| { | |
| "paper_id": "2007", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T15:50:19.853280Z" | |
| }, | |
| "title": "Beyond Repair -Testing the Limits of the Conversational Repair System", | |
| "authors": [ | |
| { | |
| "first": "David", | |
| "middle": [], | |
| "last": "Schlangen", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Institute for Linguistics University of Potsdam", | |
| "location": { | |
| "country": "Germany" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Raquel", | |
| "middle": [], | |
| "last": "Fern\u00e1ndez", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Institute for Linguistics University of Potsdam", | |
| "location": { | |
| "country": "Germany" | |
| } | |
| }, | |
| "email": "" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "We report on an experiment on the effects of inducing acoustic understanding problems in task-oriented dialogue. We found that despite causing real problems w.r.t. task performance, many instances of induced problems were not explicitly repaired by the dialogue participants. Almost all repairs referred to the immediately preceding utterance, with problems in prior utterances left unacknowledged. Clarification requests of certain forms were in this corpus more likely to trigger reformulations than repetitions, unlike in different settings.", | |
| "pdf_parse": { | |
| "paper_id": "2007", | |
| "_pdf_hash": "", | |
| "abstract": [ | |
| { | |
| "text": "We report on an experiment on the effects of inducing acoustic understanding problems in task-oriented dialogue. We found that despite causing real problems w.r.t. task performance, many instances of induced problems were not explicitly repaired by the dialogue participants. Almost all repairs referred to the immediately preceding utterance, with problems in prior utterances left unacknowledged. Clarification requests of certain forms were in this corpus more likely to trigger reformulations than repetitions, unlike in different settings.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Abstract", | |
| "sec_num": null | |
| } | |
| ], | |
| "body_text": [ | |
| { | |
| "text": "Clarification requests (CRs), i.e., utterances that request repair of understanding problems, are typically studied on corpora of transcribed conversations (see, inter alia, (Purver, 2004; Rodr\u00edguez and Schlangen, 2004) ). While much knowledge about the use of this utterance type has been gathered this way, there are principled limitations to this approach:", | |
| "cite_spans": [ | |
| { | |
| "start": 174, | |
| "end": 188, | |
| "text": "(Purver, 2004;", | |
| "ref_id": "BIBREF4" | |
| }, | |
| { | |
| "start": 189, | |
| "end": 219, | |
| "text": "Rodr\u00edguez and Schlangen, 2004)", | |
| "ref_id": "BIBREF5" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "\u2022 If there is a CR, the problem that caused it must be inferred from its form and the original speaker's reply, as it cannot be directly observed.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "\u2022 As it is not obvious for the annotator whether there has been a problem or not, strategies for avoiding to ask for clarification cannot be studied straightforwardly.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "\u2022 The effectiveness of the repair system can only indirectly be studied.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "In this paper, we present the results of an experiment where we addressed these limitations through the controlled induction of understanding problems.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "The remainder of the paper is structured as follows. In Section 2 we describe the method used in our experiment, the results of which are then presented in Section 3. A general discussion and conclusions close the paper. 1", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "The experiment consisted in a voice-only cooperative task with two participants: an instruction giver (IG) had to describe in order of the numbering the placement of pieces on a puzzle (see Figure 1 ) to an instruction follower (IF), who only had access to the unsolved puzzle with unnumbered pieces.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 190, | |
| "end": 198, | |
| "text": "Figure 1", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Overview", | |
| "sec_num": "2.1" | |
| }, | |
| { | |
| "text": "In half of the runs we manipulated one audiochannel by replacing (in real-time, at random points) all signal with noise, effectively blocking out the speech for the hearer. Around 10% of one speaker's signal was removed in this way. The random, automatic placement of noise meant that we had no control over which part of the utterance exactly was masked, but we judged this preferable over more controlled manual placement of noise, which necessarily would have disabled real-time interactivity. The design is related to (Skantze, 2005) , where distortion was introduced through a simulated ASR, although not in real-time.", | |
| "cite_spans": [ | |
| { | |
| "start": 522, | |
| "end": 537, | |
| "text": "(Skantze, 2005)", | |
| "ref_id": "BIBREF7" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Figure 1: Solution and Outline", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "We expected the manipulation to have an effect on the effort needed to complete the task and each of its steps (placing individual puzzle pieces). Further, and more specifically, given previously observed correlations between CR forms and problem types, we expected an increase in use of CR forms previously connected to clarifying acoustic problems. As our design tells us exactly which part of the stimulus was problematic, we also wanted to explore relations between this and whether, and if so, how clarification was requested.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Figure 1: Solution and Outline", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "26 subjects (13 pairs) participated in the experiment. All were native English speakers (from a variety of native countries) that responded to a public call for participation. Half of them where college students while the other half had a range of different occupations. The age range was from 20 to over 40. None of the subjects reported any hearing difficulties.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Procedure", | |
| "sec_num": "2.2" | |
| }, | |
| { | |
| "text": "The pairs of subjects were split into IG and IF and placed in different sound-proof rooms, connected by an audio-line via headsets. They were then separately briefed on the task. IG's solution was displayed on a computer screen, IF's puzzle board was implemented in a computer program. All audio was recorded; in the runs with the manipulation, both the audio before adding noise and after adding noise was recorded. IF's computer screen was video-taped.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Procedure", | |
| "sec_num": "2.2" | |
| }, | |
| { | |
| "text": "For analysis, the recordings were transcribed using Praat (Boersma, 2001 ) and annotated using MMAX (M\u00fcller and Strube, 2001) ; the annotators had access to both the textual transcripts and the audio material.", | |
| "cite_spans": [ | |
| { | |
| "start": 58, | |
| "end": 72, | |
| "text": "(Boersma, 2001", | |
| "ref_id": "BIBREF0" | |
| }, | |
| { | |
| "start": 100, | |
| "end": 125, | |
| "text": "(M\u00fcller and Strube, 2001)", | |
| "ref_id": "BIBREF3" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Data Analysis", | |
| "sec_num": "2.2.1" | |
| }, | |
| { | |
| "text": "We segmented the recordings into utterances (following the guidelines in (Meteer and Taylor, 1995) ) and moves, which we defined as all utterances be-longing to the placement of one piece. We then annotated the transition status at move boundaries, split into grounding state, where a) the participants can be explicitly confident about their placement (\"OK, I've got it. Next one!\"); b) rather unconfident (\"Well, I'll put it there. Let's see what happens.\"); c) they can put the current sub-task on hold and go back to a previous piece; d) which in turn then can be moved and placed with any of these previous grounding outcomes, or can be reconfirmed; and success, which we checked on the video recordings. Values for this feature are: success, failure, not moved (for moves that revisited previously placed pieces, but did not move them), and on hold for moves that are on hold while a previous piece is repaired.", | |
| "cite_spans": [ | |
| { | |
| "start": 73, | |
| "end": 98, | |
| "text": "(Meteer and Taylor, 1995)", | |
| "ref_id": "BIBREF2" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Data Analysis", | |
| "sec_num": "2.2.1" | |
| }, | |
| { | |
| "text": "Within the moves, we marked regions belonging together thematically, and annotated them with the following categories: a) identification of the piece that is to be placed; b) specifying its orientation and c) location on the grid; other common dialogue actions were d) talking about the task setup (\"I am supposed to do these in order\"); e) the grounding status (\"well, let's see what happens\"); f) noting problems (\"This doesn't work. Something must be wrong.\"); g) giving a description of the state of the board (\"To the left I have the Swiss cross, and next to it...\"). Everything else was coded as h) other.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Data Analysis", | |
| "sec_num": "2.2.1" | |
| }, | |
| { | |
| "text": "Finally, we identified utterances that were CRs and coded them with (Rodr\u00edguez and Schlangen, 2004)'s scheme; for reasons of space, we refer to that paper or to (Schlangen and Fern\u00e1ndez, 2007) for a description of the values.", | |
| "cite_spans": [ | |
| { | |
| "start": 161, | |
| "end": 192, | |
| "text": "(Schlangen and Fern\u00e1ndez, 2007)", | |
| "ref_id": "BIBREF6" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Data Analysis", | |
| "sec_num": "2.2.1" | |
| }, | |
| { | |
| "text": "The 13 experimental runs resulted in 9 usable recordings, as two runs had to be excluded because of equipment failure and two because subjects aborted the task or didn't follow instructions.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Recordings", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "The pairs in the noise condition finished the task in an average 1130 seconds, producing in average 653 utterances; the pairs in the control group needed 618 seconds and 422 utterances. These differences are statistically significant (Welch's t-test; t=2.7, df=4. . There are however no significant differences between the groups (\u03c7 2 ) w.r.t. how much time was spent on different sub-tasks like identifying pieces or placements: the pairs in the noise condition don't do different things, they just do the same things for longer / more often. Table 1 shows the distributions of move outcomes. The majority of moves in the no-noise condition end with confident and successful placement. In contrast, in the noise condition only just over half of the moves are actually successful, and consequently there are more moves that are repairs of previous mistakes. The differences between the groups are significant (\u03c7 2 , for both p<0.01).", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 544, | |
| "end": 551, | |
| "text": "Table 1", | |
| "ref_id": "TABREF1" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Dialogue-based Analysis", | |
| "sec_num": "3.2" | |
| }, | |
| { | |
| "text": "The mean length of moves in terms of utterances is very similar for both groups (28.5 for noise group, 30.81 for control group), and indeed the difference is not significant: there seems to be a constant upper limit on how much time is spent on each move before the players move on, confidently or not. Table 2 shows the ratio of contributions by IG and IF within each move, averaged over all moves and separated according to grounding status and description of state; e.g., the \"54/46\" in the second line means that 54% of contributions in moves in the noise group that ended in a wrong placement came from IG and 46% from IF. Problems in a move that lead to an unsuccessful conclusion and/or notconfident grounding only in the control group had an effect on the contribution ratio, leading to more contributions by IG. ( ", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 303, | |
| "end": 310, | |
| "text": "Table 2", | |
| "ref_id": "TABREF3" | |
| }, | |
| { | |
| "start": 821, | |
| "end": 822, | |
| "text": "(", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Move-based Analysis", | |
| "sec_num": "3.3" | |
| }, | |
| { | |
| "text": "The recordings of the noise group have been segmented into 3249 utterances, those of the control group into 1607. In the noise group, there were 561 utterances that contained noise, i.e., 30.1% of all IG utterances (only those can contain noise). Only 28 of those (= 5.0%) triggered a clarification request (that is, were coded as being the antecedent of one). In the noise group, there was only one CR that was not triggered by a noise utterance; in the control group there were 8 CRs altogether. The majority of turns (both of IG and IF; turn defined as sequence of utterances before speaker change), was one utterance long, this tendency being stronger in the control group (61.8% compared to 55.6% in the noise group; difference in length distribution is significant, \u03c7 2 , p<0.001). However, there were turns of length up to 13 utterances.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Utterance-based Analysis", | |
| "sec_num": "3.4" | |
| }, | |
| { | |
| "text": "In all utterances within IG turns in the noise group (i.e., at all distances from the speaker transition), noise events were equally likely to occur. However, a noise event in an utterance at the transition pointthat is, in either the last utterance of a longer turn or in a single utterance turn-had a chance of 8.33% of triggering a CR. A noise event one utterance away from the transition point only has a 0.87% likelihood of triggering a CR. There are no CRs in the corpus whose antecedent is further away.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Utterance-based Analysis", | |
| "sec_num": "3.4" | |
| }, | |
| { | |
| "text": "Lastly, we turn to a more fine-grained analysis of the clarification requests that occurred. We compared the distributions of CR-features in this corpus with that resulting from the the other task done in the same setting, where items like strings of numbers and sentences were read from a screen by IG for the IF to write them down (see (Schlangen and Fern\u00e1ndez, 2007) ).", | |
| "cite_spans": [ | |
| { | |
| "start": 338, | |
| "end": 369, | |
| "text": "(Schlangen and Fern\u00e1ndez, 2007)", | |
| "ref_id": "BIBREF6" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Utterance-based Analysis", | |
| "sec_num": "3.4" | |
| }, | |
| { | |
| "text": "What is interesting here is that despite the manipulation being the same, there were significant differences in the CRs that occurred: in the puzzle task of the present paper, there were significantly more CRs that did not point at the exact problem location (extent), more CRs that did not present a hypothesis (severity), fewer CRs constructed through repetition of material (rel-antec), and fewer replies to CRs that were repetitions, and more reformulations or elaborations (answer) . (All differences were tested with a \u03c7 2 test, p<0.01.)", | |
| "cite_spans": [ | |
| { | |
| "start": 478, | |
| "end": 486, | |
| "text": "(answer)", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Utterance-based Analysis", | |
| "sec_num": "3.4" | |
| }, | |
| { | |
| "text": "We now briefly summarise these observations: Pairs in the noise condition needed significantly longer to finish the task, and this was not due to higher effort for repairing understanding problems, but rather to higher effort needed for repairing task-level problems, i.e. wrong placements. In fact, while there were more repairs in the noise condition than in the control condition, most induced problems went unacknowledged -and as the performance differences show, it seems to be valuable information that they miss.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Discussion and Conclusions", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "That CRs typically clarify the immediately preceding utterance has been observed before (Purver, 2004; Rodr\u00edguez and Schlangen, 2004) . Our setting allows us to see the strength of this constraint: even if there are problems with earlier utterances within a turn-and we know that they are there, as we produced them-, they are a lot less likely to be repaired than those in the last utterance of a turn. We speculate that IF judged the information gain they would achieve by clarifying too low to take the step to interrupt IG's turns. They rather settled on a more independent strategy with more reliance on tentative placements (as shown by the grounding status), which for this task turned out to be less successful than understanding IG's commands. It seems that there needs to be a baseline of understanding before utterance-level clarification is even attempted.", | |
| "cite_spans": [ | |
| { | |
| "start": 88, | |
| "end": 102, | |
| "text": "(Purver, 2004;", | |
| "ref_id": "BIBREF4" | |
| }, | |
| { | |
| "start": 103, | |
| "end": 133, | |
| "text": "Rodr\u00edguez and Schlangen, 2004)", | |
| "ref_id": "BIBREF5" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Discussion and Conclusions", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "Another interesting observation is that while the forms of the CRs that are present are not significantly different from those in comparable conditions but with different task (see previous section), the CRs are interpreted differently: significantly often, forms that trigger verbatim responses in that other corpus trigger reformulations or elaborations here. There are two possible explanations (not mutually exclusive): the CR addressees are more primed to expect clarification requests that target the meaning level (Clark, 1996) and hence treat the CRs as being such. Or, given the spontaneous, rather unplanned nature of these also often rather long description utterances, there are memory limitations that make verbatim responses harder.", | |
| "cite_spans": [ | |
| { | |
| "start": 521, | |
| "end": 534, | |
| "text": "(Clark, 1996)", | |
| "ref_id": "BIBREF1" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Discussion and Conclusions", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "To summarise, our results show that a) clarification is not automatic, but underlies complex considerations about the value of the missing information; b) CR forms are interpreted in a (task-)contextdependent way.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Discussion and Conclusions", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "In future work, we will look in more detail at the dialogue acts of the utterances at turn-boundaries. We also plan to test task-performance in the same setting, but with the IF instructed to follow a clarification policy of 'always interrupt and clarify if there is noise'. 2 ", | |
| "cite_spans": [ | |
| { | |
| "start": 275, | |
| "end": 276, | |
| "text": "2", | |
| "ref_id": "BIBREF8" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Discussion and Conclusions", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "The work described here is the second part of an experiment whose first part has been described in(Schlangen and Fern\u00e1ndez, 2007). The part described here shares the general set up with that other work (i.e., introduction of noise in one channel), but uses different materials (a different task) and coding, and has a different focus for the analysis.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| } | |
| ], | |
| "back_matter": [], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "Praat, a system for doing phonetics by computer", | |
| "authors": [ | |
| { | |
| "first": "Paul", | |
| "middle": [], | |
| "last": "Boersma", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2001, | |
| "venue": "Glot International", | |
| "volume": "5", | |
| "issue": "9", | |
| "pages": "341--345", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Paul Boersma. 2001. Praat, a system for doing phonetics by computer. Glot International, 5(9-10):341-345.", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "Using Language", | |
| "authors": [ | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Herbert", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Clark", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1996, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Herbert H. Clark. 1996. Using Language. Cambridge Univer- sity Press, Cambridge.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Dysfluency annotation stylebook for the switchboard corpus", | |
| "authors": [ | |
| { | |
| "first": "Marie", | |
| "middle": [], | |
| "last": "Meteer", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ann", | |
| "middle": [], | |
| "last": "Taylor", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1995, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Marie Meteer and Ann Taylor. 1995. Dysfluency annotation stylebook for the switchboard corpus. http://www.cis.upenn.edu/\u223cbies/manuals/DFL-book.pdf.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "MMAX: A Tool for the Annotation of Multi-modal Corpora", | |
| "authors": [ | |
| { | |
| "first": "Christoph", | |
| "middle": [], | |
| "last": "M\u00fcller", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Michael", | |
| "middle": [], | |
| "last": "Strube", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2001, | |
| "venue": "Proceedings of the 2nd IJCAI Workshop on Knowledge and Reasoning in Practical Dialogue Systems", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "45--50", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Christoph M\u00fcller and Michael Strube. 2001. MMAX: A Tool for the Annotation of Multi-modal Corpora. In Proceedings of the 2nd IJCAI Workshop on Knowledge and Reasoning in Practical Dialogue Systems, pages 45-50, Seattle, USA, August.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "The Theory and Use of Clarification Requests in Dialogue", | |
| "authors": [ | |
| { | |
| "first": "Matthew", | |
| "middle": [], | |
| "last": "Purver", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2004, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Matthew Purver. 2004. The Theory and Use of Clarification Requests in Dialogue. Ph.D. thesis, King's College, Unver- sity of London, London, UK.", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "Form, intonation and function of clarification requests in german task-oriented spoken dialogues", | |
| "authors": [ | |
| { | |
| "first": "Joseba", | |
| "middle": [], | |
| "last": "Kepa", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "David", | |
| "middle": [], | |
| "last": "Rodr\u00edguez", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Schlangen", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2004, | |
| "venue": "Proceedings of Catalog (SemDial04)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "101--108", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Kepa Joseba Rodr\u00edguez and David Schlangen. 2004. Form, intonation and function of clarification requests in german task-oriented spoken dialogues. In Proceedings of Catalog (SemDial04), pages 101-108, Barcelona, Spain, July.", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "Speaking through a noisy channel -experiments on inducing clarification behaviour in human-human dialogue", | |
| "authors": [ | |
| { | |
| "first": "David", | |
| "middle": [], | |
| "last": "Schlangen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Raquel", | |
| "middle": [], | |
| "last": "Fern\u00e1ndez", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2007, | |
| "venue": "Proceedings of Interspeech", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "David Schlangen and Raquel Fern\u00e1ndez. 2007. Speaking through a noisy channel -experiments on inducing clarifi- cation behaviour in human-human dialogue. In Proceedings of Interspeech 2007, Antwerp, Belgium, August.", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "title": "Exploring human error recovery strategies: Implications for spoken dialogue systems", | |
| "authors": [ | |
| { | |
| "first": "Gabriel", | |
| "middle": [], | |
| "last": "Skantze", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2005, | |
| "venue": "Speech Communication", | |
| "volume": "45", | |
| "issue": "3", | |
| "pages": "325--341", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Gabriel Skantze. 2005. Exploring human error recovery strate- gies: Implications for spoken dialogue systems. Speech Communication, 45(3):325-341.", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "Acknowledgements: Thanks to", | |
| "authors": [], | |
| "year": null, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Acknowledgements: Thanks to: S. Bachmann, A. Stein- hilber, H. Bohle (transcription and annotation);", | |
| "links": null | |
| }, | |
| "BIBREF12": { | |
| "ref_id": "b12", | |
| "title": "This work was supported by EU (Marie Curie Programme) and DFG (Emmy Noether Programme)", | |
| "authors": [], | |
| "year": null, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "This work was supported by EU (Marie Curie Programme) and DFG (Emmy Noether Programme).", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "TABREF1": { | |
| "html": null, | |
| "type_str": "table", | |
| "num": null, | |
| "content": "<table/>", | |
| "text": "Success of Moves, in Percent of all Moves (top) and Grounding Status at Move-Transitions p<0.05 for length in seconds; t=2.8, df=7.0, p<0.05 for utterances)" | |
| }, | |
| "TABREF3": { | |
| "html": null, | |
| "type_str": "table", | |
| "num": null, | |
| "content": "<table/>", | |
| "text": "Ratio IG/IF contributions, by move success" | |
| } | |
| } | |
| } | |
| } |