Gold standard
To
build the golden standard, we extracted translations from
manually compiled pairs of K Dictionaries (KD), particularly
its Global series. The coverage of KD is not the same as
Apertium. To allow comparisons, we took the subset of KD that
is covered by Apertium to build the gold standard, i.e., those
KD translations for which the source and target terms are
present in both Apertium RDF source and target lexicons. The gold standard remained hidden to
participants. Graphically, for the FR-PT pair:
Evaluation process
For each system results file, and per language pair, we will
1.Remove duplicated translations (some systems might produce
duplicated rows, i.e., identical source and target words, POS
and confidence degree).
2. Filter out translations for which the source entry or the
target entry are not present in the golden standard
(otherwise we cannot assess whether the translation is correct
or not). Let’s call systemGS the subset of translations
that pass this filter.
3. Translations with confidence degree under a given threshold
will be removed from systemGS. In principle, the used
threshold will be the one reported by participants as the
optimal one during the training/preparation phase.
4. Compute the coverage of the system (i.e., how many
entries in the source language were translated?) with respect to
the gold standard. Graphically, for the source language:
5. Compute precision as P =(#correct translations in
systemGS) / |systemGS|
6. Compute recall as R =(#correct translations in
systemGS) / |GS|
where GS is the set of translations in the gold standard for a
given language pair
7. Compute F-measure as F=2*P*R/(P+R)
NOTE: The precision/recall metrics calculated after applying the filtering process explained in point 3 corresponds to what in [1] is defined as both-word precision and both-word recall. The idea is to reduce the penalization to a system for inferring correct translations that are missing in the golden standard dictionary because human editors might have overlooked them when elaborating the dictionary. Notice that in previous TIAD editions we only filtered out translations for which the source entry were not present in the translation (which led to computing the so-called one-word precision/recall) thus only partially covering such goal.
Baselines
We have computed results also with respect to two baselines:
Baseline 1 [Word2Vec]. The method is using Word2Vec [2] to transform the graph into a vector space. A graph edge is interpreted as a sentence and the nodes are word forms with their POS-Tag. Word2Vec iterates multiple times over the graph and learns multilingual embeddings (without additional data). We use the Gensim Word2Vec Implementation. For a given input word, we calculate a distance based on the cosine similarity of a word to every other word with the target-pos tag in the target language. The square of the distance from source to target word is interpreted as the confidence degree. For the first word the minimum distance is 0.62, for the others it is 0.82. So multiple results are only in the output if the confidence is not extremely weak. In our evaluation, we applied an arbitrary threshold of 0.5 to the confidence degree. You can find the code here. NOTE: In the TIAD'21 edition the Word2Vec baseline, although based on the same principles, has been re-implemented and re-trained to be adapted to the new Apertium RDF v2 dataset, and lead to different (generally better) results than in the previous TIAD editions.
Baseline 2 [OTIC]. The One Time Inverse
Consultation (OTIC) method was proposed by Tanaka and
Umemura [3] in 1994, and adapted by Lin et. al [4] for the
creation of multilingual lexicons. In short, the idea of the
OTIC method is to explore, for a given word, the possible
candidate translations that can be obtained through intermediate
translations in the pivot language. Then, a score is assigned to
each candidate translation based on the degree of overlap
between the pivot translations shared by both the source and
target words. In our evaluation, we have applied the OTIC method
using Spanish as pivot language, and using an arbitrary
threshold of 0.5. You can find the code here.
NOTE: Same as the Word2Vec baseline, the OTIC baseline was
re-run for TIAD'21 to be adapted to the new Apertium RDF v2
dataset. The results are generally worse than in TIAD'20 (with
the smaller Apertium RDF v1 graph).
Evaluation results
NOTE: A direct comparison between these results with such of previous TIAD editions is only relative because of two reasons: (1) the use of a new development dataset in TIAD'21, which is the larger Apertium RDF v2 graph instead of Apertium RDF v1, which lead to a different gold standard (Apertium-KD intersection) and (2) the use of improved metrics in the evaluation, as explained above: both-word precision and both-word recall [1].
Table 1: averaged systems results. Results per language pair have been averaged for every system and ordered by F-measure in descending order. The results correspond to the optimal threshold reported by every team, or an arbitraty 0.5 threshold when no preferred threshold was indicated.
Table 2.1: Systems results for EN-PT.
Table 2.2: Systems results for EN-FR.
Table 2.3: Systems results for FR-EN.
Table 2.4: Systems results for FR-PT.
Table 2.5: Systems results for PT-EN .
Table 2.6: Systems results for PT-FR.
Results with variable threshold (averaged across
language pairs):
Table 3.1: Results for ACDcat.
Table 3.2: Results for PivotAlign-F.
Table 3.3: Results for PivotAlign-P.
Table 3.4: Results for PivotAlign-R.
Table 3.5: Results for TUANMUSEca.
Table 3.6: Results for TUANMUSEes.
Table 3.7: Results for TUANWEcb.
Table 3.8: Results for TUANWEsg.
Table 3.9: Results for ULD_MUSE.
Table 3.10: Results for ULD_graphSVR.
Table 3.11: Results for ULD_mbert.
Table 3.12: Results for ULD_oneta2.
Table 3.13: Results for ULD_onetaSVR.
Table 3.14: Results for ULD_vecmap.
Table 3.15: Results for baseline-Word2Vec.
Table 3.16: Results for baseline-OTIC.
References
[1] S: Goel, J. Gracia, and M. L. Forcada, "Bilingual
dictionary generation and enrichment via graph exploration"
[UNDER REVIEW], Semantic Web Journal, 2021
[2] T. Mikolov, K. Chen, G. Corrado, and J. Dean, "Efficient Estimation of Word Representations in Vector Space", 2013.
[3] K. Tanaka and K. Umemura. "Construction of a bilingual dictionary intermediated by a third language". In COLING, pages 297–303, 1994.
[4] L. T. Lim, B. Ranaivo-Malançon, and E. K. Tang. "Low cost construction of a multilingual lexicon from bilingual lists". Polibits, 43:45–51, 2011.