Researchers developed a new way to understand language using stacked convolutions instead of recurrent neural networks. Their method is faster and can handle long sentences well. The new approach outperformed previous models and achieved top results on language tests. This means they found a better way to understand and process language quickly and accurately.