User Tools

Site Tools


This is an old revision of the document!

Shared Task Description

(Please note that the description is susceptible to be enriched in the following days)

For the shared task, we have nine different treebanks: Arabic, Basque, French, German, Hebrew, Hungarian, Korean, Polish, and Swedish.

Although not always represented in treebanks, most languages have phenomena where space delimited tokens do not correspond to words as active ingredients in syntax. This covers multi-word expressions such as in spite of in English, but also phenomena in Arabic, where conjunctions are attached to the following word.

In order to provide the means to produce and evaluate more realistic parsing models, whenever it's possible, we provide data following different tokenization schemes: one with word gold segmentation, and one in which we have unsegmented text as it would appear in newspapers, etc.

Parsing Scenarios

We have two syntactic frameworks:

  1. constituent structure
  2. dependency structure (in conll07 format)

Constituent structures are available in two formats: an extended PTB bracketed style (eg Penn Treebank with morphological features expressed at the POS or non terminal levels, see below) and, if available, the Tiger 2 format. The latter has the possibility to represent trees with crossing branches, allowing the use of more powerful parsing models (in term of expressivity) than pure PCFG-based parsers.

Dependency structures are available in the CoNLL'07 format.

All treebank instances (dep. and const.) are aligned at the token level and share the same POS tagset and morphological features.

Participants can choose either one of those frameworks, or both, or one by conversion from the other).

Input scenarios

The types of scenarios that we assume in terms of input are as follows:

  1. Gold tags: gold word segmentation and gold tags are given
  2. Predicted tags, gold segmentation: gold word segmentation is given (for the languages where it matters, otherwise standard words), POS tags and morph. features are automatically predicted (available for Basque, German, Hungarian, Korean, Polish, Swedish)
  3. Fully Predicted: sentences are tokenized; raw words are given (available for French, Arabic and Hebrew)

The evaluation task focuses on the last two scenarios but participants are strongly encouraged to provide “gold mode” parsing results so that a performance ceiling can be determined for each system/framework.

Training set size scenario

Our data set contains treebanks with different size (from 6k to 50k sentences). So in order to favor a fair comparison between treebank/parsing model pairs, we also provide training sets with a common size of 5000 sentences. Participants should then also provide results from parsing models trained on the small data set.

To sum up, we have different scenarios with regard to the size of the training sets:

  • the full training set
  • a 5000 sentence training set
  • the full test set will be used for both scenarios, we will later sample a common subset with similar properties in term of sentence length and number of tokens.

Input Formats

The input format is a variant of the CoNLL format for dependencies. This is necessary to represent word segmentation issues and easily allows to include morphological features and alternative analysis. We mark the beginning and the end of words, that does not have to correspond to what we call tokens, which can consist of more than one word.

  • In Hebrew, where several morphemes create a word, we will get something like the following: hebrewIn.pdf
  • Note that if one wants to deliver a lattice in which segmentation is ambiguous, they can do so by adding lines for alternative spans or alternative tags of spans. These lines need not be sorted. See the (real-world) example segmentation lattice here: multi.pdf or the german morphology lattice file (predicted from the SMOR analyser):
  • 0 1 Der PRELS gender=fem|case=dat|number=sg| 1
  • 0 1 Der PRELS gender=masc|case=nom|number=sg| 1
  • 0 1 Der PDS gender=fem|case=dat|number=sg| 1

The format of Form/Lemma/CPos/FPos/Feats is the exact same as in the CoNLL format, including vertical bars separating morphemes, and = separate feature values.

Output Formats

In order to evaluate all scenarios, we consider the terminals present in both the trees, and we will need to keep track of how the given parse is related to the original word tokens (this is true both for parsing over a segmentation lattice or for MWEs)

  • For dependency trees, the parsers are to deliver the standard CoNLL format
  • For constituency trees we require the standard PTB-like trees over terminals (one tree per line)

For the scenario based on fully raw text, we additionally require a file containing the token IDs. That is, for the Hebrew sentence of BCLM HNEIM, after disambiguation as follows B CL FL HM H NEIM, we would get a (constituent/ dependency) tree for these 6 morphological segments, and in a parallel file, we would get the following line:

1 1 1 1 2 2

saying that the first 4 leaves in the syntax structure correspond to the first word in the raw text, and the last 2 leaves correspond to the second word.

This may also be used for multiword expressions: for “I live in Tel Aviv” we would get the tree as usual over the 5 terminals, and the line

1 2 3 4 4

In all cases, the tree terminals hang separately under their parents (in the case of a MWE they will hang flat under a shared parent).

Evaluating All Scenarios

  • For constituent evaluation on gold word segmentation of bracketed output (eg. PTB), we will use parseval: (Note add a link to (W.Maier). How to formulate : “for output coming from parsing models trained on crossing branches trees, they'll need to be be converted to the negra format”)
  • For dependency evaluation on gold word segmentation, we will use the CoNLL 2007 evaluation:
shared_task_description.1370399016.txt.gz · Last modified: 2013/06/05 04:23 by skuebler