How to reconstruct dead languages
A step-by-step guide
Most of the everyday words that we use to describe our day-to-day, ordinary experiences are extremely old: things like food, water, hand, eye, friend, love, and think.
All of these words have been around in some form for thousands of years, and have been handed down from one generation to the next over millennia, changing — sometimes subtly, sometimes dramatically — along the way.
What if you could rewind those changes and recover how these words sounded thousands of years ago, and even hear the sounds of languages that have vanished from the face of the earth?
It might sound like a superpower, but it’s actually a surprisingly accessible science, and you don’t need a PhD in linguistics to come up with theories about what dead languages sounded like.
The basic tools are simple: patience, pattern recognition, and a stack of dictionaries. In fact, the principles of reconstructing dead languages can be grasped by anyone willing to sit down and think systematically about how language changes. It also helps if you like solving puzzles.
Today, I’m going to show you exactly how it’s done, using one of the great success stories of historical linguistics: the reconstruction of Proto-Germanic, the common ancestor of English, German, Dutch, Swedish, and all their close relatives.
By the end of this article, you’ll understand not just what Proto-Germanic sounded like, but also how we arrived at this knowledge. And you’ll be able to use the same tools to reconstruct whichever ancient language your heart desires.
What I’ll be teaching you is called the comparative method, and it’s a four-step process that feels remarkably like doing Sudoku with sounds.
You're reading The Dead Language Society. I'm Colin Gorrie, linguist, ancient language teacher, and your guide through the history of the English language and its relatives.
Subscribe for a free issue every Wednesday, or upgrade to support my mission of bringing historical linguistics out of the ivory tower and receive two extra Saturday deep-dives per month.
Before we start reconstructing, a quick announcement: there was a very positive response to the idea of a summer Beowulf book club for paid subscribers, so it’s happening!
We’ll read the whole poem in translation over the summer and meet for three live Zoom sessions, where we’ll compare translations against the original Old English and explore the historical and cultural context of the poem. Along the way, I’ll explain all the deep lore that most readers miss.
If you’re interested, scroll down to the bottom of the paid section to get access to a signup/scheduling form. I’m looking forward to meeting you face to face!
Gather ye cognates while ye may
Imagine that you met five cousins, all descended from the same couple. But not one of the cousins has a photograph of their long-departed grandparents. Would you be able to figure out what they looked like?
Not exactly. But you’d be able to make some good guesses by comparing the traits that tend to run in families. For example, if all five cousins had the same distinctive nose shape, one (or both) of the grandparents probably had it too.
Linguistic reconstruction works in a similar way. When we compare words across languages, we often find similarities. Sometimes these similarities are accidental: they arise purely by chance, like English name and Japanese namae ‘name’. No one thinks English and Japanese are related just because of that one word.
Other similarities exist due to borrowing: languages pass words as freely as kids share germs in their first week at preschool.
But when these similarities between two languages are systematic — that is, the similarities between the languages are so predictable, you could describe them with an equation — it points to a shared inheritance from a common ancestor. From there, we can work backwards to tell what that common ancestor was like.
Let’s look at an example. Here are some words for ‘hand’ across Germanic languages, along with their pronunciations in the International Phonetic Alphabet — when you see symbols inside square brackets like this [x], that means we’re dealing with a pronunciation rather than a spelling.
We need to look at the pronunciation rather than the spelling, because spelling can often deceive us as to how words are actually pronounced.1
English hand [hænd]
German Hand [hant]
Dutch hand [ɦɑnt]
Swedish hand [hand]
Gothic handus [handus]Don’t worry too much about what the symbols mean at this point: most of them are intuitive, but some are a bit strange. I’ll explain them as they come up.
The family resemblance between each of these words is obvious, but just like in a family of people, each member of the hand family looks a bit different.
All the languages but Dutch begin their word for ‘hand’ with a sound written [ɦ], which is very close to the [h] sound English has.2 They have other similarities: in the middle of the word, all five have an [n] sound. And right after that [n] sound, each language has a [d] or a [t].
What we’ve found here are linguistic correspondences. Where English has [h], so does German, Swedish, and Gothic — and Dutch has the very close sound [ɦ]. Where English has [n], so do the other four. Where English has [d], so does Swedish and Gothic — while German and Dutch have [t]. The existence of these correspondences is a clue that leads us to believe these languages might be related.
But here’s a question: how do we know these languages are actually related, in that they descended from a common ancestor? Couldn’t these similarities be from borrowing?
What if one language had a word for ‘hand’, and the other languages just borrowed that word? This is, after all, what has happened in the modern world with words like sushi and pizza. Just because English, Italian, and Japanese all have words for sushi and pizza doesn’t mean they’re related.3
This is why we look for systematic correspondences, because we need to make sure our correspondences hold across the language as a whole.
To do this, we need to find another word that’s similar in sound to hand: let’s use hound. If we end up seeing the same correspondences that we saw with hand, we’ll be more confident that we’re dealing with truly related languages. So what do we have for hound?
English hound [haʊnd]
German Hund [hʊnt]
Dutch hond [ɦɔnt]
Swedish hund [hʊnd]
Gothic hunds [hunds]Voilà. We see the same correspondences for the [h…nd] sequence in hound as we saw for hand!
Now, in a real reconstruction, you’d go farther and check more words than just two. But, for our purposes, this is enough for us to be satisfied that we’re dealing with systematic correspondences, which means genuine relatedness: descent from a common ancestor.
Lining everything up
But before we move on to figuring out what that common ancestor looked like, let’s take stock of what we’ve done.
We have collected related words across the languages we’re comparing: these related words are called cognates, and the groups of them together are called cognate sets.
Note that collecting cognates can sometimes be a bit of a subtle art: the words that we collected as part of the hound cognate sets are all better translated as dog. In English, a hound is a particular kind of dog. But it’s clearly related to dogs, so even though it’s not an exact translation of the other words in the set, hound is fair game to put in our cognate set.4
But sometimes words have changed a lot over time, either in meaning (English starve is cognate with German sterben ‘die’) or sound (English wheel is ultimately cognate with Sanskrit cakrám — the word that gives us chakra).
So, as you can see, identifying cognates is often an iterative process: as you work to reconstruct the ancestor language and understand how the daughter languages have changed, you may realize that there are more cognates than you had originally thought.
At any rate, that is Step 1 in the comparative method: Collect cognate sets.
Step 2 is one we have already started: establishing sound correspondences, or, in other words, aligning the sounds in the cognate sets and identifying the correspondences between them. So far, we’ve come up with (in the order English, German, Dutch, Swedish, Gothic):
h h ɦ h h
n n n n n
d t t d d
We’ve ignored the vowels so far, but we could do the same for them. Here’s what it looks like, for completeness’ sake:
æ a ɑ a a
aʊ ʊ ɔ ʊ u
I’m not going to deal with the vowels in detail today, since it would take a long time to describe all the vowel sounds and their changes. Vowels in the Germanic family are notoriously slippery and prone to change. But the same principles apply both to vowels and to consonants.
We also had one other difference we haven’t yet discussed: in Gothic, we saw endings on both of the words handus ‘hand’ and hunds ‘dog.’ So we need to augment our correspondence tables, this time using the special symbol Ø to indicate that a sound present in one language is absent in another. The situation with the Gothic endings looks like this:
Ø Ø Ø Ø u
Ø Ø Ø Ø s
We have one instance where Gothic [u] corresponds to nothing in the other languages, namely the [u] in handus [handus]. And we have two instances where Gothic [s] corresponds to nothing in the other languages, namely the final [s] in handus and hunds.
Unfortunately, we now have a problem. But it may be difficult to spot.
The problem is this: what does Gothic [u] correspond to? In the hound cognate set, the [u] in Gothic hunds [hunds] corresponds to English [aʊ] (the ou in hound), German [ʊ] in Hund [hʊnt], and so on. But in the hand cognate set, the [u] in Gothic handus corresponds to nothing in English hand, nothing in German Hand, and so on.
This is a problem because we want our correspondences to be systematic. If a given sound in one language can correspond to multiple different sounds in other languages, the correspondences are no longer systematic. Chaos reigns: dogs and cats living together, mass hysteria.
We don’t want that. But what can we do about it?
Stress management
Here’s where some knowledge of how language works in general — and how Germanic languages work in particular — comes in handy.
In many languages, including all the Germanic languages, syllables come in two flavours: accented and unaccented. In Germanic languages, the nature of this accent is one of stress. The accented (or stressed) syllable of a word is said louder, lasts longer, and generally has a higher pitch than the unaccented (or unstressed) syllables.
Think of the English words apple, window, and factory. All of these words have their accent, or stress, on the first syllable. This pattern of word-initial stress is actually a very old pattern in the Germanic languages. For the purposes of this article, let’s just assume this to be the case — we could prove it from first principles if you were willing to read for another 10,000 words.5 But… let’s not, for all of our sakes.
The existence of a difference between stressed and unstressed vowels gives us a wedge which we can use to separate the [u] in hunds from the [u] in handus. The [u] in hunds is stressed, while the [u] in handus is not!
So we can say that Gothic stressed [u] corresponds to [aʊ ʊ ɔ ʊ] in the other languages, while Gothic unstressed [u] corresponds to [Ø Ø Ø Ø] in the other languages.6
We can rewrite the correspondence sets like this:
h h ɦ h h
n n n n n
d t t d d
æ a ɑ a a
aʊ ʊ ɔ ʊ u (stressed)
Ø Ø Ø Ø u (unstressed)
Ø Ø Ø Ø s
And now our correspondences are systematic once again. And we’ve successfully completed Step 2: Establish sound correspondences.
This whole process works because sound change often depends on the environment the sound is in. Think of how ‘r’ was dropped (in non-rhotic dialects of English) at the end of a word, or before a consonant, but not before vowels.
Now that we’ve done all that ground work, we’re finally ready to reconstruct our first form in the ancestor language!
The fun part
It’s time for Step 3. Reconstruct proto-forms.
Keep reading with a 7-day free trial
Subscribe to Dead Language Society to keep reading this post and get 7 days of free access to the full post archives.