👋 Hi, welcome to Signal Lab where the team at Innate Language Processing digs deep into language & behavior, the limits of technologies, and the connection between what people say and do. 100% human authored, never with AI.
Guilty?
Did Adnan do it? Over 300 million audience members are still divided, 12 years after the Serial podcast was released.
Serial centred on the defendant, Adnan Syed, convicted when he was 17 of the first degree murder of his ex-girlfriend, Hae-Min Lee, and sentenced to life in prison. The accounts of neither the prosecution nor the defense fully added up at the time of conviction (or in the podcast), so listeners — like jurors — were left to draw their own conclusions, largely based on whether they believed Adnan or not. That’s what made it so provocative. Serial’s success popularized the genre of true crime podcasts.
Host Sarah Koenig spent over 40 hours on the phone with Adnan during his incarceration (in 15 minute intervals because that’s all Global Tel-Link prepaid inmate calls allow). He said some strange things in some strange ways, like the time he got genuinely angry at Koenig for calling him a nice guy.
“To be honest with you, it kinda- I feel like I want to shoot myself, if I hear someone else say, I don’t think he did it cause you’re a nice guy, Adnan. So I guess kinda, you know, cause you wouldn’t know that, but I hear people say that to me over the years and it just drives me crazy. I would love someone to hear, I would to hear love someone to say, I don’t think that you did it because I looked at the case and it looks kind of flimsy. I would rather someone say, Adnan, I think you’re a jerk, you’re selfish, you know, you’re a crazy SOB, you should just stay in there for the rest of your life except that I looked at your case and it looks, you know, like a little off. You know like something’s not right.
The evidence under our noses
We took Adnan seriously, no more “Mr. Nice Guy”.
We looked at the evidence from a new angle, still focused on Adnan himself, but not so impression-based….
We ran a linguistic analysis of his own statements on the podcast. How did he orient himself to the crime, how did his grammatical choices demonstrate dissemblance or avoidance. What could it all mean?
First, listen for yourself and see what you pick up.
How does a murderer describe the crime?
If we’re pattern matching, Adnan fits right in with the quantitative speech patterns of hundreds of real murderers we analyzed (on a project ten years ago), specifically in their accounts of the crimes that landed them in prison.
Here are three hallmarks:
Convicted murderers demonstrate low expressions of logical, formal thinking during accounts of the crime — lower than the average population by a dramatic measure. Adnan is even lower than the average for convicted murderers. What this sounds like: a story-telling register, lacking in clarity and structure. Lots of pronouns, conjunctions, adverbs.
Convicted murderers show grammatical evidence of hugely suppressed confidence during these accounts (drops to 30% the rate of the average population). Adnan matches the rate for convicted murderers. What this sounds like: heavy first-person-singular uses relative to third person, low affiliation and low command language.
Convicted murderers’ measures of disclosure spike way out of range during these accounts of the crime, and Adnan is the same. What this sounds like: too much insight language, self-reference, and reflective/hedging language.
I just sometimes wish they could look into my brain and see how I really felt about her. And no matter what else someone would say, they would see, man, this guy had no ill will toward her. Whatever the motivation is to kill someone, I had absolutely-- it didn't exist in me, you know what I mean? No one can ever say why.
People could say why. Oh, man, he was mad, this, that, or the other. But no one could ever come with any type of proof or anecdote or anything to ever say that I was ever mad at her, that I was ever angry with her, that I ever threatened her. That's the only thing I can really hold onto. That is like my only firm handhold in this whole thing, that no one's ever been able to prove it.
No one ever has been able to provide any shred of evidence that I had anything but friendship toward her, like love and respect for her. That's at the end of the day, man. The only thing I can ever say is, man, I had no reason to kill her.
But all of these trends, are they simply byproducts of an intense communicative situation? What if they’re topic-driven more than anything else?
So we ran Adnan against a baseline drawn from court transcripts, narrations of crime, just not by perpetrators themselves.
This exercise only affirms the fit of Adnan to the patterns above. The courtroom baseline is even farther from Adnan than the average population was, by huge margins. The alignment between Adnan and convicted murderers isn’t purely topic-driven; it not only survives but is strengthened against a topic-matched baseline. And what’s more, some of these patterns were confirmed in peer-reviewed academic work, comparing convicts by language analysis, placing Adnan not among the innocent, but the guilty….
Exonerated ‘murderers’
We had to more carefully examine an important subset of ‘murderers’ — exonerated ‘murderers’ — just to be sure we weren’t missing something. Broadly, we can highlight two main patterns that differentiate exonerated from convicted murderers.
First, convicted murderers grammatically avoid sensitive words or terms, using strategies like grammatical substitution or ellipsis. Here’s an example, from one convicted murderer:
I may not know everything clearly, but I think I've gained. This man is going to be like this until he dies, and why, because of my hands. And that's not something that is pleasing to live by, and I can't forget that, you know. And I know that no matter what how much work I do, no matter how good I do, he's going to remain the same."
Exonerated murderers, by and large, do not do this. They use real words that match their meanings, not euphemisms or substitutions. The difference is statistically significant, a noticeable discrepancy.
Secondly, convicted murderers (in contrast with exonerated murderers) utilize a a lot of extreme grammatical cases. Scroll back up to the big blue block quote (“I sometimes wish…”) and you’ll count “no” 7 times, “ever” 9 times, “only” 3 times, and “any” 4 times, all in less than 200 words. This is multiples above the average.
Adnan is guilty on both linguistic counts. His speech does not align with exonerated murderers’ accounts of crime.
Why Serial took off
Twelve years ago when Serial aired, there was a lot of hoopla about cell towers, Jay’s testimony, Asia’s alibi, the Nisha call… most of that evidence was unreliable, anecdotal.
Adnan was freed in a March 2025 resentencing.
Reflecting on the series years later, Sarah Koenig said, “people treating [the case] like a puzzle to be solved rather than thinking about the impact on the real people involved who have been through a lot of pain” “felt bad” and that she “felt responsible for a lot of it.”
But what drew Koenig to the case from the start was Adnan himself. In her words, “I see many problems with the state’s case. But then, I see many problems with Adnan’s story too. And so I start to doubt him, I talk to him and talk to him, and I start to doubt my doubts. And then I worry that I’m a sucker that I don’t know. That’s the cycle. Once, about six months after we’d begun our phone calls, Adnan asked me, a little nervously, what’s your interest in this case, really? Why are you doing this? And so I explained all the interesting stuff I’d read, and the people I’d talked to blah, blah, blah. But I also told him really what really hooked me most, was him. Just trying to figure out, who is this person who says he didn’t kill this girl but is serving a life sentence for killing this girl.”
One reason Serial took off is because everyone has someone in their life they can’t be sure quite what to make of, whether to trust them or not. “And then I worry that I’m a sucker that I don’t know.”
That question — how can we know about another person — is ultimately what drew us to this case, too. Not to solve Adnan Syed, or to decide whether someone is telling the truth based on a handful of linguistic clues, but to ask what can be learned from the way a person naturally speaks. Our language is full of signals we rarely notice: emphases, hedges, omissions, descriptions. They are not proof, and not a substitute for evidence, but they are clues.
And if we want to understand people as they actually are — not just as they present themselves — we have to start by listening closely.
