Construction of a Hierarchical Translation Memory

_q}:anslation memories are t)ronfising devi('es for artt;omati(t ransla t ion. Thei r main weakuess, however, is poor coverage, on llllSeell {;ex|;. ]ill this t)at)er, l;he use of a hierarchical l;ranslat ion memory, (:onsisting of a ('as(:ade of finite si;~d;e t ransducers , is t)rot)os(;d. A mmfl)er of tr~nsdu(:e, rs is al)l)]ied to (;onverl; s(;ni;enee 1)airs fl:om a t)ilingual cortms into t rans la t ion patterns, which are then used as a t rans la t ion me, mory. Pr(; l iminary results on the ( ]e rman English V ERBMOIIIL ('orl)us a,re given. 1 I n t r o d u c t i o n In reeenl; years, exa,int)le-1)ased t;l"ansl~l;i<)l~ has been 1)rol)osed as an efli<:ient ~n(;t;llo 1)lest version, examl)le-1)ased t rans la t ion boils down to l l S i l l g a (tat;fl)ase of SOllrce sel l ( ;e l l ( ;es with their l;rmlslations. For many translat;ion tasks, esl)eeially in coml)ul;er assisl;cd |;ranslation, this at)l)roa(:h works wi th greal; success. For flflly aul;onlat;i(" l;ranslal;ion the main t ) r o t ) ] e m is t ) o o r COVel'a~e oi1 l leW data . To o v e r c o l I l e this weakness, it hierar(:hi(:al t rans la t ion lllelnory is prot)osed. Al)plying a cascade of tinite sti%te |;ra, i l s (h l ( ' e r s~ a~ SOl l rce S e l l t e l l c e is {;ral islaW, d into the tin:get language. 2 T h e T r a n s d u c e r s