Multilingual Neural Semantic Parsing for Low-Resourced Languages

Abstract

Multilingual semantic parsing is a costeffective method that allows a single model to understand different languages. However, researchers face a great imbalance of availability of training data, with English being resource rich, and other languages having much less data. To tackle the data limitation problem, we propose using machine translation to bootstrap multilingual training data from the more abundant English data. To compensate for the data quality of machine translated training data, we utilize transfer learning from pretrained multilingual encoders to further improve the model. To evaluate our multilingual models on humanwritten sentences as opposed to machine translated ones, we introduce a new multilingual semantic parsing dataset in English, Italian and Japanese based on the Facebook Task Oriented Parsing (TOP) dataset. We show that joint multilingual training with pretrained encoders substantially outperforms our baselines on the TOP dataset and outperforms the state-of-theart model on the public NLMaps dataset. We also establish a new baseline for zero-shot learning on the TOP dataset. We find that a semantic parser trained only on English data achieves a zero-shot performance of 44.9% exact-match accuracy on Italian sentences.

Don't Parse, Insert: Multilingual Semantic Parsing with Insertion Based Decoding

“…We also test on the multilingual TOP dataset (Xia and Monti, 2020) to other languages providing human annotated Italian and Japanese test sets. TOP contains a much larger test set compared to ATIS.…”

“…1, the nested slots make it harder to parse using a simple sequence tagging model. We also do experiments on multilingual TOP (Xia and Monti, 2020) with Italian and Japanese data. In this dataset, the training and validation set is machine translated while the test set is annotated by human experts.…”

“…In this way, the pointer embedding will be replaced by the corresponding outputs from the encoder. We test our proposed model on several well known datasets, TOP (Gupta et al, 2018), ATIS (Price, 1990), SNIPS (Coucke et al, 2018), MultiATIS++ (Xu et al, 2020) and multilingual TOP (Xia and Monti, 2020). In summary, the main contributions of our work include:…”

FastRAT: Fast and Efficient Cross-lingual Text-to-SQL Semantic Parsing

“…The efforts of the scientific community to build systems for cross-lingual semantic parsing have led to the development of a series of relevant benchmark datasets (Min et al, 2019;Dou et al, 2023;Zhang et al, 2023). Most recent approaches for cross-lingual semantic parsing have sought to localise parsers to new target languages using backtranslations (Sherborne et al, 2020) or machine translation (Xia and Monti, 2021;Shi et al, 2022). Such solutions precondition access to high-quality machine translation or source-target language alignment systems.…”

“…While schema information of non-English databases is often available in English, there are several challenges in transferring a monolingual system to the cross-lingual setup, where the natural language question is in a different language than the database. Most existing cross-lingual solutions have focused on shared database semantic parsing setups where evaluation databases are known during training (Sherborne et al, 2020;Xia and Monti, 2021;Sherborne and Lapata, 2022).…”