The original KLM paper says:
The basic states of a bosonic qubit encoded in modes $l_1$ and $l_2$ are $|0\rangle\mapsto|0\rangle_{l_1}|1\rangle_{l_2}$ and $|1\rangle\mapsto|1\rangle_{l_1}|0\rangle_{l_2}$.
Why couldn't we simply encode qubits as $|0\rangle\mapsto|0\rangle_{l_1}$ and $|1\rangle\mapsto|1\rangle_{l_1}$?