Deepening Science Teaching and Learning with Language AI

The same artificial intelligence that drives chatbots, virtual assistants, and machine translation of one language to another is making its way into six San Francisco Bay Area middle schools as part of a research effort to support science teaching and learning.

A partnership between GSE Professor Marcia Linn’s WISE–TELS Lab and Educational Testing Service (ETS) combines the research on how students learn science with natural language processing (NLP)* AI technology, allowing teachers to see in real time students’ progress, and more importantly, giving students more engagement with their own scientific ideas as they seek answers.

The project, called Natural Language Processing Technologies to Inform Practices in Science (NLP-TIPS), aims to detect student ideas and empower students to use those ideas as a starting point for deepening their science understanding.

For example, students are asked, “How are mountains formed?” Using the NLP-TIPS, students respond by typing in their answer. Some students might think mountains are a collection of rocks and snow, or may type, “I don’t know.”

“The computer recognizes the student’s idea and engages them in a conversation that first affirms the student’s viewpoint, and then asks more probing questions,” said Libby Gerard, a research scientist in the GSE and co-principal investigator on the project. “For a student who says the mountain is caused by a volcano the NLP-TIPS might say, ‘Good idea. Why does the volcano form at that location?’”

As a student types their ideas, the NLP analyzes the text, and then the algorithm decides whether the computer’s reply should:

  • elicit more details from the student regarding the idea or combination of ideas the student has expressed;
  • distinguish the applicability of the student’s idea when tested in different contexts or when compared to a different interpretation of evidence;
  • help the student discover a new idea to fill gaps in their repertoire by reviewing evidence or another’s perspective; or
  • strengthen the connections among the ideas the student has already expressed.

Each student’s conversation with the computer is recorded and analyzed in real time, offering the teacher insights into whether a student is progressing, stagnating, or possibly off task.

Gerard noted that the research is being conducted at schools with students from diverse racial, ethnic, and linguistic groups. NLP-TIPS will explore how the technology can enable the design of computer-assigned guidance that builds on the cultural experiences of each student including views that may be neglected in typical science instruction.

“Students’ ideas are often generated from their experiences with the world. That’s a valuable starting point when we want students to develop an integrated, long-lasting understanding. Just as important, asking open-ended questions promotes their identity as a science learner,” she said.

NLP-TIPS will explore how to use the NLP to design an adaptive conversation that guides each student to reflect on the evidence underlying their idea. The guidance is designed to lead the student to determine how their ideas connect with, or are distinct from, the evidence presented by instruction. Otherwise, Gerard says, designing guidance that simply tells the students specific ideas or facts to add, the more typical approach of automated feedback programs, can lead to an accumulation of disparate ideas and short-term understanding.

NLP-TIPS, which is supported by a $2.8 million grant from the National Science Foundation, is designed to support teachers who are embedded in complex classrooms that are brimming with the talk, emotion, and activity of 30-plus students.

As Gerard noted, “Teachers want to give individualized guidance to students during inquiry to promote knowledge integration. With class sizes of 30–35 students per teacher however and the frequent re-assignment of teachers to a different grade level or a different school with a different curriculum, it is nearly impossible. Teachers frequently report they are not able to provide the type of guidance for each student they think is needed."

Teachers have also observed that students rarely ask for help with their science ideas, possibly for fear of being wrong, and students who need the most help are also the least likely to ask for it. To remedy this problem, the NLP-TIPS detects ideas and themes in students' written explanations in a science unit, and then initiates an adaptive, guided conversation that encourages students to interrogate their own thinking.

“One way to view this project is by what it isn’t. We’re not providing an automated scoring system to measure knowledge,” Gerard said. “Our research with this project is designed to examine the language and ideas that students are using when they are learning science, help science teachers amplify their impact, and deepen student learning.”

Natural language processing (NLP) is a form of artificial intelligence that, through algorithms, analyzes human language and makes decisions based on the information. For example, NLP is used as a help desk chatbot on some websites and for automatic translation between languages.