The Twelfth Turn

Keza Paredes had not slept in thirty-one hours, and the predictor on her laptop was still running, and the green line was still holding.

She sat on the cabin floor of the research vessel with her father’s headphones around her neck and the laptop propped against her knees. The hydrophone was in the water. The matriarch was two hundred meters off the bow, her blow a slow white column against the gray Atlantic. Three other whales were with her, and the rest of the clan was spread across a quarter mile of ocean, foraging.

Keza’s predictor model ran on the laptop. It took as input the inter-click intervals of a coda, the number of clicks, the rubato envelope, which was the variance in timing between clicks, and the spectral centroid of each click. It output a probability distribution over the next coda’s form. Not its content. Form. Interval structure and rubato envelope. It did not predict meaning, because Keza had not yet defined meaning in a way a model could use.

She had defined form. Form was measurable. Form was the thing she trusted.

The matriarch’s name was Marengo. SW-07 in the Project CETI catalog. Female, roughly forty years old, matriarch of a clan of eleven that worked the waters east of Dominica. Keza had first recorded her in 2019, off the coast where her father had died. Her father had been a fisherman who had spent three summers on this water with a hydrophone and a notebook and no model at all. He had written in the notebook that the whales were talking to each other and that he did not know what they were saying and that this was fine.

Keza had read that notebook at twenty-three and had spent the next seven years making sure she knew more than he had.

The exchange began at 06:14 on a Tuesday.

Keza had been running a twelve-day trial. Each day, two transmissions. One on-beat, one late. The on-beat transmission went out at 06:14:00, in the matriarch’s opening slot. The late transmission went out at 06:14:04, four seconds past the beat. Same coda both times. Nine clicks. She alternated which came first on odd and even days, to control for order effects.

This was day twelve. The last day. The matriarch had responded to every transmission so far, and Keza had predicted the form of nine of the twelve replies. The predictor locked on nine. Three misses, all in the middle of the sequence, where the matriarch had varied the rubato in ways the model almost caught.

Today the matriarch did not wait for the transmission.

At 06:13:58, two seconds before Keza’s scheduled on-beat slot, Marengo emitted a coda. Seven clicks. Inter-click intervals of 0.42, 0.38, 0.41, 0.39, 0.40, 0.38 seconds. Rubato envelope tight, variance low. Keza’s predictor, running on the previous eleven turns, output a distribution: next coda form, seven clicks, tight rubato, probability 0.81.

The matriarch’s coda matched. Keza felt the small triumph she had felt nine times before, the one that hooked her, the one that made her think the science worked.

Turn one.

The clan stopped foraging. All eleven whales oriented toward the boat. Keza noted the time, the bearing, the distance. Observable behavior. Data she could not not interpret.

She sent the on-beat transmission at 06:14:00. Nine clicks, her standard coda. The matriarch replied at 06:14:22. Eight clicks. Rubato loose, variance high. The predictor had said eight clicks, loose rubato, probability 0.74. It matched.

Turn two.

Keza sent the late transmission at 06:14:04. Four seconds past the beat. The matriarch replied at 06:14:31. Six clicks, tight rubato. The predictor had said six clicks, tight rubato, probability 0.69. It matched.

Turn three.

The clan was still oriented toward the boat. None of them were foraging. Keza wrote this down. She wrote it down every time. It was the part of the data she was least comfortable with, because it meant the whales were doing something in response to her that was not about the codas.

Turn four. The matriarch emitted first, again, before Keza’s slot. Nine clicks, loose rubato. The predictor said nine clicks, loose rubato, 0.77. Match.

Turn five.

The matriarch emitted a coda Keza had not heard before. Not new clicks, not new intervals, but a combination. Seven clicks with a rubato envelope that started tight and ended loose, a shape the predictor had not seen in the training data. The model output a distribution, and the top candidate was a seven-click coda with uniform rubato, probability 0.52. The actual coda was not in the top three.

Keza stared at the screen. The predictor had failed. She checked the input features. She checked the hydrophone gain. She checked the clock. Everything was correct. The matriarch had produced a form the model did not expect, and the clan was still oriented toward the boat, and Keza did not know what that meant.

She sent her standard coda. The matriarch replied. Eight clicks, tight rubato. The predictor said eight clicks, tight rubato, 0.83. Match.

Turn six.

The matriarch emitted first. Six clicks, tight rubato. Match. Then, before Keza could send her transmission, the matriarch emitted again. A second coda, without waiting. Four clicks, loose rubato. The predictor had not predicted a second coda. Keza had not modeled turn-taking this way. The model assumed one coda per turn.

She sat with her hands on the keyboard. She wanted to answer. She wanted to emit something back, a coda of her own, not the standard nine-click test signal but something that responded to what the matriarch had just done. She did not. She sent the standard coda. The matriarch replied. Nine clicks, loose rubato. Match.

Turn seven.

The matriarch emitted. Five clicks, tight rubato. Match. Then a second coda. Seven clicks, tight-to-loose rubato, the same shape as turn five. The predictor did not catch it. Keza noted the repeat. The matriarch was using the new form again. It was not a one-time variation. It was a coda she had added to her repertoire, or had always had and was only now using.

Keza sent her standard coda. The matriarch replied. Six clicks, loose rubato. Match.

Turn eight.

The matriarch emitted. Nine clicks, loose rubato. Match. Second coda. Four clicks, tight rubato. The predictor did not catch the second coda, but Keza, watching the screen, thought she saw the pattern. The second codas were shorter. They were the replies to her replies, or they were something else.

She sent her standard coda. The matriarch replied. Five clicks, tight rubato. Match.

Turn nine.

The matriarch emitted. Seven clicks, tight rubato. Match. Second coda. Eight clicks, tight-to-loose rubato. The predictor did not catch it. Keza noted that the clan had shifted. The whales that had been oriented toward the boat were now oriented toward the matriarch. Not toward Keza. Toward Marengo. The matriarch was addressing someone, and it was not the hydrophone.

Keza sent her standard coda. The matriarch did not reply.

The hydrophone was still on. The boat engine was off. The only sound was the water and the distant blow of a whale and the hum of the laptop fan.

Turn ten.

The matriarch emitted. Four clicks, loose rubato. Match. Second coda. Nine clicks, tight rubato. The predictor did not catch the second coda. Third coda. Six clicks, tight-to-loose rubato. The predictor had no framework for a third coda in a turn.

Keza did not send anything. She sat with her hands on the keyboard and watched the screen and did not send anything.

The clan was oriented toward the matriarch. All eleven whales. The boat was a thing in the water, and the whales were not looking at it.

Turn eleven.

The matriarch emitted. Five clicks, tight rubato. Match. Second coda. Seven clicks, loose rubato. Third coda. Four clicks, tight rubato. The predictor caught none of the second or third codas. Keza’s accuracy for the exchange was now nine of eleven first codas, zero of six second codas, zero of two third codas.

She wrote in her field notes: “The model predicts form of first coda at ~80%. It does not predict second or third codas. It does not predict turn count. It does not predict clan orientation. It predicts the thing I built it to predict and nothing else.”

She sent her standard coda. The matriarch did not reply.

Turn twelve.

The matriarch emitted a coda Keza had never heard. Not a combination. A new form. Twelve clicks. Inter-click intervals of 0.55, 0.50, 0.48, 0.45, 0.42, 0.40, 0.38, 0.36, 0.34, 0.32, 0.30 seconds. Rubato envelope descending, each click faster than the last, a shape that sounded, to Keza’s ear, like a sentence that was running out of breath.

The predictor output nothing useful. The distribution was flat. No candidate above 0.15.

The clan moved away from the boat. All eleven whales, including the matriarch, turned and swam east, toward deeper water, away from the hydrophone, away from Keza, away from the research vessel. They did not look back. The matriarch’s blow was the last thing Keza saw, a white column against the gray, and then the water was empty.

Keza sat on the cabin floor. The hydrophone was still on. She could hear the water. She could hear the distant hum of the boat’s systems. She could hear her own breathing.

She did not start the engine.

She opened her field notes. She typed: “We are not exchanging information. We are exchanging presence. The information, if there is any, is not in the codas. It is in the fact that we are both still here.”

She looked at the sentence. She crossed it out. Not because it was wrong. Because it was the kind of sentence that sounded like it meant something, and she had spent seven years making sure she did not write sentences that sounded like they meant something. She wrote instead: “Matriarch emitted novel coda, 12 clicks, descending rubato envelope. Clan departed. Hydrophone on. No response to standard coda.”

She saved the file. She closed the laptop.

The water was quiet. The hydrophone was on. Somewhere to the east, eleven whales were moving through water that Keza could not hear, saying things she could not predict, to each other and not to her.

She picked up her phone. There was a message from her brother, Toma, sent three hours ago. The preview said: “Can you call me when you’re back? It’s about Dad’s st…” The preview cut off at the ellipsis.

Keza looked at the preview. She could predict the tone. She could predict the shape of the conversation, her brother’s voice starting careful and then turning angry, and her own “I know” meaning “I don’t know how to be there for this.” She could predict all of it. She could not predict what it would mean to hear it.

She did not open the message.

She sat on the cabin floor with her father’s headphones around her neck and the hydrophone on and the whales gone, and she listened to the water, and the water did not say anything, and she stayed.