How we retrieve moments

We do not find our memories by date. We reach them through cues: a place, a person, a season, a song, the thing that happened just before or just after. Any one of these can open the door, and any one of them can be missing. Memory is not a list of entries waiting for the right search term. It is a web, and you can walk into it from almost any side.

This is a post about how that works, what decades of memory research have found, and why it is quietly comforting to know.

The moment it comes back

You know the feeling. You are standing in a kitchen that smells of cardamom, and suddenly you are eight years old at your grandmother's house. Or a song comes on in a shop, and for three minutes it is the summer you were nineteen, and you can feel the heat of a particular car seat. You did not decide to remember. Something reached in and pulled the moment out.

Writers have described this for a long time. Marcel Proust built a whole novel around a cake dipped in tea. But it is not only a literary idea. It is, very nearly, the way memory works for all of us.

Memories are episodes, not facts

In 1972, the psychologist Endel Tulving drew a line that still shapes how we think about remembering. He separated semantic memory, the facts you know (Paris is in France, your phone number), from episodic memory, the experiences you lived through, each one tied to a time, a place and a feeling.

An episode is not stored as a sentence. It is stored as a bundle: where you were, who was with you, what the light was like, what you were worried about that week. When we were designing Ember, Tulving's idea sat at the centre of everything. A journal entry is an episode. And an episode is held together by its context.

A memory opens with the right key

A year after that, Tulving and Donald Thomson described what they called the encoding specificity principle. Put simply: a memory comes back most easily when the cues around you now match the cues that were there when it was made. The smell, the place, the company, the mood.

One of the loveliest demonstrations came in 1975. Duncan Godden and Alan Baddeley asked divers to learn lists of words, some on the beach and some underwater. Later they tested them, again in one place or the other. The divers remembered noticeably more when they were tested in the same place they had learned. Their memories were waiting in the water.

This is why returning to your old school, or hearing an accent from the town you grew up in, can flood you with things you had not thought about in years. You are holding the right key again.

Who, what, where and, least of all, when

The most patient study of everyday memory may be the one the Dutch psychologist Willem Wagenaar ran on himself. For six years, starting in 1978, he wrote down one or two events from each day, about 2,400 in all. For each he noted four things: who was there, what happened, where it happened, and when.

Later he tested himself, giving himself one clue at a time and trying to recall the rest. What he found surprised him. What was the strongest clue, followed by where and who. When, the date, was by far the weakest. On its own, a date brought back almost nothing. Combine two or three clues, though, and memories he thought were gone came back. Some he could recover only when someone who had been there filled in a detail.

That small finding says a great deal. The date is the way a calendar remembers. It is not the way we do.

A life is stored in layers

So if not by date, how is it all arranged? In 2000, Martin Conway and Christopher Pleydell-Pearce proposed what they called the self-memory system. Their idea is that what we know about our own lives is layered, like a set of nested rooms.

• Lifetime periods. The broad chapters: "when I lived in Pune," "my first job," "the years with Alex."
• General events. The trips, routines and stretches inside a chapter: "our holiday in Goa," "Sunday lunches at my aunt's," "the month I was ill."
• Specific moments. The fine detail at the centre: the sound of rain on the tin roof, the joke someone made on the train back.

When we remember, we usually move down through those rooms. You think of a time in your life, then a holiday inside it, and then a single afternoon opens up, with a whole world around it: the people, the weather, the song that was playing. You rarely jump straight to the afternoon. You arrive there.

Before this, after that

Ask someone when something happened, and listen to the answer. It is almost never a date. It is "just after we moved," "the winter before the baby came," "around the time Dad retired."

The psychologist Norman Brown has spent years studying this. His work suggests that we organise our past around transitions: moves, new jobs, births, losses, the moments when the fabric of daily life changed. In one study, people who had lived through the war in Bosnia placed ordinary memories as before the war, during it, or after it. The big event became the ruler they measured everything else against.

Ember's own research leaned on a related idea from Hengchen Dai, Katherine Milkman and Jason Riis, who showed that temporal landmarks, like a birthday or the start of a new year, feel like dividing lines in a life. We use landmarks to start things. We also use them to find things again.

We remember in circles of people

There is another way we file our lives, and it is social. The sociologist Erving Goffman described how each of us plays a slightly different self in each circle we move in: one with family, another at work, another with old friends from college. Those circles are not only roles. They are places memory lives.

"Something happened with the people from work, I think in the spring." "It was one of those evenings with the college group." Often we cannot name the day or even the exact person, but we know which world it belonged to. That is enough to narrow a whole life down to a single shelf.

A web, not a list

In 1975, Allan Collins and Elizabeth Loftus described memory as a network of linked ideas, where thinking of one thing sends a little activation along to everything connected to it. Think of a beach, and sand, sunburn and a particular friend all light up slightly, ready to be reached.

That is why remembering so often feels like following a thread. The beach leads to the friend, the friend to the year, the year to the flat you rented, the flat to a conversation you had forgotten you had. No single cue holds the whole memory. Each one leads to the next.

It also explains the strange half-remembering we all know. In 1966, Roger Brown and David McNeill studied the tip-of-the-tongue state and found that people who could not recall a word could often still say its first letter or how many syllables it had. Memory gives us pieces before it gives us the whole. A missing piece is not a failure. It is simply how the web is shaped.

Every memory is rebuilt

There is one more thing worth knowing, and it is gentle once you accept it. In 1932, Frederic Bartlett showed that we do not replay memories like a recording. We rebuild them each time, from the pieces we have, filling the gaps with what seems likely.

So if you cannot remember what month it was, or which friend was there, you are not losing your mind. You are remembering the way every person always has. The pieces that are there are real. The gaps are part of being human.

Why a search box feels so wrong

Now picture how most software asks you to find a memory. Type the exact words you used. Pick the right date. Scroll a single line of entries from newest to oldest.

That is the opposite of everything above. It asks for the weakest cue first, the date. It expects the exact phrase, when memory gives you only the gist. It assumes you have every piece, when you might have only one: a place, a face, a season, a feeling.

George Miller's famous work on working memory, later refined by Nelson Cowan, suggests we can only hold about four things in mind at once. When you are reaching for a memory, those four slots are full of fragments: Goa, rain, Priya, the year after college. A good way to search would let you hand those fragments over just as they come, in any order, and trust the rest to follow.

Writing it down is part of remembering

Evan Risko and Sam Gilbert, who study what they call cognitive offloading, argue that putting memory into the world, in notes, calendars, photographs and diaries, is not a weakness. It is one of the most natural things a human mind does. We have always kept our memories partly outside our heads.

And reading them back matters. Research by Tim Wildschut, Constantine Sedikides and their colleagues has found that nostalgia, remembering meaningful moments with others, tends to make people feel more connected, and gives a stronger sense that life hangs together and means something. Memory is not only about the past. It is how we know who we are, and who we have been with.

What this means, quietly

If you take one thing from all of this, let it be this. You do not need a perfect memory. You need one thread. A place, a face, a season, a song, the thing that happened just before. Follow it, and the rest has a way of coming back.

And if it does not, that is all right too. Some moments come back on their own, years later, on an ordinary afternoon, when the right key finds you.

How search in Ember follows the same path

Everything above is why search in Ember works the way it does. We built it to match how memory actually opens, not how a database does.

• Type the fragments you have, in any order. "Priya goa rain," "dad summer 2019," "college group last winter." There are no special words to learn. Ember reads each piece and shows you how it understood it, as small labels under the box: a person, a place, a time, a circle. If it read one wrong, remove that label and the rest still holds.
• Exact about who and when, forgiving about what. People and times are matched precisely, so the right Sam is the right Sam. The words you remember are matched loosely, so a typo or a different form of a word still finds the entry.
• Time the way you say it. "Last summer," "three years ago," "around 2019," "the 90s," "Christmas," "Diwali," "Eid," or "Sarah's birthday." You can also reach for what came before or after: "after the Lisbon trip," "during the move," using the stories you have written.
• Places, at any size. A town, a country, or a whole continent. "Japan" finds the evening in Kyoto.
• The texture around a moment. "Rainy," "snow," "late night," "Sundays," "photos," or the name of a song or an artist you were playing that day.
• Answers shaped like memories. Search for one person and you get them: every moment you wrote down together, in the order it happened. Search for a stretch of time and Ember walks you through it, month by month, with the people you saw most.

All of it happens on your device. No AI reads your entries to do this, and search runs entirely on your device. For now, the smart reading of fragments works in English. Searching for the words themselves works in other languages too.

You can read more about the idea behind it in what is a life archive, or in the search guide.