Innovation Blog

Teachers use video lessons a lot these days, some they create themselves, and other videos that they find on Youtube. Not everyone has time to build a set of cued comprehension questions for each video lesson. Some teachers assign their students to summarize what they saw in the video. This is a fantastic way to keep kids engaged in the video, but it’s a lot of work scoring them.

The Innovation AI grading assistant is the perfect solution. In this post, I would like to show you how to generate a video lesson in Innovation and then quickly train the AI to help you score the summaries very rapidly.

Quick Links to Short Video Tutorials

• Add a video for students to summarize
• Student access to the video and summary space
• Part 1, Train the AI grading Assistant
• Part 2, Train the AI grading assistant
• The AI can coach students to write better summaries.

A Word about the Innovation AI

There are two broad categories of AI: those that work using a complex algorithm and those that work using large language model learning. The Innovation AI is of the former type. The latter, the machine-learning AI, is exemplified by ChatGPT and the Bing AI under development right now. They “learn” by analyzing vast amounts of data across the internet. The Innovation AI is trained on five to seven models that the teacher provides for comparison.

I developed the Innovation AI to help me grade summaries and short answer tests. When I taught social studies, I often assigned summaries of texts instead of comprehension questions. It works by comparing the student text to a number of models and scoring the comparison on a dozen features. These features include measures of similarity like cosine and Jaccard, as well as readability, number of words, level of text complexity, and so forth. The scoring rubric was designed using 500 of my students’ work submissions that I had scored manually so that the AI essentially grades as I would.

The Innovation AI is highly effective for helping you score summaries and short answer responses where the range of possible answers is fairly limited. The AI does effectively recognize different ways to say the same thing using natural language processing algorithms.

Training the AI

When you train the AI, you give it model answers to use in the comparison algorithm. For short answer tasks, the limit is presently five models. For writing samples such as summaries and compositions, you can store up to seven model answers.

The process, in summary, is this: (1) Compose your own summary of the task or let the Innovation AI generate a summary from the source text for you; (2) Manually score the work submissions of your students who usually get full credit. When you find a submission to which you would award full credit, ask the AI to score it. If the AI cannot recognize it as a full credit answer, you “add it to the corpus” of model answers. The next time you ask the AI to score a student submission, it will compare it to each of the models in the corpus and award the highest score earned by the student in those comparisons.

Since many of us reuse our assignments from year to year, you really only have to do this once. I trained the AI on most of my Global 9 and Global 10 assignments in 2018 and just continued to score with those for several years.

Creating an Inbox for a Task with Embedded Video

1. Select the Inbox button from the new course playlist element dashboard.

2. Enter the title and some optional attributes. Paste in the embed code from the video you want students to watch and summarize.

3. Once created, you can click and drag the element to its right position in your class playlist.

Students Save their Summaries

When students access the task from the course playlist, they will see the video you embedded and the space underneath to compose their summary.

AI can coach students to write better!

The AI grading assistant can be engaged to coach students along the way. As they compose their summaries, they can periodically click the “Coach” button to get an estimate of their grade so far. In my experience, this promoted student prolonged engagement for a better work product.

The stimulus-based multiple-choice test item was introduced into the New York State social studies Regents examinations starting in 2019 for Global History and Geography II and for United States history. The task poses challenges for students such that it merits some regular, focused training throughout the year.

In a stimulus-based task, the student is directed to respond to a document, map, or image using their ability to analyze and their knowledge of historical context. In the case of the New York State exams, there are eighteen “task models” used when designing questions. For example, a student may be asked to evaluate and classify (identify) best use of a source or to respond based on knowledge of historical context. Principles of reliability assessment are applied here, such as when students are asked to identify point of view, purpose, context, bias, format of source, location of source in time and/or place, and/or intended audience of sources using background knowledge.

Click here to shop stimulus-based tasks at my store for grades seven through eleven social studies.

The first important habit of thought to train students to engage is to think beyond the document. Habit since their first reading lessons has asked them to find the answer in the text somehow. It takes a lot of practice and reinforcement to get students to activate their schema on the topic; to think of the story of which the document is but a fragment. The question cannot be answered without background knowledge.

• part I of the Global and US History Regents
• 25-30 questions
• primary or secondary source documents
• M-C questions are always paired with stimulus
• primary or secondary source
• maps
• charts
• cartoons
• may have more than one stimulus tied to it
• no more stand-alone questions
• estimate 30-45 minutes for this part of the exam

I used almost exclusively primary source documents for my stimulus-based tasks. This can be challenging for weaker readers, but with practice in skills for addressing difficult texts, this obstacle can be addressed.

My custom was to assign a stimulus-based multiple-choice at the end of every unit starting in October. At first, students find these very difficult. I use a z-score standardization procedure to adjust the scores so as not to bomb out their GPA while they are just learning. Click here to read up on standardized scoring. It is a great way to score students in tasks they are not yet proficient at.

• Practice! Students are generally not good at these at first. 
• Read the question first. be certain you know what it is asking.
• Remind yourself that the answer is rarely found in the document itself.
• Identify the historical time period the documents go with.
• Consciously call to mind the historical context of the document before you read. Try to recite to yourself who, what, when, where, why of the time period.
• Use process of elimination to narrow down the options.

Some of my students always would wonder why ask questions this way. If the test writers want to know whether a student knows something, why not just ask? I don’t have a good answer for this. I strongly support instruction in social studies that calls upon students to think critically and make meaningful connections with knowledge. I also think that students should actually possess knowledge. This assessment method was no doubt inspired by AP history exams. It remains a question in my mind as to whether this level of complexity is necessary for an instrument for secondary school evaluation. If we want to know whether the student knows what caused the French Revolution, for example, maybe we should just ask them that?

The updated New York State Regents examination in United States History and Government, part II, is a short essay task designed to measure students’ ability to work with historic documents. It is a mature version of the “CRQ” found on the tenth grade Global Regents. Students are called upon to understand text, engage it with historical context, and assess a text’s reliability.

In document set 1, students describe the historical context surrounding two documents and identify and explain the relationship between the events and/or ideas found in those documents (Cause/Effect or Similarity/Difference or Turning Point).

Turning point is always the most challenging for students, mainly because it demands a strong knowledge of historical context which only the higher performing students usually possess. In stating similarities and differences, it is important to stress to students that this should be a substantial feature of the two texts, not trivial. For example, some students may respond something like this: “Document one is a cartoon and document two is a newspaper report”. This is trivial and should be discouraged. For cause and effect analyses, remind students that some events may lie outside the documents at hand, so they may need to rely on their historical knowledge.

Click here to short for short essay prompts at my store.

Document set 2 asks students to describe the historical context surrounding two documents and (for one identified document) analyze and explain how audience, or purpose, or bias, or point of view affects the document’s use as a reliable source of evidence.

A good strategy for ensuring students possess the skills to address this task is to be certain to assign one every month or so throughout the year, followed by a debriefing where the class can study their classmates’ work (anonymously) and develop strategies for improvement.

Weaker readers are particularly disadvantaged in this task, although since the test items are field tested before administration it is likely the field testing will mitigate some issues with the difficulty of reading some primary source texts. Students can be taught compensatory reading strategies to help deal with difficult texts.

As always, the challenge is to ensure that students have learned a strong body of historical context. That is, the best marks are reserved for those who actually recall the history and who can analyze it (cause-effect, turning point, etc.) This is best achieved by regularly administering quizzes on historical knowledge. I like to give students time in class to study for these. The apps here at Innovation Assessments are especially suited to that end. A lot of social studies assignments can tend to be just look-it-up and transfer kinds of exercises without real demands on students to remember. This is an easy instruction error to remedy.

Teaching factors affecting the reliability of sources is another matter. This takes a great deal of time and practice and, I would argue, is of upmost importance for a person’s education in this day and age. Students, for the most part, do not intuitively analyze and explain how audience, or purpose, or bias, or point of view affects a document’s use as a reliable source of evidence. They tend to take what we give them on face value. It is important to teach students to think critically and approach all historic documents with a healthy skepticism.

Click here to shop my store for primary source essay prompts for grades seven through ten.

I think teaching reliability should begin young, down in middle school. Engaging students with documents that have very vague reliability weights is a good practice. In a debriefing after the task, it is useful to anonymously display some student reliability evaluations for all to see and to discuss. It is important to do this regularly, starting off right at the start of the school year. There are really no stock phrases that students can learn by rote for this, given the variety of context and source material. I had the benefit of working in a small school where I had the same students grades seven through ten or eleven, so I could implement reliability assessments early in my program. In larger school districts, it would be good to consider a commitment to reliability factor training from an early age. I assigned one longer primary source to analyze each topic (so about once a month). This short essay included an extended analysis of reliability in a conclusion paragraph. The training paid off and when my eleventh graders were preparing for the Regents in US History and Government, they had little difficulty with reliability factors.

A good piece of advice on this is to assign students to do this every month in grade eleven. I suggest assigning it as a test each time. Coach students on the historical context they have to memorize in advance. Lots of teachers assign these for homework, but this entirely misses the point of such training. Student independent work practices are highly efficient in applying the minimal effort to a task, including copying their colleagues’ work or copy-pasting from a source. If they are not doing these without notes, they’re not really practicing.

If you are afraid to assign your students this as a test because they are not likely to do well at first and don’t want to bother their GPA, I recommend using standardized scoring. You can use the z-score calculator here at Innovation Assessments. Use 78 as your standard mean and 14.8 as your standardized standard deviation. Read more about standardized scoring here and where I got those figures. The beauty of this system is you can apply this to their grades every month and as the class improves, as the class average approaches the standardized mean (78 in this case), then the algorithm affects their scores less and less.

Once you have scored their papers, select out some problem responses for class discussion in a debriefing. Keeping the responses anonymous, review how to improve the answers next time.

The short essay is 14% of the Regents score, so for passing the text it’s a good practice. But I would suggest that this kind of work as a regular lesson is extremely valuable as an educational tool. Studying primary sources rightly should take center stage in our social studies lessons. It is not just teaching to the test to do this. It is developing a critical thinking skill set and insisting on recall of historical context that are the values here. Remember, that the highest valued performance is that which is based on a substantial recall of historical context!

Dear fellow social studies teachers,

I would like to play the role of a crusader in this post, if you don’t mind. There’s something I would like you to consider about your teaching practice, or maybe reconsider as the case may be. I’d like you to abandon assigning your students comprehension questions on most kinds of texts you assign in your social studies class. Instead, I’d like you to consider assigning students to summarize instead.

When students do the search and copy instead of reading the text fully, they deny the author the chance to build with the reader an information schema of their own that lends itself to long-term recall.

Okay, so who do I think I am anyway to question a practice that’s at least a century old? Well, I cannot claim to be any super authority. I taught social studies for eighteen of my thirty-two years teaching. I worked in small, rural mountain schools and my roster averaged about one hundred students a year, which is likely smaller than that of my readers. I think these are good starters to convince you to at least hear me, but I have always felt that the argument from authority to be one of the most pernicious fallacies. So in honor of my thirty-two years, grant me the indulgence of reading further. As for accepting my suggestion, let the arguments sway you themselves.

When you assign your students to read a text selection and answer questions on it, you must know what the majority of them do. They skim the text to locate the answer, often they copy it wholesale, and then they’re done. Teachers who accept this kind of work are missing an important opportunity to foster greater reading comprehension skills and greater long-term recall. Writers, even writers of textbooks, build meaning in a schema of content to deliver to the reader, who constructs meaning related to their experience. Ideas are hierarchically arranged and organized to support a set of main ideas. When students do the search and copy instead of reading the text fully, they deny the author the chance to build with the reader an information schema of their own that lends itself to long-term recall. At the risk of putting too fine a point on it, questions on text are largely a useless exercise even when the questions are of high quality.

Once students leave sixth grade in my region, the formal lessons in reading stop. But a great number of sixth graders do not read at a sixth-grade level. They cannot advance their reading without texts at their level. And they have to actually read these. It is possible and desirable in middle and junior high school to promote the development of reading in youngsters by structuring what they do to process text. Teaching students to write competent summaries is one of the best ways to let them develop their reading skills. (This should be paired with offering texts at their independent reading level whenever possible).

What does the research say?

Summarizing and note-taking are two of the most useful academic skills students can have. (Marzano, Pickering, & Pollock, 2005)

Summarizing is in the top two most powerful writing tasks that support the development of student reading. (Graham & Hebert, 2010)

Click here to shop my TpT store for passcodes to give your students access to training videos with embedded auto-corrected questions on composing summaries.

Writing to Read

I was greatly influenced by a paper entitled Writing to Read: Evidence for how Writing can Improve Reading in 2010 published by the Carnegie Corporation of New York. A meta-analysis clearly demonstrates that composing summaries is superior to answering questions on text. The reader may argue that this research was conducted to address methods for using writing to enhance reading. But I would argue that (a) this is rightly a major goal of assigning most secondary students a textbook reading task and (b) the same cognitive processes that go into reading comprehension go into long-term recall of information. Students answering questions on text are less likely to be encoding from working memory into long-term memory than those composing summaries in their own words.

The Five and Three Summary Style

Around 2006 when I was making major investments in curriculum and methodology development, I sat down with a noted professor of literacy who happened to be working in our district, Trudy Walp. I explained my qualms about assigning students questions on text and I asked her, what is the very best thing I can have students do with text from a reading specialist’s perspective. She did not hesitate to say: have them retell what they read.

When I assigned my students summaries on their textbook articles early on, my top students wrote terribly long and detailed compositions that were not brief enough to be a summary. I needed to shorten my students’ summaries and to develop a rubric for evaluating the quality of a summary. The resulting task I used for some fifteen years: the Five and Three.

The assignment for the five and three is to summarize the assigned text (usually four pages in a typical high school textbook) in five sentences exactly, no more and no less. This is the “five”. The “three” is the personal reaction to the text. Students were assigned to connect this text to their own life somehow. What does it remind them of? What do they think about what they read? Why? These had to be exactly three sentences long. This was the second element that the reading specialist recommended to me. Students need to make the text meaningful to them in some personal way that makes new information integrate into their preexisting schema.

The five-and-three became the foundation of my students’ textbook reading work. They also had the option of composing Cornell notes for an assigned reading. I would invite the reader to return to the blog for a post dedicated to extolling the virtues of Cornell note-taking. The five-and-three faded in importance in certain classes where video lessons became an effective information delivery tool when reading instruction was no longer a priority.

The Mechanical Summary

The mechanical summary evolved in 2020 when I had a group of learners who struggled and who just would not produce summaries even given time to do so in class. I offered them the opportunity to write a summary by copying the first sentence of each paragraph word-for-word, then to connect and arrange these into complex sentences such that there were only five sentences. They had to compose the three-sentence personal reaction to the text just like normally done. For this, I offered a maximum score of 76 because, I argued, it had less value not being processed in their own words. I reason that not processing in their own words likely limited the ability of encoding to happen from working memory to long-term memory.

With the Assistance of AI

At this point you are likely wondering how you could possibly grade all these summaries. I will grant you, the workload takes some management, but I would argue that it is only slightly more time consuming than grading answers to a set of questions and the benefit to your students makes it all very worth it.

I’m an amateur programmer and I developed an AI grading assistant that is very good at scoring summaries. This app is available to subscribers to Innovation Assessments. Say, why not sign up for a free 60-day trial? The price is surely right if you decide to subscribe?

The AI grading assistant was accurate enough to save me tons of time scoring summaries. The summarizer app actually let me create models on which to train the AI that were highly accurate when compared to human-generated summaries… eerily so!

This blog post describes the AI grading assistant.

The ability to skillfully summarize is a lost art worth recovering for our students. It enhances the development of their reading ability and it promotes greater recall of the content we teach. It’s my hope my experience developing this system can make this suggestion seem do-able and worthwhile.

SOURCES

Graham, S., & Hebert, M. (2010). Writing to Read: Evidence for How Writing Can Improve Reading. New York: Carnegie Corporation.

Marzano, R. J., Pickering, D. J., & Pollock, J. E. (2005). Classroom Instruction that Works: Research-Based Strategies for Increasing Student Achievement. Upper Saddle River, NJ: Pearson Education, Inc.

When the new Frameworks for social studies in New York State came out about six years ago, the first item that caught my eye was the constructed-response question sets. If you are not familiar with this task, in brief, it calls upon the student to examine historical sources (I use exclusively primary sources) by providing historical or geographic context, identifying the point of view, intended audience, or purpose of a document and then using the two documents in either compare-contrast, cause-effect analysis, or turning point identification.

Global 9 CRQ Products in my store, organized by topic.

Global 10 CRQ products in my store, organized by topic.

Originally, the task was supposed to call upon the student to evaluate the reliability of the source. This was scrapped, probably after field testing questions. I regret their decision to scrap this (though it does appear as part of the analogous task, the short essay, in US History grade 11). Students can be taught to evaluate the reliability of sources at an early age (I incorporated it into my sixth- through eighth-grade social studies work). Students can be trained to evaluate based on features like the point of view (bias), intended audience, purpose, time and place, and authorship. Teaching this does require patience and practice. What I think happened is this: few teachers address the reliability of sources and so when they field tested the new exam format, everyone did so poorly that they scrapped the question. I wish I could find out how my kids did. I had been teaching the reliability of sources since they were in middle school.

I advocate a strategy of assigning one CRQ in each unit of study in grades nine and ten without access to notes. I used this as one of the tests at the end of a unit of study. I would tell students about what kinds of documents would appear and what historical context they should be able to recall in advance. It is a challenging task for them.

Click here to purchase a set of PowerPoint slide shows for teaching reliability factors from my store.

The first challenge for novices is to understand what it means to provide context. Faced with the question “What is the historical context of this document?”, beginners will retell what the document says. The reason for this mistake is that, since they were little kids, teachers have asked them to relate what a text means to prove they understood it. This task asks students to bring to bear what they know about the history behind the document, which requires a level of recall I don’t think we’re asking students to do enough.

Another challenge for students writing the CRQ is the third question where they must use two documents in analysis. Cause-effect relationships are the easiest for students, it seems, so I teach those first. When students are asked to compare and contrast the documents, some students have to be cautioned to select significant elements to compare. The fact that one document is a map and one is a diary entry, for example, is not a significant fact. The turning point question is the hardest, mostly because it calls on students to recall history and to understand historical trends both before and after the event. I find it useful to break this up into small tasks: first say what the turning point is, then say what is so special about it.

After a difficult task, especially for writing, I like to do a “debriefing”. In the debriefing after the CRQ, I like to share student answers anonymously in a slide show and discuss them. It is a very effective strategy for stamping out common errors early and permanently. Sample debriefing PowerPoints from an eighth-grade and ninth-grade CRQ are posted below.

Certain constant reminders were recurring for all my classes:

— Things are not reliable because of what they say or show.

— Things are not reliable because they are in quotation marks.

— Things are not reliable because they happen to match what you think is already true.

— Unreliable statements can be true.

— When you’re asked for historical context, do not describe what the document says.

And what about middle school? Can the constructed-response task be modified for that level?

Students with an IEP often are very challenged by these tasks because of the reading level of the texts. Since I use primary sources exclusively in my CRQ’s, this is doubly difficult for students with reading limitations. The answer to this is to consider obtaining CRQ’s designed with a lower reading level.

If I may be so bold as to give advice, I would suggest that students do these throughout Global nine and ten, perhaps once a month. I suggest that they be as tests instead of open-book or take-home assignments because students need to prepare for recall situations. The debriefing afterward is vital to class improvement and it makes for a long-lasting correction. I would advocate resisting the temptation to leave this off for the Global ten teacher. It really does take a long time to learn to do this well.