Hashtags for Humans

Humans are pattern matching machines. The world around us is vast and incomprehensible, but we are able to make some sense of it by organizing messy reality into neat categories. We dismantle complex concepts and individuals into their constituent traits, to match them to prototypes, archetypes and stereotypes.

Categories and hierarchies are built into our culture at every level. When we give names to abstract concepts; when we say these are dogs and these are rocks and these are songs, we are categorizing. We categorize ourselves, too: categories of gender are baked into nouns and pronouns, and family lineage — literally, “the patriarchy” — is written into our names.

Categories are rules for identity. The only thing we love more than making rules is making rules for making rules, so we make formal systems to manage our hierarchies.

These systems reflect the times in which they are created. Linnaeus’ hierarchy of life — kingdom phylum class order family genus species — is an echo of king prince duke count baron knight peasant. What would a taxonomy look like if we created it today? Allow me to thought-lead for a moment:

The information age promises the democratization and decentralization of media, with news and ideas and conversation spreading across the world in real time. Hierarchy and linear thought have no place in cyberspace. In the world of tomorrow, information is coming in from every direction at once — its a network where every node is connected.

How do we organize knowledge in 2015? The hashtag.

First, let’s get our terms straight. Many of us already have a pretty good idea of what a hashtag is, but I want to focus on what I consider to be the three essential properties of hashtags:

1. Hashtags are relational. An item in a collection can have multiple tags, and tags can have multiple items. Hashtag membership follows the same rules as mathematical sets.
2. Hashtags are vernacular. Any string can be a hashtag, and a hashtag’s meaning comes entirely through usage. There is no authority that decides what a hashtag can be or what content can be tagged with it.
3. Hashtags are conversational. A hashtag isn’t just collection; its a channel of communication. One participates in and responds to a hashtag.

Relational

Let’s start with tags as a categorization system. The concepts of relational categorization predates the web and even computers by hundreds of years. I don’t have time for quite that much background. Instead, let’s go to the library.

Libraries are kind of like an analog database, right? They contain information, and they organize it in a way that you can find it. But books take up physical space. A book needs to be somewhere. If you have to put a book somewhere its often helpful to put it with other books of the same kind — then you don’t need to know exactly where every book is, just how to find the category.

If you’re the super-organized type, like I imagine most librarians are, you’d organize your categories into a hierarchy. At the top of the hierarchy you have broad classifications like science, philosophy, arts; below them you segment into finer-grained sections like geology, ethics, sculpture. You keep subdividing into smaller and smaller sections until you have the right place for everyhthing in your collection.

The problem with this is that you can only impose one hierarchy at a time; it must necessarily be compromised. If you want arts and technology to be top-level classifications, that means that architecture and structural engineering books could be on opposite sides of the library. What does one do with topics that don’t fit into your hierarchy?

Library of Congress Subject Headings

Create a system of categorization on top of your hierarchy, with a vocabulary of keywords for attributes of the item that you couldn’t express in the hierarchy. For example, the Library of Congress Classifications is the hierarchy that dictates where books are located. The Subject Headings map topics to the different branches of the classifications hierarchy where that topic is discussed.

This indirection is a necessary inconvenience in a physical collection — an item in that collection needs a canonical address to prevent duplication. But indirection is how everything works on a computer. Web developers don’t need their data to be stored in a way that makes sense to them as long as it makes sense to the computer. In other words, we do not need to impose a hierarchy on our data if it doesn’t make sense.

Hierachies were designed to be searched by hand, and on foot; keywords are meant to be searched by computer.

So I’ve been discussing keywords and set theory in sort of abstract terms. Let’s make it a little more concrete.

game cards do not actually talk

There’s a children’s game called Guess Who?, where players take turns asking yes-or-no questions about the characters, like “do they have brown hair?” or “are they wearing glasses?”, in order to figure out who their opponent has picked. It’s like 20 questions for people without imagination.

Well that’s not fair. I thought this was trememdous fun in 1991. (Plus I think I had a little crush on Maria.) But the skills it taught me were how to judge people and how to do an advanced Google search.

When you’re asking a succession of yes-or-no questions as you do in this game, you’re filtering a dataset with a series of logical predicates ANDed together.

This is how we could express our questions this in javascript:

people.filter((p) => p.tags.blond && p.tags.glasses)

And this is one way to say that in SQL:

SELECT * FROM people WHERE tags LIKE “%blond%” AND tags LIKE “%glasses%”

We’ve been looking at this in terms of the people; let’s flip our perspective around and look at this in terms of the traits. Before we would say that Anita, Charles, Eric, David and Joe all have blond hair; instead, we’ll say the set of people with blond hair includes Anita, Charles, Eric, David and Joe. The set of people with glasses are Claire, Joe, Paul, Sam and Tom. The intersection of blondes and people wearing glasses is Joe.

What does this have to do with hashtags or humans? Well, when we see tags in the wild, we tend to see them in clusters — Twitter’s length constraints prevent exhaustive tagging, but most sites do not, and posts will have many tags. Likewise, in many cases a tag will be associated with many posts. There’s a many-to-many relationship between posts and tags. This is the same relationship model Guess Who? uses for people and their traits.

Vernacular

Let’s fast-forward a little bit to the early 2000s, the Web 2.0 Years. The internet had always been a social place on Usenet and BBSes, but the direct ancestors of today’s Social Media were springing up in this era.

The library model doesn’t quite work for the social web. In a typical library, the hierarchy and the keyword vocabulary are fixed, and the collection is curated by the librarians. A patron cannot add books to the collection, nor can they alter the classifications or keyword vocabulary.

But the photos on Flickr, or the links on del.icio.us belong to the community — so why shouldn’t the vocabulary of categorization also belong the them? And thus the rise of user-generated taxonomy, or “folksonomy”, became one of the defining characteristics of the Web 2.0 Era. Not only had we democratized information, we had democratized the organization of information.

Also, we stopped calling them “keywords” — probably because of shameful associations with AOL — and started calling them tags.

This is a typical metafilter post from 2002. See the different kinds of information that the tags cover: Tim Cavanaugh is the author, and warblogs are the subject. #brokenlink is a site-wide convention for posts with broken links — the tags on this post are used both for the content of the linked article and to describe the state of the link itself.

The freeform nature of tags let us categorize via neologisms — “blog” first appeared in 1999 and “warblog” was coined three months before this post — a decentralized vocabulary can include new concepts much more quickly than a centralized one. If we needed an approval committee for tag terms, warblogs would have become totally irrelevant by the time they were added to the tag vocbulary.

On the other hand, a centralized system can have rules for dealing with synonyms and inflections — a controlled vocabulary, once it were able to reference blogs at all, wouldn’t have distinct terms for #blog and #blogs. Nor does a controlled vocabulary have namespace collisions: posts tagged with #football will either be about soccer or hand-egg, but not both. A decentralized system can only hope its users figure out some conventions and choose to follow them.

One solution for this is an ontology that includes both a fixed vocabulary of categories, often mapping to classical hierarchies, and an open vocabulary of non-hierarchical tags. This is the approach that’s used in wordpress blogs and ask metafilter questions.

This is also the approach taken by browsers. We have a fixed vocabulary of general-purpose tags — some of them have special functionality but most of them are just common classifictions of containers. But we also use CSS classes to create our own terms with meaning specific to the site. Every element on the page must have one and only tag type, but they can have as few or as many classes as they want.

Another common solution is to have user-defined tags as the sole form of taxonomy, but encourage users, through the UI, to prefer existing terms over novel ones. This is the preferred system for newer networks that do not have distinct categories and tags, like Twitter and Tumblr.

There’s one particular use of this that I want to explore in depth: Facebook’s gender selector. Before this came out, most of the gender options I had seen were either strictly male/female or freeform text fields. A few offered more choices.

This particularly progressive social networking site offered these options for gender. This is a lot of options, but it reveals a paradox: the more options you provide, the more you appear to be excluding — if genderqueer can make the list, why not pangender? Should you use a term that used to be correct but is now archaic?

It’s kind of a mess from a UI perspective, too — there are a lot of items and they don’t seem to be in any particular order. There are redundant and ambiguous items that would work better if you could combine terms.

But when Facebook adopted the tag-based gender options, they didn’t just drop in a jQuery library and call it a day; there are specific design choices being made in this form. Here are a few of them:

• It cannot be empty. You have to select at least one option, though that option may be one of several variations on genderless
• While most tag fields are abstract and can end up like junk drawers for metadata, tags in the gender field have inherent context, even if the meaning is not known.
• The form reflects the subject matter: gender is simple for many, but complex for others. Therefore it has multiple tiers of defaults: male / female, known tags, and freeform tags.
• It has a separate (limited option) selector for pronouns — this is the only part that has any implications for the UI.
• There are a lot of options that seem to be redundant. This is intentional. These tags are being used for personal expression as much as they are for selection.
• It wasn’t initially freeform; that was added later. It launched with over 50 gender options but even that was not sufficient. 50 is more than enough options for many people, but its quite limited if you’re thinking of it as the vocabulary for expressing the nuance of how one exactly relates to society’s expectations of masculinity and femininity.

A third approach is that you can use human moderators to enforce community standards. One can do this with a hired staff, but in many cases its more efficient to harness the internet’s limitless supply of pedantry and deputize users as moderators.

We’ll take it as given that community management is necessary for dealing with spam and trolls. Sites like Instagram that have policies against nudity and illegal activity blacklist tags like #pussy and #weed from search results. (They also block #🍆 for similar reasons.) Things get a little more complicated when you regulate for quality.

Knowledge bases like stack overflow and wikipedia depend on their posts being well organized for both editors and readers, and managing metadata is as much a part of that as managing content. Editorial policy isn’t just normalizing #blogs and #blogging to #blog, it also involves choosing what information is purposefully not captured. This post is discussing why qualitative tags like #best-practices don’t fit the mission of stack overflow, and why they have been removed. They’re not imposing a system as rigid as a formally defined vocabulary, but they’re far from the anarchy of pure folksonomy. These sites compromise expressivity to provide more important features.

But in many cases freedom of expression is the feature. Twitter could do a lot to improve the quality of discourse on its service but removing subjective hashtags ain’t it. The restrictions that are useful for a purpose-made knowledge base are detrimental to a general means of communication.

Conversational

We’ve gotten this far and we still haven’t answered the biggest question of them all: how did the hashtag get its hash?

A little background: IRC is a chatroom protocol that predates the web. Rooms — or ‘channels’ — are marked with a hash symbol and are intended to act like categoies: a single, long-running discussion with real-time participation. IRC channels are vernacular and conversational, but they are not relational; there is no system for cross-referencing posts across channels.

When hashtags came to Twitter, they began as conventions that would only later be proper features, much like retweets and mentions. This meant that early hashtags functioned like keywords for Twitter’s existing search system, which was completely agnostic towards search terms.

There was a whole syntax for messaging and subscribing to tags proposed, but it was never implemented; nor did Twitter ever distinguish between posts starting with tags vs those containing them, as they would with replies and mentions.

Twitter’s benign neglect allowed the hashtag to adapt to the unique nature of the network: because Twitter combines the real-time conversational aspects of IRC with the authorial collection aspects of blogging, it needs systems for both grouping conversations into a channel and tagging posts with metadata. Hashtags combined these features into a single idiom and achieved both.

A hashtag is more than the sum of its tweets. Tags at their best (and their worst) are entities in their own right — there is as meaning in the connections as there is in the content. People talk about “posting in this tag” and tell each other to “get out of this tag” because tags are shared spaces, and tagging is inherently a group participation. Unless you take pains to identify the tag without naming it, e.g. “👾🐊”, commentary and discussion about a tag also becomes part of the tag.

When the digerati of 2007 were first exploring the modering hashtag, there was an understanding that it wasn’t going to be limited to any single site. This was the heady days of microformats and the semantic web; my forebears in thought-leadership had dreams of tags providing the metadata to unify human and machine interfaces across the whole internet.

That’s not quite how it worked out. Hashtags couldn’t live up to their utopian vision, but they’ve become much more a part of society than they could have imagined. Tags are not only a means of conversation but now also part of conversation.

Because there are no rules for tags, our usage can reflect all kinds of purposes — we develop idioms for hashtags. They’re used as interjections, parenthetical asides, and as a means to #mock the #overuse of #hashtags.

This is all powered by the duality of hashtags as a collection of ideas and the channel for broadcasting them. I would bet that very few of us participate in any hashtags on a daily basis, but everyone here is familiar with them as a presence in society.

Tags can be trite (#TheDress), and tags can be meaningful (#BlackLivesMatter), but in either case, tags are more than their content and more than their context. Tags can be the signifier for massive social movements, whether they be 24 hours of civilization-paralyzing mania or something that, you know, matters.

For all the ways that tags can unite people around a common idea, they’re not an unmitigated force for good. Any woman on the internet can tell you that.

But lets go back to the beginning. Humans are pattern-matching machines. The world around us is vast and incomprehensible, but we are able to make some sense of it by organizing messy reality into neat categories.

In order to do this, in order to make sense of the world, we need to have a system of categorization that works for the world we’re in. And the systems of hierarchy and clustering that we’re relying on now are insufficient.

We’ve always intuitively understood tags and traits — children don’t have to be taught how to read the Guess Who? people’s faces. But hashtags give us a more structured system of organization that maps pretty well to naive set theory. They provide a framework for thinking about ideas like intersectionality — that race and gender aren’t just single values but complex constellations of interrelated attributes and experiences.

And what about kids growing up in a world where tags have always been the predominant paradigm? Will they leave us hierarchy-thinkers behind? Is this like the last chapter of Childhood’s End; should this essay be called Hashtags for Posthumans?

It’s fashionable to mock people who use the idioms of hashtags to define themselves. I will admit that I don’t especially understand it myself; part of why I wrote this was to try to understand it. But I look back on myself at that age and I think of all the ways I felt weird and gross and worst of all, alone in my weird grossness. As an adult I can think a little more clearly about myself without labels but man, I sure wish I had some labels that made sense for my version of gender and my version of sexuality. I know I’m not alone; there are a lot of people who have rounded up to straight, or male, or to any sort of identity because they thought they had to pick one-and-only-one from the standard list.

So in a lot of ways, I envy the generation that’s growing up in a tagged world. Now, I may be a thought leader but I’m not an idiot — hashtags will not save the world. But the dream of the internet is a place where people all over the world can meet others that share their esoteric interests while also exchanging new ideas. As asinine as this sounds, hashtags give us both a way to conceptualize this and a way to make it happen, and I’m feeling optimistic.