The Copy

The hydrophone was in at 0600, before the swell built, and by 0640 Keza had the matriarch’s voice on the spectrogram as a clean trace. Five clicks. The decoder gave her a label she had not seen before: NOVEL-17, confidence 0.91, clan EC1 (Eastern Caribbean 1), pattern 1+1+3, inter-click intervals 0.12, 0.12, 0.08, 0.08 seconds. A greeting variant, the literature said. The literature was wrong about what it meant, but the literature was all she had.

She sat in the cabin with the playback speaker off and her hands on the keyboard. The boat was Rosalind, a fourteen-meter research catamaran, engine idling at 650 rpm because the hydrophone cable ran off the stern and the drift had to be controlled. She could feel the diesel through the soles of her boots. The matriarch was 1.2 kilometers east, at depth, and her codas were coming up through the water column like bubbles.

Keza had recorded this social unit for three seasons. Seven whales: the matriarch, two adult females, three juveniles, one calf. She knew their fluke photographs, their acoustic signatures, their preferred depths. She knew that the matriarch’s codas ran to the 1+1+3 pattern when she was surfacing with the unit, and to the 5-regular pattern when she was separated from them. She knew the matriarch’s name was not a name. The matriarch did not have a name. The matriarch had a coda repertoire, and the repertoire was the clan, and the clan was the only thing that mattered to her.

Keza wanted to answer.

She had spent fourteen months building the synthesized click library. Each coda was a train of clicks, each click a 2-millisecond pulse at 2 to 20 kHz, and the library contained 312 variants drawn from recordings of EC1 whales off Dominica, Guadeloupe, and Martinique. She could emit any of them through the underwater speaker at calibrated amplitude, 185 dB re 1 µPa at 1 meter, the same source level a whale would produce. She had a permit for passive recording. She did not have a permit for playback. The permit application had been pending with the Dominican Fisheries Division for eleven months. Her supervisor at the University of the West Indies had told her to wait. Keza had not waited.

The question was simple. If the matriarch emitted a coda, and Keza emitted a synthetic coda back, and the matriarch emitted another coda, was that an exchange? Was it reciprocal? And if it was reciprocal, then the codas were addressable. They were not just identity badges. They were signals that could be directed at a specific receiver. That was the difference between a signal and a language, between her grant and her field, between her and the next person who would come along and record the same whales for another three seasons and call it science.

She knew the risk. If she played back and the matriarch changed her behavior, if the playback altered the unit’s movement or vocalization pattern, then Keza had not studied whales. She had manipulated them. She would have become the thing she studied. That was the cost.

She looked at the decoder screen. NOVEL-17 again. Confidence 0.89. The matriarch was still emitting it.

Keza typed the reply. Variant EC1-5R, five regular clicks, the most common EC1 coda, the one the matriarch used with her unit. She set the amplitude to 180 dB, five below source level, because she was not a whale and she did not want to pretend to be one louder than she was. She pressed enter.

The speaker fired. Five clicks, 0.1 seconds apart, clean in the water.

Then she waited.

The hydrophone picked up the boat’s engine at 650 rpm. The diesel had a rhythm. Four beats per second, uneven, the third beat slightly heavy because the port injector was worn. Keza had heard it for three seasons and had stopped hearing it. The decoder did not stop hearing it. The decoder flagged the engine as an unknown acoustic event, confidence 0.12, and logged it in the same column as the whale codas, which Keza had told it not to do, which it did anyway, because the engine was in the water and the hydrophone was in the water and the decoder classified everything.

At 0647 the matriarch emitted again. The decoder said: NOVEL-17, confidence 0.93. Same pattern. 1+1+3. Same intervals. She had not answered with a different coda. She had repeated herself.

Keza stared at the screen. Reciprocal exchange, attempt one. The matriarch had emitted, Keza had replied, the matriarch had emitted again. Three turns. That was an exchange by any definition. But the matriarch had not changed her coda. She had not acknowledged Keza’s reply. She had simply done what she was already doing, which was emit NOVEL-17 at intervals of roughly seven minutes, which was what she had been doing before Keza pressed enter.

Keza did not know if that counted. She sat with the not-knowing for four minutes, which was long enough for the matriarch to emit again, NOVEL-17, confidence 0.90, and for the decoder to log the engine again, confidence 0.11, and for Keza to decide that she did not know and would not know until she tried again.

She typed EC1-5R again. Pressed enter. Five clicks.

At 0658 the matriarch emitted. NOVEL-17. Confidence 0.94. Closer to the first one, not farther. The intervals were tightening. 0.11, 0.11, 0.07, 0.07. She was getting more precise, or she was getting closer, or she was doing neither and Keza was reading precision into noise because she wanted it to be there.

Keza’s hands were cold. She put them in her jacket pockets and left them there for a minute. The boat rocked. The engine hummed its four-beat rhythm, the third beat heavy.

She thought about her brother. Toma lived in Roseau and worked at the customs house and had not spoken to her in two years, not since she had told him that the recording work was more important than coming home for their mother’s birthday. He had said something about her and her whales and her grant, and she had said something about his customs forms and his life, and the conversation had ended the way conversations end when both people are right and neither is going to say so. She had sent him a message three weeks ago. He had not replied. She had not sent another.

She took her hands out of her pockets.

Second attempt. She changed the reply. EC1-7I, seven clicks, the irregular pattern, the one the matriarch used when she was separated from the unit. She chose it because it was different, because if the matriarch was testing for clan membership, then the irregular pattern was the test coda, the one that said I am here, are you? She pressed enter.

The speaker fired. Seven clicks, uneven intervals, 0.15, 0.09, 0.15, 0.09, 0.15, 0.09.

At 0709 the matriarch emitted. The decoder said: NOVEL-18. New label. Confidence 0.76. A different pattern. 1+1+3+2. Six clicks. The inter-click intervals: 0.11, 0.11, 0.07, 0.07, 0.05, 0.05. She had added two clicks. She had lengthened the coda.

Keza’s mouth went dry. That was a response. That was a different coda after a different reply. The matriarch had changed her output based on Keza’s input. That was not just an exchange. That was a turn. That was the matriarch saying I heard you and I am adjusting.

Or it was the matriarch doing what matriarchs do, which is emit codas at intervals that change over time for reasons that have nothing to do with a boat. Keza could not tell. She could not tell and she knew she could not tell and that was the problem with the whole field, the whole grant, the whole three seasons. You recorded. You classified. You named the patterns. You did not know what they meant. You did not know if they meant anything. You did not know if the matriarch was speaking or singing or just being a whale, and the difference between those things was the difference between a signal and a language and you could not tell from the outside.

Keza typed EC1-7I again. The same reply. She wanted to see if the matriarch would repeat NOVEL-18 or go back to NOVEL-17 or do something else.

At 0721 the matriarch emitted. NOVEL-18. Confidence 0.82. Same pattern. 1+1+3+2. She had held the new coda. She had not gone back.

Keza pressed her palms flat on the console. The engine thrummed through her hands. She could feel the worn injector in the third beat.

Third attempt. She changed the reply again. EC1-3S, three clicks, short. A juvenile pattern. She was not sure why she chose it. She was not sure she was choosing. She pressed enter.

At 0734 the matriarch emitted. NOVEL-19. Confidence 0.68. Seven clicks. 1+1+3+2+1. She had added another click. The confidence was dropping because the pattern was getting longer and the decoder had less training data for long patterns. The decoder was doing what decoders do. It was classifying. It was not understanding. There was a difference and Keza had spent three seasons trying not to think about it.

The matriarch’s codas were getting longer. Each time Keza replied with a different pattern, the matriarch added clicks. She was escalating, probing whether the boat was kin, and each coda was a longer, more specific question, and the boat was failing every test because the boat was not kin. The boat was a diesel engine and a hydrophone and a woman with a keyboard, and the matriarch did not know that and could not know that and was doing the only thing she knew how to do, which was ask.

Keza felt something in her chest. She had been doing the same thing. Emitting into the water, waiting for an answer that was not coming in a form she could read, and asking anyway. She had known that about herself for two years and had not told Toma and had not told anyone.

At 0745 the matriarch emitted again. NOVEL-19. Confidence 0.71. Same pattern. She was holding. Keza had not replied. She was waiting to see if the matriarch would escalate on her own, without a reply, and the matriarch had not. She was waiting for Keza.

Keza violated protocol. She knew she was violating protocol. The protocol said: emit one reply per matriarch coda, wait for the matriarch’s response, log the response, do not emit again until the matriarch emits. The protocol was written by her supervisor and approved by the ethics committee and was the reason the permit was still pending. Keza violated it because the matriarch was waiting and Keza could not stand the waiting.

She emitted EC1-5R. Five regular clicks. The most common coda. The one the matriarch used with her unit. She was saying, in the only language she had, I am here, I am kin, I am one of you. She knew she was not. She knew the matriarch would know she was not. She pressed enter anyway.

The speaker fired.

At 0752 the matriarch emitted. NOVEL-20. Confidence 0.54. Eight clicks. 1+1+3+2+1+1. She had added two more. The confidence was below the threshold Keza had set for a reliable classification, which was 0.70. The decoder was guessing. The decoder was doing what decoders do when the pattern gets too long and too variable. It was reaching.

Keza watched the confidence score. 0.54. She had set the threshold at 0.70 because that was what the literature said was reliable, and the literature was written by people who had never played a coda back to a whale, and the literature did not account for what happened when the whale started changing her coda in response to a boat. The literature did not account for NOVEL-20. The literature did not account for any of this.

She sat with the 0.54 for a long time. The boat drifted. The engine hummed. The hydrophone logged everything, including the engine, which the decoder flagged at confidence 0.10 and filed in the unknown column.

At 0810 the matriarch emitted again. NOVEL-20. Confidence 0.58. She was holding. She was waiting for Keza.

Keza did not reply. She sat in the cabin with her hands on the console and the engine in her feet and the matriarch’s coda on the screen and she did not reply because she did not know what to say. She did not have a coda that meant I am not kin and I know I am not kin and I am sorry. She did not have a coda that meant I am a woman on a boat and I am studying you and I am sorry for that too. She had 312 variants and none of them said what she needed to say.

At 0823 the matriarch emitted. NOVEL-21. Confidence 0.41. Nine clicks. The longest pattern Keza had ever recorded. The decoder was barely classifying it. The confidence was below 0.50, which meant the decoder was not classifying it, it was guessing, and the guess was probably wrong, and the label NOVEL-21 was probably not a real label, it was just the decoder doing what it did when it did not know, which was make something up.

Keza looked at the nine clicks. The last two intervals were 0.03 and 0.03, which was faster than any EC1 coda she had ever recorded. The matriarch was clicking faster. She was speeding up. She was getting louder, or closer, or both. Keza checked the amplitude. The NOVEL-21 coda was arriving at 178 dB at the hydrophone, which was louder than the previous codas by 4 dB. The matriarch was closer. She was coming toward the boat.

Keza’s hands were shaking. She put them flat on the console again. The engine hummed. The third beat was heavy.

She thought about Toma. She thought about the message she had sent three weeks ago and the silence that had followed it. She thought about the matriarch, who was coming toward the boat, and about the fact that the matriarch knew codas and knew clan and was doing the only thing she knew how to do, which was ask, and ask, and ask, and Keza was doing the only thing she knew how to do, which was record, and record, and record, and neither of them was answering the other.

At 0831 Keza made a decision. She would emit one final coda. She would not reply with a variant. She would not choose from the library. She would copy. She would emit the exact pattern the matriarch had just emitted, NOVEL-21, nine clicks, the intervals she had recorded, 0.11, 0.11, 0.07, 0.07, 0.05, 0.05, 0.03, 0.03. She would match it. Not answer it. Match it. A copy, not a reply. She did not know what that meant. She did not know if it meant anything. She did it because she could not think of anything else to do, and because the matriarch was coming toward the boat, and because Keza had been asking for three seasons and had never once been answered, and because maybe the answer was not an answer but a mirror, and maybe the matriarch already knew that, and maybe Keza was the last to know.

She built the coda in the synthesizer. Nine clicks. She matched the intervals to the recorded values. She set the amplitude to 178 dB, to match the matriarch’s source level. She pressed enter.

The speaker fired. Nine clicks. 0.11, 0.11, 0.07, 0.07, 0.05, 0.05, 0.03, 0.03. The same pattern. The same rhythm. A copy.

Then she waited.

At 0838 the matriarch emitted. The decoder said: UNKNOWN. Confidence 0.00. No label. No pattern match. The coda was not a coda. It was not a click train. It was not anything the decoder could classify.

It was the boat’s engine.

Four beats per second. 0.25, 0.25, 0.25, 0.25. The third beat slightly heavy. 650 rpm. The diesel rhythm, transmitted through the water, through the hydrophone, through the decoder, which could not classify it because it was not a whale and it was not a coda and it was not anything the decoder had been trained on.

The matriarch had emitted a coda that matched the engine rhythm. She had copied the boat. She had done exactly what Keza had done. She had mirrored. She had not answered. She had echoed.

Then she dove. The hydrophone picked up the fluke slap, a single sharp pulse at 15 Hz, and then nothing. The spectrogram went flat. The matriarch was gone. The unit was gone. The water was empty.

Keza sat in the cabin. The engine hummed. The diesel’s four-beat rhythm filled the water. The decoder logged the engine at confidence 0.10 and filed it in the unknown column. It did not relabel it. It did not call it NOVEL-22. It did not call it a coda. It called it unknown, and Keza did not change the label, and the label stayed unknown, and the file stayed open on the screen, and the engine kept running.

She left her hand on the console. The diesel vibrated through her palm, four beats to the second, the third one heavy.