Discord

Checklist AICC 2026 Round 9 · Task 1

Word Lookups

Segment unspaced Mandarin Chinese sentences into words by predicting a BMES tag for every character, without pretrained models or labelled data.

  • NLP
  • Unsupervised Chinese word segmentation (BMES sequence tagging)
Read the official statement

The task

A friend's browser extension looks up words by hovering over them in subtitles. It works for languages that separate words with spaces, but Mandarin Chinese does not mark word boundaries, so a segmentation model is required. Because the extension must run in real time, no pretrained models may be used.

Each input is a Chinese sentence given as a stringified Python list of characters. The model must output one BMES tag per character: B (beginning of a multi-character word), M (middle), E (end) or S (single-character word); the output sequence has the same length as the input.

Both train.csv and test.csv contain only the ID and chars columns; no word-boundary labels are supplied, and pretrained models, pretrained embeddings, external dictionaries and manually labelled data are all forbidden.

Abridged by SOTA from the official materials. The official statement has the exact rules, and it wins wherever this summary differs.

At a glance

You get
train.csv and test.csv with columns ID and chars (a Chinese sentence as a stringified list of characters, to be parsed with ast.literal_eval). The ID column must not be changed.
You submit
CSV with columns id and bio_tags, where bio_tags is the predicted BMES sequence as a string representation of a Python list, e.g. "['B','E','S']".
Scoring
Boundary F1: predicted and true BMES sequences are converted to sets of word-boundary positions (E and S end a word; B and M continue it), and F1 is computed over all samples after concatenating the boundary positions. Higher is better.
Rules
  • No pretrained models, pretrained embeddings, external dictionaries or manually labelled data.
  • Maximum team size 10 on Kaggle; at most 25 submissions per day.
  • AICC contest rules (stated on each Kaggle rules page, not enforceable): no use of LLMs for writing code or getting task ideas; no internet use other than official library documentation and the contest platform; no communication with anyone during the contest; clarifications only via the #clarification-requests channel on the AICC Discord server.
Format
AICC Round 9, online on Kaggle, 24 Jul 2026 18:00 UTC – 26 Jul 2026 17:00 UTC (Kaggle deadline for this task).

Competition

Solutions

Details

Year
2026, Online (Kaggle)
Round
Round 9 · Task 1
Language
English
License
Varies by task: Word Lookups — MIT; Buried Fault and Shuffled — CC BY-NC-ND 4.0 (Kaggle competition licences). Solutions repository: MIT., as stated by the source