The research team hails from Korea University as well as NCSOFT, a South Korean video game developer. Their newly developed question-aware sentence-gating networks allow for sentence-level information to be taken into account during word-level encoding.
The ability for machines to derive meaning from text is no longer some far-off fantasy, but being able to answer questions based on a passage requires a level of semantic reasoning that extends beyond mere lookup and basic compositionality. Typically, a question-answering task is a matching game, whereby the machine aligns the word-level encoding vectors in the question with a corresponding set of word-level vectors in the passage. Yet, to effectively find the correct answer, a machine should synthesize sentence-level and even paragraph-level context, making sense of the higher-level semantics these grammatical structures entail. While prior methods have concentrated on making improvements to word-level understanding, this new research implements networks that “directly impose the sentence-level information into individual word encoding vectors.”ranking chwilowek
To form the question-aware sentence-gating networks, the researchers created encoding vectors of individual words, in which the original word vector was combined with its complementary sentence vector. Model performance was measured with the Exact Match (EM) score, which identifies the ratio of questions where the model prediction aligns with the ground truth answer. The sentence-gating model exceeded the EM of the established baseline by 1.53%, suggesting that a blend of word and sentence context results in more accurate answer predictions.
As evidenced by the experiment, the sentence-gating technique provides words with greater contextual information, which enriches their semantic standing in the text. Looking into the future, the researchers see potential for the success of sentence gating to be replicated among other machine comprehension tasks. Automatic summarization, machine translation, and named entity recognition are among the various applications that could benefit from more powerful comprehension tools.
Find out more here.
Kaylen Sanders, ODSC
I currently study Computational Linguistics as an M.S. candidate at Brandeis University. I received my Bachelor's degree from the University of Pittsburgh where I explored linguistics, computer science, and nonfiction writing. I'm interested in the crossroads where language and technology meet.
ODSC’s Accelerate AI focuses on three key areas: Innovation, Expertise, and Management. Learn what the latest advances in AI and applied data science are, how they can affect your company, and how to build an effective team around their potential. Ready to learn more? Learn more here.
- Reviewing Amazon’s Machine Learning University – Is it Worth All of the Hype? 39 views | by Daniel Gutierrez, ODSC | under Featured Post, Modeling
- 7 Reasons Your Data Science Resume is Suboptimal 35 views | by ODSC Team | under Career Insights, Featured Post
- Cracking the Box: Interpreting Black Box Machine Learning Models 33 views | by Yuriy Gavrilin | under Machine Learning, Modeling