Semantic　communication,　by　the　extraction　of　essential　semantic　information　from　source　data,　reduces　data　volume　and　enhances　transmission　efficiency,　making　it　increasingly　popular　among　candidate　technologies　in　6G.　In　this　letter,　a　covert　transmission　scheme　for　text　semantic　communications　is　proposed,　where　a　transmitter　attempts　to　send　text　semantic　information　to　a　legitimate　receiver　under　the　surveillance　of　a　warden,　while　the　receiver　emits　artificial　noise　(AN)　to　interfere　the　warden.　To　maximize　semantic　spectral　efficiency,　we　formulate　an　optimization　problem　while　considering　constraints　on　covertness,　the　minimum　semantic　similarity,　and　the　number　of　semantic　symbols　mapped　per　word　K.　We　derive　the　closed-form　expressions　for　the　optimal　transmit　power　and　AN　power　when　K　is　fixed,　and　employ　a　one-dimension　searching　method　to　find　the　optimal　K*.　Numerical　results　demonstrate　that　fixed　AN　power　can　contribute　to　covert　transmission　and　in　the　semantic　model,　K*　is　the　minimum　allowable　K.