |
| Pronoun resolution feeds aspect-level sentiment | Any system that resolves pronouns well | One cited study reports about a 10% average gain over baseline coreference modules on its sentiment task. Another reaches about half the gain of perfect anaphora resolution [arxiv:1805.11824 §11]. The survey does not name the systems in these studies. |
| You need a transparent, rule-based pipeline | Rule-based | The survey names rule systems as the early approach [arxiv:1805.11824 §8.1]. Its rule-based row (54.82 CoNLL F1) is well below the neural rows in its table [arxiv:1805.11824 tab. 7], so expect a lower score. |

Training details are missing. The survey states no learning rate, optimiser,
epoch count or batch size for any system [arxiv:1805.11824 tab. 6]. Do not
implement a neural variant from this page alone. Read the original paper.

## Implementation sketch

The common core of mention-ranking systems. Lines follow the survey's
descriptions of how candidate antecedents are scored [arxiv:1805.11824 §8.2]
[arxiv:1805.11824 §8.3].

```pseudocode
# Resolve each mention to an earlier mention or to "new entity"
# Input:  mentions in document order, score(m, c) for a pair, window size
# Output: dict mapping each mention to its antecedent, or None

def resolve(mentions, score, window):
    links = {}
    for i, m in enumerate(mentions):
        # Only look back a bounded number of mentions. Lee et al. 2017
        # used 250 [arxiv:1805.11824 §8.3]
        candidates = mentions[max(0, i - window):i]
        # Score each candidate and the "no antecedent" option. The survey
        # does not say how the null score is computed for every system
        best, best_score = None, score(m, None)
        for c in candidates:
            s = score(m, c)
            if s > best_score:
                best, best_score = c, s
        links[m] = best
    return links
```

End-to-end systems add mention detection and span pruning before this loop.
The survey gives no pruning fraction for Lee et al. 2017 [arxiv:1805.11824 tab. 6].

## Evaluation

Results are not comparable across papers unless the metric and the setup match.
The survey says the papers use different metrics, splits and mention-detection
setups [arxiv:1805.11824 §5.8] [arxiv:1805.11824 §9].

- **CoNLL F1** is the unweighted mean of MUC, B-cubed and CEAFe F1
  [arxiv:1805.11824 §5.5] [arxiv:1805.11824 eq. 12]. Most rows in the survey's
  table use this score.
- **MUC** is link-based and ignores singleton clusters [arxiv:1805.11824 §5.2]
  [arxiv:1805.11824 §5.8].
- **B-cubed** is per-mention. The survey's formula is ambiguous about the
  weighting, so check the original definition [arxiv:1805.11824 §5.1].
- **CEAF** first finds a one-to-one mapping between output and gold clusters,
  then scores it. The survey's table uses CEAFe [arxiv:1805.11824 §5.3]
  [arxiv:1805.11824 tab. 7].
- **LEA** weights links by entity size, which the survey says is meant to avoid
  the mention-identification effect [arxiv:1805.11824 §5.7].

## Contradictions

- **Clark and Manning 2016 (improving), CoNLL F1.** The survey's text gives
  65.39 [arxiv:1805.11824 §8.3]. Its table gives 65.29 for the same system
  [arxiv:1805.11824 tab. 7].
- **Clark and Manning 2016 (deep), which score is which.** The survey's text
  calls 65.73 and 63.40 the CEAF and B-cubed scores [arxiv:1805.11824 §8.3].
  Its table lists 65.73 as the CoNLL score, with CEAFe 59.23 and B-cubed 63.40
  [arxiv:1805.11824 tab. 7]. The survey gives no way to settle this.
- **Lee et al. 2017, score definition.** The survey's text says the final score
  adds the two mention scores and the pair score [arxiv:1805.11824 §8.3]. Its
  equation gives only the pair term [arxiv:1805.11824 eq. 28].

## Papers

- [Anaphora and Coreference Resolution: A Review](/papers/1805.11824): a survey
  of AR and CR. It defines the metrics, groups systems into rule-based,
  statistical and neural, and tabulates CoNLL 2012 English scores. It runs no
  experiments of its own.