a7152205e00835c98a41755248833131fa260fbc
[itlast1617.git] / exam2 / q1.tex
1 \begin{enumerate}[label=\alph*.]
2 % 1a
3 \item
4 There are several differences between dialogues of native speakers
5 without a translator and with an intermediate translator.
6
7 When native speakers are aware of the fact that there is a translation
8 process going they will adapt the conversation style to be more
9 suitable for translation. This is because translation suffers from
10 common problems such as overlap between words and differences in
11 lexical structure that can cause confusion between the speakers.
12
13 For one the native speakers might use a lot more grounding to make sure
14 the translation was correct and the partner understands the thing that
15 has been said. Moreover, the grounding is probably a lot more
16 deliberate in the case of machine translation because a simple
17 backchannel like \emph{uhu} might not be enough to convince the partner
18 that the utterance was understood. This elaborate grounding might be
19 necessary since translation happens both ways and an utterance in
20 language \emph{A} translated to language \emph{B} and back to \emph{A}
21 could be different from the utterance that they started with. When two
22 native speakers converse the grounding can be very automatic and
23 simple. Speakers can use nuances in the grounding to determine whether
24 the listener understood the utterance and adapt thereupon.
25
26 Secondly, regarding conversation turn taking there is also likely a
27 change in conversation. Since intricate turn taking behaviour is a lot
28 more difficult to translate the turn-taking will likely be much more
29 concrete and structured. Because the turn-taking is so concrete it is
30 much harder to use references that span over turns.
31
32 Thirdly, the language use will be much more simple in the case of a
33 translator intervening. Lexical divergence can only be solved by a
34 translator when the context is explaining enough. When native speakers
35 converse the context might be implicit and subtle whereas in translator
36 the context must be concrete. Speakers will adapt to this and provide
37 more context, for example in the form of more adjectives. This remark
38 also includes the use of linguistic constructions that are very
39 difficult to translate such as metaphors and complex referencing.
40
41 Fourth and lastly, the use of conversational implicature will have to
42 be minimised since by mistranslations it might not be clear what the
43 implicature is. Moreover, the speakers are probably not sure what to
44 imply since the partner might come from a very different culture with a
45 very different language structure. For example when an English speaker
46 says \emph{a couple} it probably means \emph{two or more}. In languages
47 that also posses a dual case, such as Russian, next to the common
48 singular and plural this might be translated as \emph{exactly two}.
49 Because of errors of this kind the maxims proposed by Grice are extra
50 important.
51
52 % 1b
53 \item
54 The dialogue manager component should use the aforementioned techniques
55 to improve the understanding and clearness of the conversation.
56 It must use sophisticated grounding techniques such as \emph{explicit
57 confirmation} and \emph{rejection} when there is even a minimal amount
58 of doubt.
59
60 When the system is still not sure it can use rapid reprompting to get
61 the details clear, even when in the native language of the user it
62 might have been clear from other signals such as context.
63
64 This means that the dialogue manager probably should not produce a lot
65 of references, must expect very little implicature from the user and
66 must not use ambiguities.
67
68 When the user reprompts the system it should use a different, maybe a
69 little bit more illogical, construction to say the same. By doing this
70 the translation might be a bit better and therefore easier to
71 understand for the non-native speaker.
72
73 % 1c
74 \item
75 Evaluating dialogue systems can be done via multiple perspectives.
76
77 Expectations on task completion are pretty high. When an error occurs
78 the system can just reprompt. Moreover, when the user does not
79 understand an utterance it can also ask for a reprompt to the system
80 which will then hopefully reformulate the utterance.
81
82 Expectations on efficiency will probably be a lot worse than a system
83 without translation. There will be a lot more grounding, reprompting
84 and other clarification and confirmation techniques. All these
85 technique increase understanding at the cost of efficiency. Luckily in
86 an information providence system the conversations are often short and
87 therefore the overhead will not be as devastating.
88
89 Quality cost expectations will also not be as good as without
90 translation since due to all the problems mentioned above the system
91 has to use more recovery techniques which lower the quality.
92 \end{enumerate}