I didn’t feel like going back into the game for the rest of the weekend.
I couldn’t wrap my head around the idea that SparkleCo would tailor and release an adventure just for me. It was entirely possible that it was a new first player module that was being beta tested on a subset of players, though; the Dauphine might have told two dozen players tonight that they alone were chosen to go on her expedition, and we wouldn’t know about each other’s quests. I scanned the forums—no one had reported the Dauphine missing from the desert, which made sense, because her appearance there was always unpredictable. But certainly no one had reported her sudden appearance anywhere else in the blingdom either.
Ordinarily, I would have tried DMing the Keeper of the Moonlight Prism to chat about this whole situation. Except now that I knew the Keeper was Jordon Connelly, I suddenly felt awkward about that relationship. In the game, we were pretty chummy, had each other’s backs in a fight; as a client of Jenning & Reece, she was basically one of my bosses.
I needed something to distract me since my original plan to game all weekend was no longer appealing. I spent the weekend learning power morphemes instead.
My second week in the lab at Jenning & Reece was a blur. Olivia recognized I had an aptitude and proceeded to take advantage of it. “Sympathetic intonation” was the first major component of power morphemes that made sense to me. The next one on the list was something Olivia called “phonetic overtones.” It was a term she made up, not a concept I could find on any linguistics wiki pages or whatever, rather something quite specific to how power morphemes worked.
Power morphemes really leaned on the concept of double meaning to work. Morphemes can have double meaning in the sense that they might mean one thing in one context and a different thing in a different context. Or like—if you say “sheep” you could mean a single sheep or plural sheep and the word sheep is just wacky like that. But sheep has, what, three phonemes in it, and what if all of those phonemes were coded with their own additional meanings? And then you managed to add a layer of sympathetic intonation on top of the whole resulting morpheme that resonated to add extra layers of meaning to both the morpheme and the individual phonemes?
An analogy might be—think about overtone singing, where people can seemingly generate multiple pitches at the same time. Really they’re just generating one tone with their vocal cords, but then using the shape of their mouth to create a harmonic overtone at the same time. Like shooting light through a prism to get more colors. Now imagine when you pronounced a power morpheme, you also created overtones—but overtones of meaning, not just sound. Now add in sympathetic intonation—imagine your intonation affected each tone individually, as well as the whole stack of tones as a group. You might at that point express several layers of meaning with a single phoneme, before you ever got to an entire morpheme.
Now, for the final twist, imagine that power morphemes required you to pronounce multiple distinct phonemes at the same time. That’s the part that was truly hurting my brain, and my vocal cords. The overtone often needed to resonate as a completely different phoneme from the base phoneme, and you often needed to pull that off for several phonemes in a row to get a full morpheme that was truly denser with meaning than it should be.
It was all pretty ludicrous.
After all, how could that much complexity of meaning in your pronunciation even decode properly in the receiving mind? You couldn’t walk up to someone who didn’t speak French and just speak power French at them and get any comprehension.
So what else was happening here?
I learned faster than Olivia expected: twenty-four power morphemes by the end of my second week. I gently wondered aloud if my learning might be accelerated even more if she shared Maddy’s notes with me, which to my surprise actually convinced her. She gave me access to an archive on one of the lab’s workstations and showed me how to navigate. And as I began browsing the treasure trove of data, she shared with me the story of why Maddy was no longer with the lab.
“She was consulting for the Department of Defense when I found her,” Olivia said, “working on nondestructive methods to defeat voiceprint security systems. I don’t think I was supposed to know that, but sometimes I could get her to loosen up. The project we handed her was: here are a set of unnatural sounds that can alter human behavior in unpredictable ways, help us understand what they are and how they work. The project was catnip for her.
“She set up a deep neural network, intending to find commonalities across all one hundred and eight power morpheme recordings, looking for factors and attributes that were distinct and measurable, and attempting to catalog what made each one unique. She also had Alexander record hundreds of ordinary morphemes to feed the neural net as mundane examples for comparison, so that the neural net could learn the difference. This wasn’t going to tell her what any of the hundred and eight meant; it wasn’t going to magically tell her how they worked. It was just barely going to give her the start of a vocabulary to describe them.
“But that’s often how you do things: you peel the onion layer by layer looking for answers. You train algorithms on data sets starting from scratch, establishing an initial baseline and then, depending on how many layers you’ve deployed in your neural net, eventually the algorithm starts emitting guesses about new or similar data that you feed it. Then a human grades the results and starts the process over, this time with a theoretically more accurate baseline for the neural net to work from; and you can do this many many times, depending on available processing, or how robust the data sets are that you’re analyzing or comparing.”
“How could you determine the initial baseline?” I wondered.
“Alexander contributed key insights based on his subjective understanding of what was physically required to express power morphemes out loud. I mean, you don’t simply hear sounds with your ears; your brain formulates an interpretation of what you hear that might not correspond perfectly to the specific soundwaves your ears detected. In this case, the neural net could identify acoustic differences in the recordings, but then Alexander would step in and analyze how the components identified by the neural net correlated to his actual cognitive experience, and slowly they could build weights for each attribute. These were by definition very subjective measures, but they had no other way to develop a classification system. So Alexander was definitely integral to all of it, the whole project, until…”
Pour one out for our fallen comrade, Alexander Reece.
“Without Alexander to guide the project, Maddy … took her research in an unplanned direction. She set out to discover new power morphemes. She could iterate through an enormous number of possible combinations of attributes in search of candidates that might actually work like one of the hundred and eight. She found four that we know of before she vanished.”
“You mean she quit?”
“No, I mean she’s officially a missing person.”
“Seriously?”
“Bradford pulled in favors to try to find her, but she hasn’t left a trail.”
I couldn’t resist asking, “So what do the four new ones sound like? What do they do?”
I was a little surprised by the very sharp look I received in response.
“What, you weren’t curious?” I said. “If you’ve got software on your hands that can generate new power morphemes, that’s like—free money or something. What’s stopping you from taking a closer look?”
“Alexander delivered the hundred and eight to us as a holistic library,” she replied, “and proved by his own example they were relatively safe to handle. We have no reason to assume that any ‘synthetic’ power morphemes will be equally safe.”
“But you still have her neural net running in the lab?”
“I’ve repurposed it. We’re designing full sequences of power morphemes now. The software is modeling combinations of the hundred and eight, and predicting which sequences should be useful.”
“Useful for what?”
“Alexander wanted to call this inevitable avenue of research ‘advanced persuasive tactics,’” she replied. Then she smiled and said, “I prefer the term ‘combat linguistics.’”
My face must have lit up like I’d been given an early Christmas present.
“Oh, of course, I don’t mean literal combat,” she quickly clarified. “The marketplace of ideas is inherently adversarial. The best ideas don’t simply survive based on merit. They must cut down opposing ideas with no mercy in order to gain competitive advantage.”
Uh-huh. She could say that—we were working for an advertising agency, after all—but I was pretty sure that she really did mean literal combat.
Then it hit me. Olivia might have repurposed the software running in the Jenning & Reece lab. But if Maddy was able to clone it on her way out the door … If she’d archived all those recordings of Alexander somewhere outside the firewall … And if she no longer had any “adult supervision” from Olivia to distract her … She could be up and running with her own instance of her neural net. She might have more than four brand new power morphemes by now.
Maybe Olivia was developing “combat linguistics” because she knew someday she might face Maddy again in “the marketplace of ideas.” And maybe when that day came, Olivia expected me to have her back somehow.
I suspected I might be crossing a line to ask about any of this directly. But after falling silent for several moments to let me process what she’d told me, Olivia suddenly plugged a gap in the story for me.
“Maddy didn’t quite vanish without a trace,” she said. “After what would be our final argument about her new research, I didn’t see her for a couple days, and then she sent me an email. A warning not to follow her. Apparently she had placed her trust in … new friends.”
“Like … one of your competitors?”
“Jenning & Reece has no competitors in this arena,” Olivia said quickly. Then she sighed and said, “But it would not surprise me to learn that we’ve acquired enemies.”