Table of Links

Abstract and Introduction

Related Work

Semantic Parsing for Wikidata

WikiWebQuestions (WWQ) Dataset

Implementation

Experiments

Experiment with QALD-7

Conclusions, Limitations, Ethical Considerations, Acknowledgements, and References

A. Examples of Recovering from Entity Linking Errors

A. Examples of Recovering from Entity Linking Errors

Here, we illustrate our proposal of using entity mentions to recover from entity linking errors. In the training set, we have the following example:

• Query: What year did giants win the world series?

• Original Gold SPARQL:

SELECT DISTINCT ?x WHERE { ?y wdt:sports_season_of_league_or_competition wd:Q265538; wdt:winner wd:Q308966; wdt:point_in_time ?x. }

• Gold Entity linker result:

World Series (QID Q265538),

San Francisco Giants (QID Q308966);

• ReFinED result:

San Francisco Giants (QID Q308966);

Here, the ReFinED entity linker model fails to identify the “World Series” entity. Our proposal of mentions gives the semantic parser a chance to recover from entity linker failures. To train the parser to generate mentions, our training includes samples like this:

• Query: what year did giants win the world series?

• ReFinED result:

San Francisco Giants (QID Q308966);

• Gold target:

SELECT DISTINCT ?x WHERE { ?y wdt:sports_season_of_league_or_competition; wd:world_series; wdt:winner wd:Q308966; wdt:point_in_time ?x. }

The gold query mentions “world_series”. At inference time, our heuristics use the predicted mention to look up the actual Wikidata entity. For example, if wd:world_series is predicted at inference time, our heuristics maps it back to wd:Q265538.